1 /*
2 * Copyright (c) 1999, 2014, 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/orderAccess.inline.hpp"
53 #include "runtime/osThread.hpp"
54 #include "runtime/perfMemory.hpp"
55 #include "runtime/sharedRuntime.hpp"
56 #include "runtime/statSampler.hpp"
57 #include "runtime/stubRoutines.hpp"
58 #include "runtime/thread.inline.hpp"
59 #include "runtime/threadCritical.hpp"
60 #include "runtime/timer.hpp"
61 #include "services/attachListener.hpp"
62 #include "services/memTracker.hpp"
63 #include "services/runtimeService.hpp"
64 #include "utilities/decoder.hpp"
65 #include "utilities/defaultStream.hpp"
66 #include "utilities/events.hpp"
67 #include "utilities/elfFile.hpp"
68 #include "utilities/growableArray.hpp"
69 #include "utilities/vmError.hpp"
70
71 // put OS-includes here
72 # include <sys/types.h>
73 # include <sys/mman.h>
74 # include <sys/stat.h>
75 # include <sys/select.h>
76 # include <pthread.h>
77 # include <signal.h>
78 # include <errno.h>
79 # include <dlfcn.h>
80 # include <stdio.h>
81 # include <unistd.h>
82 # include <sys/resource.h>
83 # include <pthread.h>
84 # include <sys/stat.h>
85 # include <sys/time.h>
86 # include <sys/times.h>
87 # include <sys/utsname.h>
88 # include <sys/socket.h>
89 # include <sys/wait.h>
90 # include <pwd.h>
91 # include <poll.h>
92 # include <semaphore.h>
93 # include <fcntl.h>
94 # include <string.h>
95 # include <syscall.h>
96 # include <sys/sysinfo.h>
97 # include <gnu/libc-version.h>
98 # include <sys/ipc.h>
99 # include <sys/shm.h>
100 # include <link.h>
101 # include <stdint.h>
102 # include <inttypes.h>
103 # include <sys/ioctl.h>
104
105 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
106
107 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
108 // getrusage() is prepared to handle the associated failure.
109 #ifndef RUSAGE_THREAD
110 #define RUSAGE_THREAD (1) /* only the calling thread */
111 #endif
112
113 #define MAX_PATH (2 * K)
114
115 #define MAX_SECS 100000000
116
117 // for timer info max values which include all bits
118 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
119
120 #define LARGEPAGES_BIT (1 << 6)
121 ////////////////////////////////////////////////////////////////////////////////
122 // global variables
123 julong os::Linux::_physical_memory = 0;
124
125 address os::Linux::_initial_thread_stack_bottom = NULL;
126 uintptr_t os::Linux::_initial_thread_stack_size = 0;
127
128 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
129 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
130 Mutex* os::Linux::_createThread_lock = NULL;
131 pthread_t os::Linux::_main_thread;
132 int os::Linux::_page_size = -1;
133 const int os::Linux::_vm_default_page_size = (8 * K);
134 bool os::Linux::_is_floating_stack = false;
135 bool os::Linux::_is_NPTL = false;
136 bool os::Linux::_supports_fast_thread_cpu_time = false;
137 const char * os::Linux::_glibc_version = NULL;
138 const char * os::Linux::_libpthread_version = NULL;
139 pthread_condattr_t os::Linux::_condattr[1];
140
141 static jlong initial_time_count=0;
142
143 static int clock_tics_per_sec = 100;
144
145 // For diagnostics to print a message once. see run_periodic_checks
146 static sigset_t check_signal_done;
147 static bool check_signals = true;
148
149 static pid_t _initial_pid = 0;
150
151 /* Signal number used to suspend/resume a thread */
152
153 /* do not use any signal number less than SIGSEGV, see 4355769 */
154 static int SR_signum = SIGUSR2;
155 sigset_t SR_sigset;
156
157 /* Used to protect dlsym() calls */
158 static pthread_mutex_t dl_mutex;
159
160 // Declarations
161 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
162
163 #ifdef JAVASE_EMBEDDED
164 class MemNotifyThread: public Thread {
165 friend class VMStructs;
166 public:
167 virtual void run();
168
169 private:
170 static MemNotifyThread* _memnotify_thread;
171 int _fd;
172
173 public:
174
175 // Constructor
176 MemNotifyThread(int fd);
177
178 // Tester
179 bool is_memnotify_thread() const { return true; }
180
181 // Printing
182 char* name() const { return (char*)"Linux MemNotify Thread"; }
183
184 // Returns the single instance of the MemNotifyThread
185 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
186
187 // Create and start the single instance of MemNotifyThread
188 static void start();
189 };
190 #endif // JAVASE_EMBEDDED
191
192 // utility functions
193
194 static int SR_initialize();
195
196 julong os::available_memory() {
197 return Linux::available_memory();
198 }
199
200 julong os::Linux::available_memory() {
201 // values in struct sysinfo are "unsigned long"
202 struct sysinfo si;
203 sysinfo(&si);
204
205 return (julong)si.freeram * si.mem_unit;
206 }
207
208 julong os::physical_memory() {
209 return Linux::physical_memory();
210 }
211
212 ////////////////////////////////////////////////////////////////////////////////
213 // environment support
214
215 bool os::getenv(const char* name, char* buf, int len) {
216 const char* val = ::getenv(name);
217 if (val != NULL && strlen(val) < (size_t)len) {
218 strcpy(buf, val);
219 return true;
220 }
221 if (len > 0) buf[0] = 0; // return a null string
222 return false;
223 }
224
225
226 // Return true if user is running as root.
227
228 bool os::have_special_privileges() {
229 static bool init = false;
230 static bool privileges = false;
231 if (!init) {
232 privileges = (getuid() != geteuid()) || (getgid() != getegid());
233 init = true;
234 }
235 return privileges;
236 }
237
238
239 #ifndef SYS_gettid
240 // i386: 224, ia64: 1105, amd64: 186, sparc 143
241 #ifdef __ia64__
242 #define SYS_gettid 1105
243 #elif __i386__
244 #define SYS_gettid 224
245 #elif __amd64__
246 #define SYS_gettid 186
247 #elif __sparc__
248 #define SYS_gettid 143
249 #else
250 #error define gettid for the arch
251 #endif
252 #endif
253
254 // Cpu architecture string
255 #if defined(ZERO)
256 static char cpu_arch[] = ZERO_LIBARCH;
257 #elif defined(IA64)
258 static char cpu_arch[] = "ia64";
259 #elif defined(IA32)
260 static char cpu_arch[] = "i386";
261 #elif defined(AMD64)
262 static char cpu_arch[] = "amd64";
263 #elif defined(ARM)
264 static char cpu_arch[] = "arm";
265 #elif defined(PPC32)
266 static char cpu_arch[] = "ppc";
267 #elif defined(PPC64)
268 static char cpu_arch[] = "ppc64";
269 #elif defined(SPARC)
270 # ifdef _LP64
271 static char cpu_arch[] = "sparcv9";
272 # else
273 static char cpu_arch[] = "sparc";
274 # endif
275 #else
276 #error Add appropriate cpu_arch setting
277 #endif
278
279
280 // pid_t gettid()
281 //
282 // Returns the kernel thread id of the currently running thread. Kernel
283 // thread id is used to access /proc.
284 //
285 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
286 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
287 //
288 pid_t os::Linux::gettid() {
289 int rslt = syscall(SYS_gettid);
290 if (rslt == -1) {
291 // old kernel, no NPTL support
292 return getpid();
293 } else {
294 return (pid_t)rslt;
295 }
296 }
297
298 // Most versions of linux have a bug where the number of processors are
299 // determined by looking at the /proc file system. In a chroot environment,
300 // the system call returns 1. This causes the VM to act as if it is
301 // a single processor and elide locking (see is_MP() call).
302 static bool unsafe_chroot_detected = false;
303 static const char *unstable_chroot_error = "/proc file system not found.\n"
304 "Java may be unstable running multithreaded in a chroot "
305 "environment on Linux when /proc filesystem is not mounted.";
306
307 void os::Linux::initialize_system_info() {
308 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
309 if (processor_count() == 1) {
310 pid_t pid = os::Linux::gettid();
311 char fname[32];
312 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
313 FILE *fp = fopen(fname, "r");
314 if (fp == NULL) {
315 unsafe_chroot_detected = true;
316 } else {
317 fclose(fp);
318 }
319 }
320 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
321 assert(processor_count() > 0, "linux error");
322 }
323
324 void os::init_system_properties_values() {
325 // The next steps are taken in the product version:
326 //
327 // Obtain the JAVA_HOME value from the location of libjvm.so.
328 // This library should be located at:
329 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
330 //
331 // If "/jre/lib/" appears at the right place in the path, then we
332 // assume libjvm.so is installed in a JDK and we use this path.
333 //
334 // Otherwise exit with message: "Could not create the Java virtual machine."
335 //
336 // The following extra steps are taken in the debugging version:
337 //
338 // If "/jre/lib/" does NOT appear at the right place in the path
339 // instead of exit check for $JAVA_HOME environment variable.
340 //
341 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
342 // then we append a fake suffix "hotspot/libjvm.so" to this path so
343 // it looks like libjvm.so is installed there
344 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
345 //
346 // Otherwise exit.
347 //
348 // Important note: if the location of libjvm.so changes this
349 // code needs to be changed accordingly.
350
351 // See ld(1):
352 // The linker uses the following search paths to locate required
353 // shared libraries:
354 // 1: ...
355 // ...
356 // 7: The default directories, normally /lib and /usr/lib.
357 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
358 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
359 #else
360 #define DEFAULT_LIBPATH "/lib:/usr/lib"
361 #endif
362
363 // Base path of extensions installed on the system.
364 #define SYS_EXT_DIR "/usr/java/packages"
365 #define EXTENSIONS_DIR "/lib/ext"
366 #define ENDORSED_DIR "/lib/endorsed"
367
368 // Buffer that fits several sprintfs.
369 // Note that the space for the colon and the trailing null are provided
370 // by the nulls included by the sizeof operator.
371 const size_t bufsize =
372 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
373 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
374 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
375 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
376
377 // sysclasspath, java_home, dll_dir
378 {
379 char *pslash;
380 os::jvm_path(buf, bufsize);
381
382 // Found the full path to libjvm.so.
383 // Now cut the path to <java_home>/jre if we can.
384 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
385 pslash = strrchr(buf, '/');
386 if (pslash != NULL) {
387 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
388 }
389 Arguments::set_dll_dir(buf);
390
391 if (pslash != NULL) {
392 pslash = strrchr(buf, '/');
393 if (pslash != NULL) {
394 *pslash = '\0'; // Get rid of /<arch>.
395 pslash = strrchr(buf, '/');
396 if (pslash != NULL) {
397 *pslash = '\0'; // Get rid of /lib.
398 }
399 }
400 }
401 Arguments::set_java_home(buf);
402 set_boot_path('/', ':');
403 }
404
405 // Where to look for native libraries.
406 //
407 // Note: Due to a legacy implementation, most of the library path
408 // is set in the launcher. This was to accomodate linking restrictions
409 // on legacy Linux implementations (which are no longer supported).
410 // Eventually, all the library path setting will be done here.
411 //
412 // However, to prevent the proliferation of improperly built native
413 // libraries, the new path component /usr/java/packages is added here.
414 // Eventually, all the library path setting will be done here.
415 {
416 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
417 // should always exist (until the legacy problem cited above is
418 // addressed).
419 const char *v = ::getenv("LD_LIBRARY_PATH");
420 const char *v_colon = ":";
421 if (v == NULL) { v = ""; v_colon = ""; }
422 // That's +1 for the colon and +1 for the trailing '\0'.
423 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
424 strlen(v) + 1 +
425 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
426 mtInternal);
427 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
428 Arguments::set_library_path(ld_library_path);
429 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
430 }
431
432 // Extensions directories.
433 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
434 Arguments::set_ext_dirs(buf);
435
436 // Endorsed standards default directory.
437 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
438 Arguments::set_endorsed_dirs(buf);
439
440 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
441
442 #undef DEFAULT_LIBPATH
443 #undef SYS_EXT_DIR
444 #undef EXTENSIONS_DIR
445 #undef ENDORSED_DIR
446 }
447
448 ////////////////////////////////////////////////////////////////////////////////
449 // breakpoint support
450
451 void os::breakpoint() {
452 BREAKPOINT;
453 }
454
455 extern "C" void breakpoint() {
456 // use debugger to set breakpoint here
457 }
458
459 ////////////////////////////////////////////////////////////////////////////////
460 // signal support
461
462 debug_only(static bool signal_sets_initialized = false);
463 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
464
465 bool os::Linux::is_sig_ignored(int sig) {
466 struct sigaction oact;
467 sigaction(sig, (struct sigaction*)NULL, &oact);
468 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
469 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
470 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
471 return true;
472 else
473 return false;
474 }
475
476 void os::Linux::signal_sets_init() {
477 // Should also have an assertion stating we are still single-threaded.
478 assert(!signal_sets_initialized, "Already initialized");
479 // Fill in signals that are necessarily unblocked for all threads in
480 // the VM. Currently, we unblock the following signals:
481 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
482 // by -Xrs (=ReduceSignalUsage));
483 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
484 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
485 // the dispositions or masks wrt these signals.
486 // Programs embedding the VM that want to use the above signals for their
487 // own purposes must, at this time, use the "-Xrs" option to prevent
488 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
489 // (See bug 4345157, and other related bugs).
490 // In reality, though, unblocking these signals is really a nop, since
491 // these signals are not blocked by default.
492 sigemptyset(&unblocked_sigs);
493 sigemptyset(&allowdebug_blocked_sigs);
494 sigaddset(&unblocked_sigs, SIGILL);
495 sigaddset(&unblocked_sigs, SIGSEGV);
496 sigaddset(&unblocked_sigs, SIGBUS);
497 sigaddset(&unblocked_sigs, SIGFPE);
498 #if defined(PPC64)
499 sigaddset(&unblocked_sigs, SIGTRAP);
500 #endif
501 sigaddset(&unblocked_sigs, SR_signum);
502
503 if (!ReduceSignalUsage) {
504 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
505 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
506 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
507 }
508 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
509 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
510 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
511 }
512 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
513 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
514 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
515 }
516 }
517 // Fill in signals that are blocked by all but the VM thread.
518 sigemptyset(&vm_sigs);
519 if (!ReduceSignalUsage)
520 sigaddset(&vm_sigs, BREAK_SIGNAL);
521 debug_only(signal_sets_initialized = true);
522
523 }
524
525 // These are signals that are unblocked while a thread is running Java.
526 // (For some reason, they get blocked by default.)
527 sigset_t* os::Linux::unblocked_signals() {
528 assert(signal_sets_initialized, "Not initialized");
529 return &unblocked_sigs;
530 }
531
532 // These are the signals that are blocked while a (non-VM) thread is
533 // running Java. Only the VM thread handles these signals.
534 sigset_t* os::Linux::vm_signals() {
535 assert(signal_sets_initialized, "Not initialized");
536 return &vm_sigs;
537 }
538
539 // These are signals that are blocked during cond_wait to allow debugger in
540 sigset_t* os::Linux::allowdebug_blocked_signals() {
541 assert(signal_sets_initialized, "Not initialized");
542 return &allowdebug_blocked_sigs;
543 }
544
545 void os::Linux::hotspot_sigmask(Thread* thread) {
546
547 //Save caller's signal mask before setting VM signal mask
548 sigset_t caller_sigmask;
549 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
550
551 OSThread* osthread = thread->osthread();
552 osthread->set_caller_sigmask(caller_sigmask);
553
554 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
555
556 if (!ReduceSignalUsage) {
557 if (thread->is_VM_thread()) {
558 // Only the VM thread handles BREAK_SIGNAL ...
559 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
560 } else {
561 // ... all other threads block BREAK_SIGNAL
562 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
563 }
564 }
565 }
566
567 //////////////////////////////////////////////////////////////////////////////
568 // detecting pthread library
569
570 void os::Linux::libpthread_init() {
571 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
572 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
573 // generic name for earlier versions.
574 // Define macros here so we can build HotSpot on old systems.
575 # ifndef _CS_GNU_LIBC_VERSION
576 # define _CS_GNU_LIBC_VERSION 2
577 # endif
578 # ifndef _CS_GNU_LIBPTHREAD_VERSION
579 # define _CS_GNU_LIBPTHREAD_VERSION 3
580 # endif
581
582 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
583 if (n > 0) {
584 char *str = (char *)malloc(n, mtInternal);
585 confstr(_CS_GNU_LIBC_VERSION, str, n);
586 os::Linux::set_glibc_version(str);
587 } else {
588 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
589 static char _gnu_libc_version[32];
590 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
591 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
592 os::Linux::set_glibc_version(_gnu_libc_version);
593 }
594
595 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
596 if (n > 0) {
597 char *str = (char *)malloc(n, mtInternal);
598 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
599 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
600 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
601 // is the case. LinuxThreads has a hard limit on max number of threads.
602 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
603 // On the other hand, NPTL does not have such a limit, sysconf()
604 // will return -1 and errno is not changed. Check if it is really NPTL.
605 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
606 strstr(str, "NPTL") &&
607 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
608 free(str);
609 os::Linux::set_libpthread_version("linuxthreads");
610 } else {
611 os::Linux::set_libpthread_version(str);
612 }
613 } else {
614 // glibc before 2.3.2 only has LinuxThreads.
615 os::Linux::set_libpthread_version("linuxthreads");
616 }
617
618 if (strstr(libpthread_version(), "NPTL")) {
619 os::Linux::set_is_NPTL();
620 } else {
621 os::Linux::set_is_LinuxThreads();
622 }
623
624 // LinuxThreads have two flavors: floating-stack mode, which allows variable
625 // stack size; and fixed-stack mode. NPTL is always floating-stack.
626 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
627 os::Linux::set_is_floating_stack();
628 }
629 }
630
631 /////////////////////////////////////////////////////////////////////////////
632 // thread stack
633
634 // Force Linux kernel to expand current thread stack. If "bottom" is close
635 // to the stack guard, caller should block all signals.
636 //
637 // MAP_GROWSDOWN:
638 // A special mmap() flag that is used to implement thread stacks. It tells
639 // kernel that the memory region should extend downwards when needed. This
640 // allows early versions of LinuxThreads to only mmap the first few pages
641 // when creating a new thread. Linux kernel will automatically expand thread
642 // stack as needed (on page faults).
643 //
644 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
645 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
646 // region, it's hard to tell if the fault is due to a legitimate stack
647 // access or because of reading/writing non-exist memory (e.g. buffer
648 // overrun). As a rule, if the fault happens below current stack pointer,
649 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
650 // application (see Linux kernel fault.c).
651 //
652 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
653 // stack overflow detection.
654 //
655 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
656 // not use this flag. However, the stack of initial thread is not created
657 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
658 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
659 // and then attach the thread to JVM.
660 //
661 // To get around the problem and allow stack banging on Linux, we need to
662 // manually expand thread stack after receiving the SIGSEGV.
663 //
664 // There are two ways to expand thread stack to address "bottom", we used
665 // both of them in JVM before 1.5:
666 // 1. adjust stack pointer first so that it is below "bottom", and then
667 // touch "bottom"
668 // 2. mmap() the page in question
669 //
670 // Now alternate signal stack is gone, it's harder to use 2. For instance,
671 // if current sp is already near the lower end of page 101, and we need to
672 // call mmap() to map page 100, it is possible that part of the mmap() frame
673 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
674 // That will destroy the mmap() frame and cause VM to crash.
675 //
676 // The following code works by adjusting sp first, then accessing the "bottom"
677 // page to force a page fault. Linux kernel will then automatically expand the
678 // stack mapping.
679 //
680 // _expand_stack_to() assumes its frame size is less than page size, which
681 // should always be true if the function is not inlined.
682
683 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
684 #define NOINLINE
685 #else
686 #define NOINLINE __attribute__ ((noinline))
687 #endif
688
689 static void _expand_stack_to(address bottom) NOINLINE;
690
691 static void _expand_stack_to(address bottom) {
692 address sp;
693 size_t size;
694 volatile char *p;
695
696 // Adjust bottom to point to the largest address within the same page, it
697 // gives us a one-page buffer if alloca() allocates slightly more memory.
698 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
699 bottom += os::Linux::page_size() - 1;
700
701 // sp might be slightly above current stack pointer; if that's the case, we
702 // will alloca() a little more space than necessary, which is OK. Don't use
703 // os::current_stack_pointer(), as its result can be slightly below current
704 // stack pointer, causing us to not alloca enough to reach "bottom".
705 sp = (address)&sp;
706
707 if (sp > bottom) {
708 size = sp - bottom;
709 p = (volatile char *)alloca(size);
710 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
711 p[0] = '\0';
712 }
713 }
714
715 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
716 assert(t!=NULL, "just checking");
717 assert(t->osthread()->expanding_stack(), "expand should be set");
718 assert(t->stack_base() != NULL, "stack_base was not initialized");
719
720 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
721 sigset_t mask_all, old_sigset;
722 sigfillset(&mask_all);
723 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
724 _expand_stack_to(addr);
725 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
726 return true;
727 }
728 return false;
729 }
730
731 //////////////////////////////////////////////////////////////////////////////
732 // create new thread
733
734 static address highest_vm_reserved_address();
735
736 // check if it's safe to start a new thread
737 static bool _thread_safety_check(Thread* thread) {
738 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
739 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
740 // Heap is mmap'ed at lower end of memory space. Thread stacks are
741 // allocated (MAP_FIXED) from high address space. Every thread stack
742 // occupies a fixed size slot (usually 2Mbytes, but user can change
743 // it to other values if they rebuild LinuxThreads).
744 //
745 // Problem with MAP_FIXED is that mmap() can still succeed even part of
746 // the memory region has already been mmap'ed. That means if we have too
747 // many threads and/or very large heap, eventually thread stack will
748 // collide with heap.
749 //
750 // Here we try to prevent heap/stack collision by comparing current
751 // stack bottom with the highest address that has been mmap'ed by JVM
752 // plus a safety margin for memory maps created by native code.
753 //
754 // This feature can be disabled by setting ThreadSafetyMargin to 0
755 //
756 if (ThreadSafetyMargin > 0) {
757 address stack_bottom = os::current_stack_base() - os::current_stack_size();
758
759 // not safe if our stack extends below the safety margin
760 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
761 } else {
762 return true;
763 }
764 } else {
765 // Floating stack LinuxThreads or NPTL:
766 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
767 // there's not enough space left, pthread_create() will fail. If we come
768 // here, that means enough space has been reserved for stack.
769 return true;
770 }
771 }
772
773 // Thread start routine for all newly created threads
774 static void *java_start(Thread *thread) {
775 // Try to randomize the cache line index of hot stack frames.
776 // This helps when threads of the same stack traces evict each other's
777 // cache lines. The threads can be either from the same JVM instance, or
778 // from different JVM instances. The benefit is especially true for
779 // processors with hyperthreading technology.
780 static int counter = 0;
781 int pid = os::current_process_id();
782 alloca(((pid ^ counter++) & 7) * 128);
783
784 ThreadLocalStorage::set_thread(thread);
785
786 OSThread* osthread = thread->osthread();
787 Monitor* sync = osthread->startThread_lock();
788
789 // non floating stack LinuxThreads needs extra check, see above
790 if (!_thread_safety_check(thread)) {
791 // notify parent thread
792 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
793 osthread->set_state(ZOMBIE);
794 sync->notify_all();
795 return NULL;
796 }
797
798 // thread_id is kernel thread id (similar to Solaris LWP id)
799 osthread->set_thread_id(os::Linux::gettid());
800
801 if (UseNUMA) {
802 int lgrp_id = os::numa_get_group_id();
803 if (lgrp_id != -1) {
804 thread->set_lgrp_id(lgrp_id);
805 }
806 }
807 // initialize signal mask for this thread
808 os::Linux::hotspot_sigmask(thread);
809
810 // initialize floating point control register
811 os::Linux::init_thread_fpu_state();
812
813 // handshaking with parent thread
814 {
815 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
816
817 // notify parent thread
818 osthread->set_state(INITIALIZED);
819 sync->notify_all();
820
821 // wait until os::start_thread()
822 while (osthread->get_state() == INITIALIZED) {
823 sync->wait(Mutex::_no_safepoint_check_flag);
824 }
825 }
826
827 // call one more level start routine
828 thread->run();
829
830 return 0;
831 }
832
833 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
834 assert(thread->osthread() == NULL, "caller responsible");
835
836 // Allocate the OSThread object
837 OSThread* osthread = new OSThread(NULL, NULL);
838 if (osthread == NULL) {
839 return false;
840 }
841
842 // set the correct thread state
843 osthread->set_thread_type(thr_type);
844
845 // Initial state is ALLOCATED but not INITIALIZED
846 osthread->set_state(ALLOCATED);
847
848 thread->set_osthread(osthread);
849
850 // init thread attributes
851 pthread_attr_t attr;
852 pthread_attr_init(&attr);
853 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
854
855 // stack size
856 if (os::Linux::supports_variable_stack_size()) {
857 // calculate stack size if it's not specified by caller
858 if (stack_size == 0) {
859 stack_size = os::Linux::default_stack_size(thr_type);
860
861 switch (thr_type) {
862 case os::java_thread:
863 // Java threads use ThreadStackSize which default value can be
864 // changed with the flag -Xss
865 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
866 stack_size = JavaThread::stack_size_at_create();
867 break;
868 case os::compiler_thread:
869 if (CompilerThreadStackSize > 0) {
870 stack_size = (size_t)(CompilerThreadStackSize * K);
871 break;
872 } // else fall through:
873 // use VMThreadStackSize if CompilerThreadStackSize is not defined
874 case os::vm_thread:
875 case os::pgc_thread:
876 case os::cgc_thread:
877 case os::watcher_thread:
878 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
879 break;
880 }
881 }
882
883 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
884 pthread_attr_setstacksize(&attr, stack_size);
885 } else {
886 // let pthread_create() pick the default value.
887 }
888
889 // glibc guard page
890 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
891
892 ThreadState state;
893
894 {
895 // Serialize thread creation if we are running with fixed stack LinuxThreads
896 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
897 if (lock) {
898 os::Linux::createThread_lock()->lock_without_safepoint_check();
899 }
900
901 pthread_t tid;
902 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
903
904 pthread_attr_destroy(&attr);
905
906 if (ret != 0) {
907 if (PrintMiscellaneous && (Verbose || WizardMode)) {
908 perror("pthread_create()");
909 }
910 // Need to clean up stuff we've allocated so far
911 thread->set_osthread(NULL);
912 delete osthread;
913 if (lock) os::Linux::createThread_lock()->unlock();
914 return false;
915 }
916
917 // Store pthread info into the OSThread
918 osthread->set_pthread_id(tid);
919
920 // Wait until child thread is either initialized or aborted
921 {
922 Monitor* sync_with_child = osthread->startThread_lock();
923 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
924 while ((state = osthread->get_state()) == ALLOCATED) {
925 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
926 }
927 }
928
929 if (lock) {
930 os::Linux::createThread_lock()->unlock();
931 }
932 }
933
934 // Aborted due to thread limit being reached
935 if (state == ZOMBIE) {
936 thread->set_osthread(NULL);
937 delete osthread;
938 return false;
939 }
940
941 // The thread is returned suspended (in state INITIALIZED),
942 // and is started higher up in the call chain
943 assert(state == INITIALIZED, "race condition");
944 return true;
945 }
946
947 /////////////////////////////////////////////////////////////////////////////
948 // attach existing thread
949
950 // bootstrap the main thread
951 bool os::create_main_thread(JavaThread* thread) {
952 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
953 return create_attached_thread(thread);
954 }
955
956 bool os::create_attached_thread(JavaThread* thread) {
957 #ifdef ASSERT
958 thread->verify_not_published();
959 #endif
960
961 // Allocate the OSThread object
962 OSThread* osthread = new OSThread(NULL, NULL);
963
964 if (osthread == NULL) {
965 return false;
966 }
967
968 // Store pthread info into the OSThread
969 osthread->set_thread_id(os::Linux::gettid());
970 osthread->set_pthread_id(::pthread_self());
971
972 // initialize floating point control register
973 os::Linux::init_thread_fpu_state();
974
975 // Initial thread state is RUNNABLE
976 osthread->set_state(RUNNABLE);
977
978 thread->set_osthread(osthread);
979
980 if (UseNUMA) {
981 int lgrp_id = os::numa_get_group_id();
982 if (lgrp_id != -1) {
983 thread->set_lgrp_id(lgrp_id);
984 }
985 }
986
987 if (os::Linux::is_initial_thread()) {
988 // If current thread is initial thread, its stack is mapped on demand,
989 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
990 // the entire stack region to avoid SEGV in stack banging.
991 // It is also useful to get around the heap-stack-gap problem on SuSE
992 // kernel (see 4821821 for details). We first expand stack to the top
993 // of yellow zone, then enable stack yellow zone (order is significant,
994 // enabling yellow zone first will crash JVM on SuSE Linux), so there
995 // is no gap between the last two virtual memory regions.
996
997 JavaThread *jt = (JavaThread *)thread;
998 address addr = jt->stack_yellow_zone_base();
999 assert(addr != NULL, "initialization problem?");
1000 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1001
1002 osthread->set_expanding_stack();
1003 os::Linux::manually_expand_stack(jt, addr);
1004 osthread->clear_expanding_stack();
1005 }
1006
1007 // initialize signal mask for this thread
1008 // and save the caller's signal mask
1009 os::Linux::hotspot_sigmask(thread);
1010
1011 return true;
1012 }
1013
1014 void os::pd_start_thread(Thread* thread) {
1015 OSThread * osthread = thread->osthread();
1016 assert(osthread->get_state() != INITIALIZED, "just checking");
1017 Monitor* sync_with_child = osthread->startThread_lock();
1018 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1019 sync_with_child->notify();
1020 }
1021
1022 // Free Linux resources related to the OSThread
1023 void os::free_thread(OSThread* osthread) {
1024 assert(osthread != NULL, "osthread not set");
1025
1026 if (Thread::current()->osthread() == osthread) {
1027 // Restore caller's signal mask
1028 sigset_t sigmask = osthread->caller_sigmask();
1029 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1030 }
1031
1032 delete osthread;
1033 }
1034
1035 //////////////////////////////////////////////////////////////////////////////
1036 // thread local storage
1037
1038 // Restore the thread pointer if the destructor is called. This is in case
1039 // someone from JNI code sets up a destructor with pthread_key_create to run
1040 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1041 // will hang or crash. When detachCurrentThread is called the key will be set
1042 // to null and we will not be called again. If detachCurrentThread is never
1043 // called we could loop forever depending on the pthread implementation.
1044 static void restore_thread_pointer(void* p) {
1045 Thread* thread = (Thread*) p;
1046 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1047 }
1048
1049 int os::allocate_thread_local_storage() {
1050 pthread_key_t key;
1051 int rslt = pthread_key_create(&key, restore_thread_pointer);
1052 assert(rslt == 0, "cannot allocate thread local storage");
1053 return (int)key;
1054 }
1055
1056 // Note: This is currently not used by VM, as we don't destroy TLS key
1057 // on VM exit.
1058 void os::free_thread_local_storage(int index) {
1059 int rslt = pthread_key_delete((pthread_key_t)index);
1060 assert(rslt == 0, "invalid index");
1061 }
1062
1063 void os::thread_local_storage_at_put(int index, void* value) {
1064 int rslt = pthread_setspecific((pthread_key_t)index, value);
1065 assert(rslt == 0, "pthread_setspecific failed");
1066 }
1067
1068 extern "C" Thread* get_thread() {
1069 return ThreadLocalStorage::thread();
1070 }
1071
1072 //////////////////////////////////////////////////////////////////////////////
1073 // initial thread
1074
1075 // Check if current thread is the initial thread, similar to Solaris thr_main.
1076 bool os::Linux::is_initial_thread(void) {
1077 char dummy;
1078 // If called before init complete, thread stack bottom will be null.
1079 // Can be called if fatal error occurs before initialization.
1080 if (initial_thread_stack_bottom() == NULL) return false;
1081 assert(initial_thread_stack_bottom() != NULL &&
1082 initial_thread_stack_size() != 0,
1083 "os::init did not locate initial thread's stack region");
1084 if ((address)&dummy >= initial_thread_stack_bottom() &&
1085 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1086 return true;
1087 else return false;
1088 }
1089
1090 // Find the virtual memory area that contains addr
1091 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1092 FILE *fp = fopen("/proc/self/maps", "r");
1093 if (fp) {
1094 address low, high;
1095 while (!feof(fp)) {
1096 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1097 if (low <= addr && addr < high) {
1098 if (vma_low) *vma_low = low;
1099 if (vma_high) *vma_high = high;
1100 fclose (fp);
1101 return true;
1102 }
1103 }
1104 for (;;) {
1105 int ch = fgetc(fp);
1106 if (ch == EOF || ch == (int)'\n') break;
1107 }
1108 }
1109 fclose(fp);
1110 }
1111 return false;
1112 }
1113
1114 // Locate initial thread stack. This special handling of initial thread stack
1115 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1116 // bogus value for initial thread.
1117 void os::Linux::capture_initial_stack(size_t max_size) {
1118 // stack size is the easy part, get it from RLIMIT_STACK
1119 size_t stack_size;
1120 struct rlimit rlim;
1121 getrlimit(RLIMIT_STACK, &rlim);
1122 stack_size = rlim.rlim_cur;
1123
1124 // 6308388: a bug in ld.so will relocate its own .data section to the
1125 // lower end of primordial stack; reduce ulimit -s value a little bit
1126 // so we won't install guard page on ld.so's data section.
1127 stack_size -= 2 * page_size();
1128
1129 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1130 // 7.1, in both cases we will get 2G in return value.
1131 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1132 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1133 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1134 // in case other parts in glibc still assumes 2M max stack size.
1135 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1136 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1137 if (stack_size > 2 * K * K IA64_ONLY(*2))
1138 stack_size = 2 * K * K IA64_ONLY(*2);
1139 // Try to figure out where the stack base (top) is. This is harder.
1140 //
1141 // When an application is started, glibc saves the initial stack pointer in
1142 // a global variable "__libc_stack_end", which is then used by system
1143 // libraries. __libc_stack_end should be pretty close to stack top. The
1144 // variable is available since the very early days. However, because it is
1145 // a private interface, it could disappear in the future.
1146 //
1147 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1148 // to __libc_stack_end, it is very close to stack top, but isn't the real
1149 // stack top. Note that /proc may not exist if VM is running as a chroot
1150 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1151 // /proc/<pid>/stat could change in the future (though unlikely).
1152 //
1153 // We try __libc_stack_end first. If that doesn't work, look for
1154 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1155 // as a hint, which should work well in most cases.
1156
1157 uintptr_t stack_start;
1158
1159 // try __libc_stack_end first
1160 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1161 if (p && *p) {
1162 stack_start = *p;
1163 } else {
1164 // see if we can get the start_stack field from /proc/self/stat
1165 FILE *fp;
1166 int pid;
1167 char state;
1168 int ppid;
1169 int pgrp;
1170 int session;
1171 int nr;
1172 int tpgrp;
1173 unsigned long flags;
1174 unsigned long minflt;
1175 unsigned long cminflt;
1176 unsigned long majflt;
1177 unsigned long cmajflt;
1178 unsigned long utime;
1179 unsigned long stime;
1180 long cutime;
1181 long cstime;
1182 long prio;
1183 long nice;
1184 long junk;
1185 long it_real;
1186 uintptr_t start;
1187 uintptr_t vsize;
1188 intptr_t rss;
1189 uintptr_t rsslim;
1190 uintptr_t scodes;
1191 uintptr_t ecode;
1192 int i;
1193
1194 // Figure what the primordial thread stack base is. Code is inspired
1195 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1196 // followed by command name surrounded by parentheses, state, etc.
1197 char stat[2048];
1198 int statlen;
1199
1200 fp = fopen("/proc/self/stat", "r");
1201 if (fp) {
1202 statlen = fread(stat, 1, 2047, fp);
1203 stat[statlen] = '\0';
1204 fclose(fp);
1205
1206 // Skip pid and the command string. Note that we could be dealing with
1207 // weird command names, e.g. user could decide to rename java launcher
1208 // to "java 1.4.2 :)", then the stat file would look like
1209 // 1234 (java 1.4.2 :)) R ... ...
1210 // We don't really need to know the command string, just find the last
1211 // occurrence of ")" and then start parsing from there. See bug 4726580.
1212 char * s = strrchr(stat, ')');
1213
1214 i = 0;
1215 if (s) {
1216 // Skip blank chars
1217 do s++; while (isspace(*s));
1218
1219 #define _UFM UINTX_FORMAT
1220 #define _DFM INTX_FORMAT
1221
1222 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1223 /* 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 */
1224 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,
1225 &state, /* 3 %c */
1226 &ppid, /* 4 %d */
1227 &pgrp, /* 5 %d */
1228 &session, /* 6 %d */
1229 &nr, /* 7 %d */
1230 &tpgrp, /* 8 %d */
1231 &flags, /* 9 %lu */
1232 &minflt, /* 10 %lu */
1233 &cminflt, /* 11 %lu */
1234 &majflt, /* 12 %lu */
1235 &cmajflt, /* 13 %lu */
1236 &utime, /* 14 %lu */
1237 &stime, /* 15 %lu */
1238 &cutime, /* 16 %ld */
1239 &cstime, /* 17 %ld */
1240 &prio, /* 18 %ld */
1241 &nice, /* 19 %ld */
1242 &junk, /* 20 %ld */
1243 &it_real, /* 21 %ld */
1244 &start, /* 22 UINTX_FORMAT */
1245 &vsize, /* 23 UINTX_FORMAT */
1246 &rss, /* 24 INTX_FORMAT */
1247 &rsslim, /* 25 UINTX_FORMAT */
1248 &scodes, /* 26 UINTX_FORMAT */
1249 &ecode, /* 27 UINTX_FORMAT */
1250 &stack_start); /* 28 UINTX_FORMAT */
1251 }
1252
1253 #undef _UFM
1254 #undef _DFM
1255
1256 if (i != 28 - 2) {
1257 assert(false, "Bad conversion from /proc/self/stat");
1258 // product mode - assume we are the initial thread, good luck in the
1259 // embedded case.
1260 warning("Can't detect initial thread stack location - bad conversion");
1261 stack_start = (uintptr_t) &rlim;
1262 }
1263 } else {
1264 // For some reason we can't open /proc/self/stat (for example, running on
1265 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1266 // most cases, so don't abort:
1267 warning("Can't detect initial thread stack location - no /proc/self/stat");
1268 stack_start = (uintptr_t) &rlim;
1269 }
1270 }
1271
1272 // Now we have a pointer (stack_start) very close to the stack top, the
1273 // next thing to do is to figure out the exact location of stack top. We
1274 // can find out the virtual memory area that contains stack_start by
1275 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1276 // and its upper limit is the real stack top. (again, this would fail if
1277 // running inside chroot, because /proc may not exist.)
1278
1279 uintptr_t stack_top;
1280 address low, high;
1281 if (find_vma((address)stack_start, &low, &high)) {
1282 // success, "high" is the true stack top. (ignore "low", because initial
1283 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1284 stack_top = (uintptr_t)high;
1285 } else {
1286 // failed, likely because /proc/self/maps does not exist
1287 warning("Can't detect initial thread stack location - find_vma failed");
1288 // best effort: stack_start is normally within a few pages below the real
1289 // stack top, use it as stack top, and reduce stack size so we won't put
1290 // guard page outside stack.
1291 stack_top = stack_start;
1292 stack_size -= 16 * page_size();
1293 }
1294
1295 // stack_top could be partially down the page so align it
1296 stack_top = align_size_up(stack_top, page_size());
1297
1298 if (max_size && stack_size > max_size) {
1299 _initial_thread_stack_size = max_size;
1300 } else {
1301 _initial_thread_stack_size = stack_size;
1302 }
1303
1304 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1305 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1306 }
1307
1308 ////////////////////////////////////////////////////////////////////////////////
1309 // time support
1310
1311 // Time since start-up in seconds to a fine granularity.
1312 // Used by VMSelfDestructTimer and the MemProfiler.
1313 double os::elapsedTime() {
1314
1315 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1316 }
1317
1318 jlong os::elapsed_counter() {
1319 return javaTimeNanos() - initial_time_count;
1320 }
1321
1322 jlong os::elapsed_frequency() {
1323 return NANOSECS_PER_SEC; // nanosecond resolution
1324 }
1325
1326 bool os::supports_vtime() { return true; }
1327 bool os::enable_vtime() { return false; }
1328 bool os::vtime_enabled() { return false; }
1329
1330 double os::elapsedVTime() {
1331 struct rusage usage;
1332 int retval = getrusage(RUSAGE_THREAD, &usage);
1333 if (retval == 0) {
1334 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1335 } else {
1336 // better than nothing, but not much
1337 return elapsedTime();
1338 }
1339 }
1340
1341 jlong os::javaTimeMillis() {
1342 timeval time;
1343 int status = gettimeofday(&time, NULL);
1344 assert(status != -1, "linux error");
1345 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1346 }
1347
1348 #ifndef CLOCK_MONOTONIC
1349 #define CLOCK_MONOTONIC (1)
1350 #endif
1351
1352 void os::Linux::clock_init() {
1353 // we do dlopen's in this particular order due to bug in linux
1354 // dynamical loader (see 6348968) leading to crash on exit
1355 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1356 if (handle == NULL) {
1357 handle = dlopen("librt.so", RTLD_LAZY);
1358 }
1359
1360 if (handle) {
1361 int (*clock_getres_func)(clockid_t, struct timespec*) =
1362 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1363 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1364 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1365 if (clock_getres_func && clock_gettime_func) {
1366 // See if monotonic clock is supported by the kernel. Note that some
1367 // early implementations simply return kernel jiffies (updated every
1368 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1369 // for nano time (though the monotonic property is still nice to have).
1370 // It's fixed in newer kernels, however clock_getres() still returns
1371 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1372 // resolution for now. Hopefully as people move to new kernels, this
1373 // won't be a problem.
1374 struct timespec res;
1375 struct timespec tp;
1376 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1377 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1378 // yes, monotonic clock is supported
1379 _clock_gettime = clock_gettime_func;
1380 return;
1381 } else {
1382 // close librt if there is no monotonic clock
1383 dlclose(handle);
1384 }
1385 }
1386 }
1387 warning("No monotonic clock was available - timed services may " \
1388 "be adversely affected if the time-of-day clock changes");
1389 }
1390
1391 #ifndef SYS_clock_getres
1392
1393 #if defined(IA32) || defined(AMD64)
1394 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1395 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1396 #else
1397 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1398 #define sys_clock_getres(x,y) -1
1399 #endif
1400
1401 #else
1402 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1403 #endif
1404
1405 void os::Linux::fast_thread_clock_init() {
1406 if (!UseLinuxPosixThreadCPUClocks) {
1407 return;
1408 }
1409 clockid_t clockid;
1410 struct timespec tp;
1411 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1412 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1413
1414 // Switch to using fast clocks for thread cpu time if
1415 // the sys_clock_getres() returns 0 error code.
1416 // Note, that some kernels may support the current thread
1417 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1418 // returned by the pthread_getcpuclockid().
1419 // If the fast Posix clocks are supported then the sys_clock_getres()
1420 // must return at least tp.tv_sec == 0 which means a resolution
1421 // better than 1 sec. This is extra check for reliability.
1422
1423 if(pthread_getcpuclockid_func &&
1424 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1425 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1426
1427 _supports_fast_thread_cpu_time = true;
1428 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1429 }
1430 }
1431
1432 jlong os::javaTimeNanos() {
1433 if (Linux::supports_monotonic_clock()) {
1434 struct timespec tp;
1435 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1436 assert(status == 0, "gettime error");
1437 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1438 return result;
1439 } else {
1440 timeval time;
1441 int status = gettimeofday(&time, NULL);
1442 assert(status != -1, "linux error");
1443 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1444 return 1000 * usecs;
1445 }
1446 }
1447
1448 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1449 if (Linux::supports_monotonic_clock()) {
1450 info_ptr->max_value = ALL_64_BITS;
1451
1452 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1453 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1454 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1455 } else {
1456 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1457 info_ptr->max_value = ALL_64_BITS;
1458
1459 // gettimeofday is a real time clock so it skips
1460 info_ptr->may_skip_backward = true;
1461 info_ptr->may_skip_forward = true;
1462 }
1463
1464 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1465 }
1466
1467 // Return the real, user, and system times in seconds from an
1468 // arbitrary fixed point in the past.
1469 bool os::getTimesSecs(double* process_real_time,
1470 double* process_user_time,
1471 double* process_system_time) {
1472 struct tms ticks;
1473 clock_t real_ticks = times(&ticks);
1474
1475 if (real_ticks == (clock_t) (-1)) {
1476 return false;
1477 } else {
1478 double ticks_per_second = (double) clock_tics_per_sec;
1479 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1480 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1481 *process_real_time = ((double) real_ticks) / ticks_per_second;
1482
1483 return true;
1484 }
1485 }
1486
1487
1488 char * os::local_time_string(char *buf, size_t buflen) {
1489 struct tm t;
1490 time_t long_time;
1491 time(&long_time);
1492 localtime_r(&long_time, &t);
1493 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1494 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1495 t.tm_hour, t.tm_min, t.tm_sec);
1496 return buf;
1497 }
1498
1499 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1500 return localtime_r(clock, res);
1501 }
1502
1503 ////////////////////////////////////////////////////////////////////////////////
1504 // runtime exit support
1505
1506 // Note: os::shutdown() might be called very early during initialization, or
1507 // called from signal handler. Before adding something to os::shutdown(), make
1508 // sure it is async-safe and can handle partially initialized VM.
1509 void os::shutdown() {
1510
1511 // allow PerfMemory to attempt cleanup of any persistent resources
1512 perfMemory_exit();
1513
1514 // needs to remove object in file system
1515 AttachListener::abort();
1516
1517 // flush buffered output, finish log files
1518 ostream_abort();
1519
1520 // Check for abort hook
1521 abort_hook_t abort_hook = Arguments::abort_hook();
1522 if (abort_hook != NULL) {
1523 abort_hook();
1524 }
1525
1526 }
1527
1528 // Note: os::abort() might be called very early during initialization, or
1529 // called from signal handler. Before adding something to os::abort(), make
1530 // sure it is async-safe and can handle partially initialized VM.
1531 void os::abort(bool dump_core) {
1532 os::shutdown();
1533 if (dump_core) {
1534 #ifndef PRODUCT
1535 fdStream out(defaultStream::output_fd());
1536 out.print_raw("Current thread is ");
1537 char buf[16];
1538 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1539 out.print_raw_cr(buf);
1540 out.print_raw_cr("Dumping core ...");
1541 #endif
1542 ::abort(); // dump core
1543 }
1544
1545 ::exit(1);
1546 }
1547
1548 // Die immediately, no exit hook, no abort hook, no cleanup.
1549 void os::die() {
1550 // _exit() on LinuxThreads only kills current thread
1551 ::abort();
1552 }
1553
1554
1555 // This method is a copy of JDK's sysGetLastErrorString
1556 // from src/solaris/hpi/src/system_md.c
1557
1558 size_t os::lasterror(char *buf, size_t len) {
1559
1560 if (errno == 0) return 0;
1561
1562 const char *s = ::strerror(errno);
1563 size_t n = ::strlen(s);
1564 if (n >= len) {
1565 n = len - 1;
1566 }
1567 ::strncpy(buf, s, n);
1568 buf[n] = '\0';
1569 return n;
1570 }
1571
1572 intx os::current_thread_id() { return (intx)pthread_self(); }
1573 int os::current_process_id() {
1574
1575 // Under the old linux thread library, linux gives each thread
1576 // its own process id. Because of this each thread will return
1577 // a different pid if this method were to return the result
1578 // of getpid(2). Linux provides no api that returns the pid
1579 // of the launcher thread for the vm. This implementation
1580 // returns a unique pid, the pid of the launcher thread
1581 // that starts the vm 'process'.
1582
1583 // Under the NPTL, getpid() returns the same pid as the
1584 // launcher thread rather than a unique pid per thread.
1585 // Use gettid() if you want the old pre NPTL behaviour.
1586
1587 // if you are looking for the result of a call to getpid() that
1588 // returns a unique pid for the calling thread, then look at the
1589 // OSThread::thread_id() method in osThread_linux.hpp file
1590
1591 return (int)(_initial_pid ? _initial_pid : getpid());
1592 }
1593
1594 // DLL functions
1595
1596 const char* os::dll_file_extension() { return ".so"; }
1597
1598 // This must be hard coded because it's the system's temporary
1599 // directory not the java application's temp directory, ala java.io.tmpdir.
1600 const char* os::get_temp_directory() { return "/tmp"; }
1601
1602 static bool file_exists(const char* filename) {
1603 struct stat statbuf;
1604 if (filename == NULL || strlen(filename) == 0) {
1605 return false;
1606 }
1607 return os::stat(filename, &statbuf) == 0;
1608 }
1609
1610 bool os::dll_build_name(char* buffer, size_t buflen,
1611 const char* pname, const char* fname) {
1612 bool retval = false;
1613 // Copied from libhpi
1614 const size_t pnamelen = pname ? strlen(pname) : 0;
1615
1616 // Return error on buffer overflow.
1617 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1618 return retval;
1619 }
1620
1621 if (pnamelen == 0) {
1622 snprintf(buffer, buflen, "lib%s.so", fname);
1623 retval = true;
1624 } else if (strchr(pname, *os::path_separator()) != NULL) {
1625 int n;
1626 char** pelements = split_path(pname, &n);
1627 if (pelements == NULL) {
1628 return false;
1629 }
1630 for (int i = 0 ; i < n ; i++) {
1631 // Really shouldn't be NULL, but check can't hurt
1632 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1633 continue; // skip the empty path values
1634 }
1635 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1636 if (file_exists(buffer)) {
1637 retval = true;
1638 break;
1639 }
1640 }
1641 // release the storage
1642 for (int i = 0 ; i < n ; i++) {
1643 if (pelements[i] != NULL) {
1644 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1645 }
1646 }
1647 if (pelements != NULL) {
1648 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1649 }
1650 } else {
1651 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1652 retval = true;
1653 }
1654 return retval;
1655 }
1656
1657 // check if addr is inside libjvm.so
1658 bool os::address_is_in_vm(address addr) {
1659 static address libjvm_base_addr;
1660 Dl_info dlinfo;
1661
1662 if (libjvm_base_addr == NULL) {
1663 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1664 libjvm_base_addr = (address)dlinfo.dli_fbase;
1665 }
1666 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1667 }
1668
1669 if (dladdr((void *)addr, &dlinfo) != 0) {
1670 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1671 }
1672
1673 return false;
1674 }
1675
1676 bool os::dll_address_to_function_name(address addr, char *buf,
1677 int buflen, int *offset) {
1678 // buf is not optional, but offset is optional
1679 assert(buf != NULL, "sanity check");
1680
1681 Dl_info dlinfo;
1682
1683 if (dladdr((void*)addr, &dlinfo) != 0) {
1684 // see if we have a matching symbol
1685 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1686 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1687 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1688 }
1689 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1690 return true;
1691 }
1692 // no matching symbol so try for just file info
1693 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1694 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1695 buf, buflen, offset, dlinfo.dli_fname)) {
1696 return true;
1697 }
1698 }
1699 }
1700
1701 buf[0] = '\0';
1702 if (offset != NULL) *offset = -1;
1703 return false;
1704 }
1705
1706 struct _address_to_library_name {
1707 address addr; // input : memory address
1708 size_t buflen; // size of fname
1709 char* fname; // output: library name
1710 address base; // library base addr
1711 };
1712
1713 static int address_to_library_name_callback(struct dl_phdr_info *info,
1714 size_t size, void *data) {
1715 int i;
1716 bool found = false;
1717 address libbase = NULL;
1718 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1719
1720 // iterate through all loadable segments
1721 for (i = 0; i < info->dlpi_phnum; i++) {
1722 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1723 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1724 // base address of a library is the lowest address of its loaded
1725 // segments.
1726 if (libbase == NULL || libbase > segbase) {
1727 libbase = segbase;
1728 }
1729 // see if 'addr' is within current segment
1730 if (segbase <= d->addr &&
1731 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1732 found = true;
1733 }
1734 }
1735 }
1736
1737 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1738 // so dll_address_to_library_name() can fall through to use dladdr() which
1739 // can figure out executable name from argv[0].
1740 if (found && info->dlpi_name && info->dlpi_name[0]) {
1741 d->base = libbase;
1742 if (d->fname) {
1743 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1744 }
1745 return 1;
1746 }
1747 return 0;
1748 }
1749
1750 bool os::dll_address_to_library_name(address addr, char* buf,
1751 int buflen, int* offset) {
1752 // buf is not optional, but offset is optional
1753 assert(buf != NULL, "sanity check");
1754
1755 Dl_info dlinfo;
1756 struct _address_to_library_name data;
1757
1758 // There is a bug in old glibc dladdr() implementation that it could resolve
1759 // to wrong library name if the .so file has a base address != NULL. Here
1760 // we iterate through the program headers of all loaded libraries to find
1761 // out which library 'addr' really belongs to. This workaround can be
1762 // removed once the minimum requirement for glibc is moved to 2.3.x.
1763 data.addr = addr;
1764 data.fname = buf;
1765 data.buflen = buflen;
1766 data.base = NULL;
1767 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1768
1769 if (rslt) {
1770 // buf already contains library name
1771 if (offset) *offset = addr - data.base;
1772 return true;
1773 }
1774 if (dladdr((void*)addr, &dlinfo) != 0) {
1775 if (dlinfo.dli_fname != NULL) {
1776 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1777 }
1778 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1779 *offset = addr - (address)dlinfo.dli_fbase;
1780 }
1781 return true;
1782 }
1783
1784 buf[0] = '\0';
1785 if (offset) *offset = -1;
1786 return false;
1787 }
1788
1789 // Loads .dll/.so and
1790 // in case of error it checks if .dll/.so was built for the
1791 // same architecture as Hotspot is running on
1792
1793
1794 // Remember the stack's state. The Linux dynamic linker will change
1795 // the stack to 'executable' at most once, so we must safepoint only once.
1796 bool os::Linux::_stack_is_executable = false;
1797
1798 // VM operation that loads a library. This is necessary if stack protection
1799 // of the Java stacks can be lost during loading the library. If we
1800 // do not stop the Java threads, they can stack overflow before the stacks
1801 // are protected again.
1802 class VM_LinuxDllLoad: public VM_Operation {
1803 private:
1804 const char *_filename;
1805 char *_ebuf;
1806 int _ebuflen;
1807 void *_lib;
1808 public:
1809 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1810 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1811 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1812 void doit() {
1813 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1814 os::Linux::_stack_is_executable = true;
1815 }
1816 void* loaded_library() { return _lib; }
1817 };
1818
1819 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1820 {
1821 void * result = NULL;
1822 bool load_attempted = false;
1823
1824 // Check whether the library to load might change execution rights
1825 // of the stack. If they are changed, the protection of the stack
1826 // guard pages will be lost. We need a safepoint to fix this.
1827 //
1828 // See Linux man page execstack(8) for more info.
1829 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1830 ElfFile ef(filename);
1831 if (!ef.specifies_noexecstack()) {
1832 if (!is_init_completed()) {
1833 os::Linux::_stack_is_executable = true;
1834 // This is OK - No Java threads have been created yet, and hence no
1835 // stack guard pages to fix.
1836 //
1837 // This should happen only when you are building JDK7 using a very
1838 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1839 //
1840 // Dynamic loader will make all stacks executable after
1841 // this function returns, and will not do that again.
1842 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1843 } else {
1844 warning("You have loaded library %s which might have disabled stack guard. "
1845 "The VM will try to fix the stack guard now.\n"
1846 "It's highly recommended that you fix the library with "
1847 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1848 filename);
1849
1850 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1851 JavaThread *jt = JavaThread::current();
1852 if (jt->thread_state() != _thread_in_native) {
1853 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1854 // that requires ExecStack. Cannot enter safe point. Let's give up.
1855 warning("Unable to fix stack guard. Giving up.");
1856 } else {
1857 if (!LoadExecStackDllInVMThread) {
1858 // This is for the case where the DLL has an static
1859 // constructor function that executes JNI code. We cannot
1860 // load such DLLs in the VMThread.
1861 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1862 }
1863
1864 ThreadInVMfromNative tiv(jt);
1865 debug_only(VMNativeEntryWrapper vew;)
1866
1867 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1868 VMThread::execute(&op);
1869 if (LoadExecStackDllInVMThread) {
1870 result = op.loaded_library();
1871 }
1872 load_attempted = true;
1873 }
1874 }
1875 }
1876 }
1877
1878 if (!load_attempted) {
1879 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1880 }
1881
1882 if (result != NULL) {
1883 // Successful loading
1884 return result;
1885 }
1886
1887 Elf32_Ehdr elf_head;
1888 int diag_msg_max_length=ebuflen-strlen(ebuf);
1889 char* diag_msg_buf=ebuf+strlen(ebuf);
1890
1891 if (diag_msg_max_length==0) {
1892 // No more space in ebuf for additional diagnostics message
1893 return NULL;
1894 }
1895
1896
1897 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1898
1899 if (file_descriptor < 0) {
1900 // Can't open library, report dlerror() message
1901 return NULL;
1902 }
1903
1904 bool failed_to_read_elf_head=
1905 (sizeof(elf_head)!=
1906 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1907
1908 ::close(file_descriptor);
1909 if (failed_to_read_elf_head) {
1910 // file i/o error - report dlerror() msg
1911 return NULL;
1912 }
1913
1914 typedef struct {
1915 Elf32_Half code; // Actual value as defined in elf.h
1916 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1917 char elf_class; // 32 or 64 bit
1918 char endianess; // MSB or LSB
1919 char* name; // String representation
1920 } arch_t;
1921
1922 #ifndef EM_486
1923 #define EM_486 6 /* Intel 80486 */
1924 #endif
1925
1926 static const arch_t arch_array[]={
1927 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1928 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1929 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1930 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1931 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1932 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1933 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1934 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1935 #if defined(VM_LITTLE_ENDIAN)
1936 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1937 #else
1938 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1939 #endif
1940 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1941 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1942 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1943 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1944 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1945 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1946 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1947 };
1948
1949 #if (defined IA32)
1950 static Elf32_Half running_arch_code=EM_386;
1951 #elif (defined AMD64)
1952 static Elf32_Half running_arch_code=EM_X86_64;
1953 #elif (defined IA64)
1954 static Elf32_Half running_arch_code=EM_IA_64;
1955 #elif (defined __sparc) && (defined _LP64)
1956 static Elf32_Half running_arch_code=EM_SPARCV9;
1957 #elif (defined __sparc) && (!defined _LP64)
1958 static Elf32_Half running_arch_code=EM_SPARC;
1959 #elif (defined __powerpc64__)
1960 static Elf32_Half running_arch_code=EM_PPC64;
1961 #elif (defined __powerpc__)
1962 static Elf32_Half running_arch_code=EM_PPC;
1963 #elif (defined ARM)
1964 static Elf32_Half running_arch_code=EM_ARM;
1965 #elif (defined S390)
1966 static Elf32_Half running_arch_code=EM_S390;
1967 #elif (defined ALPHA)
1968 static Elf32_Half running_arch_code=EM_ALPHA;
1969 #elif (defined MIPSEL)
1970 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1971 #elif (defined PARISC)
1972 static Elf32_Half running_arch_code=EM_PARISC;
1973 #elif (defined MIPS)
1974 static Elf32_Half running_arch_code=EM_MIPS;
1975 #elif (defined M68K)
1976 static Elf32_Half running_arch_code=EM_68K;
1977 #else
1978 #error Method os::dll_load requires that one of following is defined:\
1979 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1980 #endif
1981
1982 // Identify compatability class for VM's architecture and library's architecture
1983 // Obtain string descriptions for architectures
1984
1985 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1986 int running_arch_index=-1;
1987
1988 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1989 if (running_arch_code == arch_array[i].code) {
1990 running_arch_index = i;
1991 }
1992 if (lib_arch.code == arch_array[i].code) {
1993 lib_arch.compat_class = arch_array[i].compat_class;
1994 lib_arch.name = arch_array[i].name;
1995 }
1996 }
1997
1998 assert(running_arch_index != -1,
1999 "Didn't find running architecture code (running_arch_code) in arch_array");
2000 if (running_arch_index == -1) {
2001 // Even though running architecture detection failed
2002 // we may still continue with reporting dlerror() message
2003 return NULL;
2004 }
2005
2006 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2007 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2008 return NULL;
2009 }
2010
2011 #ifndef S390
2012 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2013 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2014 return NULL;
2015 }
2016 #endif // !S390
2017
2018 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2019 if ( lib_arch.name!=NULL ) {
2020 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2021 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2022 lib_arch.name, arch_array[running_arch_index].name);
2023 } else {
2024 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2025 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2026 lib_arch.code,
2027 arch_array[running_arch_index].name);
2028 }
2029 }
2030
2031 return NULL;
2032 }
2033
2034 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2035 void * result = ::dlopen(filename, RTLD_LAZY);
2036 if (result == NULL) {
2037 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2038 ebuf[ebuflen-1] = '\0';
2039 }
2040 return result;
2041 }
2042
2043 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2044 void * result = NULL;
2045 if (LoadExecStackDllInVMThread) {
2046 result = dlopen_helper(filename, ebuf, ebuflen);
2047 }
2048
2049 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2050 // library that requires an executable stack, or which does not have this
2051 // stack attribute set, dlopen changes the stack attribute to executable. The
2052 // read protection of the guard pages gets lost.
2053 //
2054 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2055 // may have been queued at the same time.
2056
2057 if (!_stack_is_executable) {
2058 JavaThread *jt = Threads::first();
2059
2060 while (jt) {
2061 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2062 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2063 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2064 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2065 warning("Attempt to reguard stack yellow zone failed.");
2066 }
2067 }
2068 jt = jt->next();
2069 }
2070 }
2071
2072 return result;
2073 }
2074
2075 /*
2076 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2077 * chances are you might want to run the generated bits against glibc-2.0
2078 * libdl.so, so always use locking for any version of glibc.
2079 */
2080 void* os::dll_lookup(void* handle, const char* name) {
2081 pthread_mutex_lock(&dl_mutex);
2082 void* res = dlsym(handle, name);
2083 pthread_mutex_unlock(&dl_mutex);
2084 return res;
2085 }
2086
2087 void* os::get_default_process_handle() {
2088 return (void*)::dlopen(NULL, RTLD_LAZY);
2089 }
2090
2091 static bool _print_ascii_file(const char* filename, outputStream* st) {
2092 int fd = ::open(filename, O_RDONLY);
2093 if (fd == -1) {
2094 return false;
2095 }
2096
2097 char buf[32];
2098 int bytes;
2099 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2100 st->print_raw(buf, bytes);
2101 }
2102
2103 ::close(fd);
2104
2105 return true;
2106 }
2107
2108 void os::print_dll_info(outputStream *st) {
2109 st->print_cr("Dynamic libraries:");
2110
2111 char fname[32];
2112 pid_t pid = os::Linux::gettid();
2113
2114 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2115
2116 if (!_print_ascii_file(fname, st)) {
2117 st->print("Can not get library information for pid = %d\n", pid);
2118 }
2119 }
2120
2121 void os::print_os_info_brief(outputStream* st) {
2122 os::Linux::print_distro_info(st);
2123
2124 os::Posix::print_uname_info(st);
2125
2126 os::Linux::print_libversion_info(st);
2127
2128 }
2129
2130 void os::print_os_info(outputStream* st) {
2131 st->print("OS:");
2132
2133 os::Linux::print_distro_info(st);
2134
2135 os::Posix::print_uname_info(st);
2136
2137 // Print warning if unsafe chroot environment detected
2138 if (unsafe_chroot_detected) {
2139 st->print("WARNING!! ");
2140 st->print_cr("%s", unstable_chroot_error);
2141 }
2142
2143 os::Linux::print_libversion_info(st);
2144
2145 os::Posix::print_rlimit_info(st);
2146
2147 os::Posix::print_load_average(st);
2148
2149 os::Linux::print_full_memory_info(st);
2150 }
2151
2152 // Try to identify popular distros.
2153 // Most Linux distributions have a /etc/XXX-release file, which contains
2154 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2155 // file that also contains the OS version string. Some have more than one
2156 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2157 // /etc/redhat-release.), so the order is important.
2158 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2159 // their own specific XXX-release file as well as a redhat-release file.
2160 // Because of this the XXX-release file needs to be searched for before the
2161 // redhat-release file.
2162 // Since Red Hat has a lsb-release file that is not very descriptive the
2163 // search for redhat-release needs to be before lsb-release.
2164 // Since the lsb-release file is the new standard it needs to be searched
2165 // before the older style release files.
2166 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2167 // next to last resort. The os-release file is a new standard that contains
2168 // distribution information and the system-release file seems to be an old
2169 // standard that has been replaced by the lsb-release and os-release files.
2170 // Searching for the debian_version file is the last resort. It contains
2171 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2172 // "Debian " is printed before the contents of the debian_version file.
2173 void os::Linux::print_distro_info(outputStream* st) {
2174 if (!_print_ascii_file("/etc/oracle-release", st) &&
2175 !_print_ascii_file("/etc/mandriva-release", st) &&
2176 !_print_ascii_file("/etc/mandrake-release", st) &&
2177 !_print_ascii_file("/etc/sun-release", st) &&
2178 !_print_ascii_file("/etc/redhat-release", st) &&
2179 !_print_ascii_file("/etc/lsb-release", st) &&
2180 !_print_ascii_file("/etc/SuSE-release", st) &&
2181 !_print_ascii_file("/etc/turbolinux-release", st) &&
2182 !_print_ascii_file("/etc/gentoo-release", st) &&
2183 !_print_ascii_file("/etc/ltib-release", st) &&
2184 !_print_ascii_file("/etc/angstrom-version", st) &&
2185 !_print_ascii_file("/etc/system-release", st) &&
2186 !_print_ascii_file("/etc/os-release", st)) {
2187
2188 if (file_exists("/etc/debian_version")) {
2189 st->print("Debian ");
2190 _print_ascii_file("/etc/debian_version", st);
2191 } else {
2192 st->print("Linux");
2193 }
2194 }
2195 st->cr();
2196 }
2197
2198 void os::Linux::print_libversion_info(outputStream* st) {
2199 // libc, pthread
2200 st->print("libc:");
2201 st->print("%s ", os::Linux::glibc_version());
2202 st->print("%s ", os::Linux::libpthread_version());
2203 if (os::Linux::is_LinuxThreads()) {
2204 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2205 }
2206 st->cr();
2207 }
2208
2209 void os::Linux::print_full_memory_info(outputStream* st) {
2210 st->print("\n/proc/meminfo:\n");
2211 _print_ascii_file("/proc/meminfo", st);
2212 st->cr();
2213 }
2214
2215 void os::print_memory_info(outputStream* st) {
2216
2217 st->print("Memory:");
2218 st->print(" %dk page", os::vm_page_size()>>10);
2219
2220 // values in struct sysinfo are "unsigned long"
2221 struct sysinfo si;
2222 sysinfo(&si);
2223
2224 st->print(", physical " UINT64_FORMAT "k",
2225 os::physical_memory() >> 10);
2226 st->print("(" UINT64_FORMAT "k free)",
2227 os::available_memory() >> 10);
2228 st->print(", swap " UINT64_FORMAT "k",
2229 ((jlong)si.totalswap * si.mem_unit) >> 10);
2230 st->print("(" UINT64_FORMAT "k free)",
2231 ((jlong)si.freeswap * si.mem_unit) >> 10);
2232 st->cr();
2233 }
2234
2235 void os::pd_print_cpu_info(outputStream* st) {
2236 st->print("\n/proc/cpuinfo:\n");
2237 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2238 st->print(" <Not Available>");
2239 }
2240 st->cr();
2241 }
2242
2243 void os::print_siginfo(outputStream* st, void* siginfo) {
2244 const siginfo_t* si = (const siginfo_t*)siginfo;
2245
2246 os::Posix::print_siginfo_brief(st, si);
2247 #if INCLUDE_CDS
2248 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2249 UseSharedSpaces) {
2250 FileMapInfo* mapinfo = FileMapInfo::current_info();
2251 if (mapinfo->is_in_shared_space(si->si_addr)) {
2252 st->print("\n\nError accessing class data sharing archive." \
2253 " Mapped file inaccessible during execution, " \
2254 " possible disk/network problem.");
2255 }
2256 }
2257 #endif
2258 st->cr();
2259 }
2260
2261
2262 static void print_signal_handler(outputStream* st, int sig,
2263 char* buf, size_t buflen);
2264
2265 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2266 st->print_cr("Signal Handlers:");
2267 print_signal_handler(st, SIGSEGV, buf, buflen);
2268 print_signal_handler(st, SIGBUS , buf, buflen);
2269 print_signal_handler(st, SIGFPE , buf, buflen);
2270 print_signal_handler(st, SIGPIPE, buf, buflen);
2271 print_signal_handler(st, SIGXFSZ, buf, buflen);
2272 print_signal_handler(st, SIGILL , buf, buflen);
2273 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2274 print_signal_handler(st, SR_signum, buf, buflen);
2275 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2276 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2277 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2278 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2279 #if defined(PPC64)
2280 print_signal_handler(st, SIGTRAP, buf, buflen);
2281 #endif
2282 }
2283
2284 static char saved_jvm_path[MAXPATHLEN] = {0};
2285
2286 // Find the full path to the current module, libjvm.so
2287 void os::jvm_path(char *buf, jint buflen) {
2288 // Error checking.
2289 if (buflen < MAXPATHLEN) {
2290 assert(false, "must use a large-enough buffer");
2291 buf[0] = '\0';
2292 return;
2293 }
2294 // Lazy resolve the path to current module.
2295 if (saved_jvm_path[0] != 0) {
2296 strcpy(buf, saved_jvm_path);
2297 return;
2298 }
2299
2300 char dli_fname[MAXPATHLEN];
2301 bool ret = dll_address_to_library_name(
2302 CAST_FROM_FN_PTR(address, os::jvm_path),
2303 dli_fname, sizeof(dli_fname), NULL);
2304 assert(ret, "cannot locate libjvm");
2305 char *rp = NULL;
2306 if (ret && dli_fname[0] != '\0') {
2307 rp = realpath(dli_fname, buf);
2308 }
2309 if (rp == NULL)
2310 return;
2311
2312 if (Arguments::created_by_gamma_launcher()) {
2313 // Support for the gamma launcher. Typical value for buf is
2314 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2315 // the right place in the string, then assume we are installed in a JDK and
2316 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2317 // up the path so it looks like libjvm.so is installed there (append a
2318 // fake suffix hotspot/libjvm.so).
2319 const char *p = buf + strlen(buf) - 1;
2320 for (int count = 0; p > buf && count < 5; ++count) {
2321 for (--p; p > buf && *p != '/'; --p)
2322 /* empty */ ;
2323 }
2324
2325 if (strncmp(p, "/jre/lib/", 9) != 0) {
2326 // Look for JAVA_HOME in the environment.
2327 char* java_home_var = ::getenv("JAVA_HOME");
2328 if (java_home_var != NULL && java_home_var[0] != 0) {
2329 char* jrelib_p;
2330 int len;
2331
2332 // Check the current module name "libjvm.so".
2333 p = strrchr(buf, '/');
2334 assert(strstr(p, "/libjvm") == p, "invalid library name");
2335
2336 rp = realpath(java_home_var, buf);
2337 if (rp == NULL)
2338 return;
2339
2340 // determine if this is a legacy image or modules image
2341 // modules image doesn't have "jre" subdirectory
2342 len = strlen(buf);
2343 assert(len < buflen, "Ran out of buffer room");
2344 jrelib_p = buf + len;
2345 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2346 if (0 != access(buf, F_OK)) {
2347 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2348 }
2349
2350 if (0 == access(buf, F_OK)) {
2351 // Use current module name "libjvm.so"
2352 len = strlen(buf);
2353 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2354 } else {
2355 // Go back to path of .so
2356 rp = realpath(dli_fname, buf);
2357 if (rp == NULL)
2358 return;
2359 }
2360 }
2361 }
2362 }
2363
2364 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2365 }
2366
2367 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2368 // no prefix required, not even "_"
2369 }
2370
2371 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2372 // no suffix required
2373 }
2374
2375 ////////////////////////////////////////////////////////////////////////////////
2376 // sun.misc.Signal support
2377
2378 static volatile jint sigint_count = 0;
2379
2380 static void
2381 UserHandler(int sig, void *siginfo, void *context) {
2382 // 4511530 - sem_post is serialized and handled by the manager thread. When
2383 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2384 // don't want to flood the manager thread with sem_post requests.
2385 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2386 return;
2387
2388 // Ctrl-C is pressed during error reporting, likely because the error
2389 // handler fails to abort. Let VM die immediately.
2390 if (sig == SIGINT && is_error_reported()) {
2391 os::die();
2392 }
2393
2394 os::signal_notify(sig);
2395 }
2396
2397 void* os::user_handler() {
2398 return CAST_FROM_FN_PTR(void*, UserHandler);
2399 }
2400
2401 class Semaphore : public StackObj {
2402 public:
2403 Semaphore();
2404 ~Semaphore();
2405 void signal();
2406 void wait();
2407 bool trywait();
2408 bool timedwait(unsigned int sec, int nsec);
2409 private:
2410 sem_t _semaphore;
2411 };
2412
2413 Semaphore::Semaphore() {
2414 sem_init(&_semaphore, 0, 0);
2415 }
2416
2417 Semaphore::~Semaphore() {
2418 sem_destroy(&_semaphore);
2419 }
2420
2421 void Semaphore::signal() {
2422 sem_post(&_semaphore);
2423 }
2424
2425 void Semaphore::wait() {
2426 sem_wait(&_semaphore);
2427 }
2428
2429 bool Semaphore::trywait() {
2430 return sem_trywait(&_semaphore) == 0;
2431 }
2432
2433 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2434
2435 struct timespec ts;
2436 // Semaphore's are always associated with CLOCK_REALTIME
2437 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2438 // see unpackTime for discussion on overflow checking
2439 if (sec >= MAX_SECS) {
2440 ts.tv_sec += MAX_SECS;
2441 ts.tv_nsec = 0;
2442 } else {
2443 ts.tv_sec += sec;
2444 ts.tv_nsec += nsec;
2445 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2446 ts.tv_nsec -= NANOSECS_PER_SEC;
2447 ++ts.tv_sec; // note: this must be <= max_secs
2448 }
2449 }
2450
2451 while (1) {
2452 int result = sem_timedwait(&_semaphore, &ts);
2453 if (result == 0) {
2454 return true;
2455 } else if (errno == EINTR) {
2456 continue;
2457 } else if (errno == ETIMEDOUT) {
2458 return false;
2459 } else {
2460 return false;
2461 }
2462 }
2463 }
2464
2465 extern "C" {
2466 typedef void (*sa_handler_t)(int);
2467 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2468 }
2469
2470 void* os::signal(int signal_number, void* handler) {
2471 struct sigaction sigAct, oldSigAct;
2472
2473 sigfillset(&(sigAct.sa_mask));
2474 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2475 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2476
2477 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2478 // -1 means registration failed
2479 return (void *)-1;
2480 }
2481
2482 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2483 }
2484
2485 void os::signal_raise(int signal_number) {
2486 ::raise(signal_number);
2487 }
2488
2489 /*
2490 * The following code is moved from os.cpp for making this
2491 * code platform specific, which it is by its very nature.
2492 */
2493
2494 // Will be modified when max signal is changed to be dynamic
2495 int os::sigexitnum_pd() {
2496 return NSIG;
2497 }
2498
2499 // a counter for each possible signal value
2500 static volatile jint pending_signals[NSIG+1] = { 0 };
2501
2502 // Linux(POSIX) specific hand shaking semaphore.
2503 static sem_t sig_sem;
2504 static Semaphore sr_semaphore;
2505
2506 void os::signal_init_pd() {
2507 // Initialize signal structures
2508 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2509
2510 // Initialize signal semaphore
2511 ::sem_init(&sig_sem, 0, 0);
2512 }
2513
2514 void os::signal_notify(int sig) {
2515 Atomic::inc(&pending_signals[sig]);
2516 ::sem_post(&sig_sem);
2517 }
2518
2519 static int check_pending_signals(bool wait) {
2520 Atomic::store(0, &sigint_count);
2521 for (;;) {
2522 for (int i = 0; i < NSIG + 1; i++) {
2523 jint n = pending_signals[i];
2524 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2525 return i;
2526 }
2527 }
2528 if (!wait) {
2529 return -1;
2530 }
2531 JavaThread *thread = JavaThread::current();
2532 ThreadBlockInVM tbivm(thread);
2533
2534 bool threadIsSuspended;
2535 do {
2536 thread->set_suspend_equivalent();
2537 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2538 ::sem_wait(&sig_sem);
2539
2540 // were we externally suspended while we were waiting?
2541 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2542 if (threadIsSuspended) {
2543 //
2544 // The semaphore has been incremented, but while we were waiting
2545 // another thread suspended us. We don't want to continue running
2546 // while suspended because that would surprise the thread that
2547 // suspended us.
2548 //
2549 ::sem_post(&sig_sem);
2550
2551 thread->java_suspend_self();
2552 }
2553 } while (threadIsSuspended);
2554 }
2555 }
2556
2557 int os::signal_lookup() {
2558 return check_pending_signals(false);
2559 }
2560
2561 int os::signal_wait() {
2562 return check_pending_signals(true);
2563 }
2564
2565 ////////////////////////////////////////////////////////////////////////////////
2566 // Virtual Memory
2567
2568 int os::vm_page_size() {
2569 // Seems redundant as all get out
2570 assert(os::Linux::page_size() != -1, "must call os::init");
2571 return os::Linux::page_size();
2572 }
2573
2574 // Solaris allocates memory by pages.
2575 int os::vm_allocation_granularity() {
2576 assert(os::Linux::page_size() != -1, "must call os::init");
2577 return os::Linux::page_size();
2578 }
2579
2580 // Rationale behind this function:
2581 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2582 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2583 // samples for JITted code. Here we create private executable mapping over the code cache
2584 // and then we can use standard (well, almost, as mapping can change) way to provide
2585 // info for the reporting script by storing timestamp and location of symbol
2586 void linux_wrap_code(char* base, size_t size) {
2587 static volatile jint cnt = 0;
2588
2589 if (!UseOprofile) {
2590 return;
2591 }
2592
2593 char buf[PATH_MAX+1];
2594 int num = Atomic::add(1, &cnt);
2595
2596 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2597 os::get_temp_directory(), os::current_process_id(), num);
2598 unlink(buf);
2599
2600 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2601
2602 if (fd != -1) {
2603 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2604 if (rv != (off_t)-1) {
2605 if (::write(fd, "", 1) == 1) {
2606 mmap(base, size,
2607 PROT_READ|PROT_WRITE|PROT_EXEC,
2608 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2609 }
2610 }
2611 ::close(fd);
2612 unlink(buf);
2613 }
2614 }
2615
2616 static bool recoverable_mmap_error(int err) {
2617 // See if the error is one we can let the caller handle. This
2618 // list of errno values comes from JBS-6843484. I can't find a
2619 // Linux man page that documents this specific set of errno
2620 // values so while this list currently matches Solaris, it may
2621 // change as we gain experience with this failure mode.
2622 switch (err) {
2623 case EBADF:
2624 case EINVAL:
2625 case ENOTSUP:
2626 // let the caller deal with these errors
2627 return true;
2628
2629 default:
2630 // Any remaining errors on this OS can cause our reserved mapping
2631 // to be lost. That can cause confusion where different data
2632 // structures think they have the same memory mapped. The worst
2633 // scenario is if both the VM and a library think they have the
2634 // same memory mapped.
2635 return false;
2636 }
2637 }
2638
2639 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2640 int err) {
2641 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2642 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2643 strerror(err), err);
2644 }
2645
2646 static void warn_fail_commit_memory(char* addr, size_t size,
2647 size_t alignment_hint, bool exec,
2648 int err) {
2649 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2650 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2651 alignment_hint, exec, strerror(err), err);
2652 }
2653
2654 // NOTE: Linux kernel does not really reserve the pages for us.
2655 // All it does is to check if there are enough free pages
2656 // left at the time of mmap(). This could be a potential
2657 // problem.
2658 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2659 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2660 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2661 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2662 if (res != (uintptr_t) MAP_FAILED) {
2663 if (UseNUMAInterleaving) {
2664 numa_make_global(addr, size);
2665 }
2666 return 0;
2667 }
2668
2669 int err = errno; // save errno from mmap() call above
2670
2671 if (!recoverable_mmap_error(err)) {
2672 warn_fail_commit_memory(addr, size, exec, err);
2673 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2674 }
2675
2676 return err;
2677 }
2678
2679 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2680 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2681 }
2682
2683 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2684 const char* mesg) {
2685 assert(mesg != NULL, "mesg must be specified");
2686 int err = os::Linux::commit_memory_impl(addr, size, exec);
2687 if (err != 0) {
2688 // the caller wants all commit errors to exit with the specified mesg:
2689 warn_fail_commit_memory(addr, size, exec, err);
2690 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2691 }
2692 }
2693
2694 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2695 #ifndef MAP_HUGETLB
2696 #define MAP_HUGETLB 0x40000
2697 #endif
2698
2699 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2700 #ifndef MADV_HUGEPAGE
2701 #define MADV_HUGEPAGE 14
2702 #endif
2703
2704 int os::Linux::commit_memory_impl(char* addr, size_t size,
2705 size_t alignment_hint, bool exec) {
2706 int err = os::Linux::commit_memory_impl(addr, size, exec);
2707 if (err == 0) {
2708 realign_memory(addr, size, alignment_hint);
2709 }
2710 return err;
2711 }
2712
2713 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2714 bool exec) {
2715 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2716 }
2717
2718 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2719 size_t alignment_hint, bool exec,
2720 const char* mesg) {
2721 assert(mesg != NULL, "mesg must be specified");
2722 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2723 if (err != 0) {
2724 // the caller wants all commit errors to exit with the specified mesg:
2725 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2726 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2727 }
2728 }
2729
2730 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2731 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2732 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2733 // be supported or the memory may already be backed by huge pages.
2734 ::madvise(addr, bytes, MADV_HUGEPAGE);
2735 }
2736 }
2737
2738 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2739 // This method works by doing an mmap over an existing mmaping and effectively discarding
2740 // the existing pages. However it won't work for SHM-based large pages that cannot be
2741 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2742 // small pages on top of the SHM segment. This method always works for small pages, so we
2743 // allow that in any case.
2744 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2745 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2746 }
2747 }
2748
2749 void os::numa_make_global(char *addr, size_t bytes) {
2750 Linux::numa_interleave_memory(addr, bytes);
2751 }
2752
2753 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2754 // bind policy to MPOL_PREFERRED for the current thread.
2755 #define USE_MPOL_PREFERRED 0
2756
2757 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2758 // To make NUMA and large pages more robust when both enabled, we need to ease
2759 // the requirements on where the memory should be allocated. MPOL_BIND is the
2760 // default policy and it will force memory to be allocated on the specified
2761 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2762 // the specified node, but will not force it. Using this policy will prevent
2763 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2764 // free large pages.
2765 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2766 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2767 }
2768
2769 bool os::numa_topology_changed() { return false; }
2770
2771 size_t os::numa_get_groups_num() {
2772 int max_node = Linux::numa_max_node();
2773 return max_node > 0 ? max_node + 1 : 1;
2774 }
2775
2776 int os::numa_get_group_id() {
2777 int cpu_id = Linux::sched_getcpu();
2778 if (cpu_id != -1) {
2779 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2780 if (lgrp_id != -1) {
2781 return lgrp_id;
2782 }
2783 }
2784 return 0;
2785 }
2786
2787 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2788 for (size_t i = 0; i < size; i++) {
2789 ids[i] = i;
2790 }
2791 return size;
2792 }
2793
2794 bool os::get_page_info(char *start, page_info* info) {
2795 return false;
2796 }
2797
2798 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2799 return end;
2800 }
2801
2802
2803 int os::Linux::sched_getcpu_syscall(void) {
2804 unsigned int cpu;
2805 int retval = -1;
2806
2807 #if defined(IA32)
2808 # ifndef SYS_getcpu
2809 # define SYS_getcpu 318
2810 # endif
2811 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2812 #elif defined(AMD64)
2813 // Unfortunately we have to bring all these macros here from vsyscall.h
2814 // to be able to compile on old linuxes.
2815 # define __NR_vgetcpu 2
2816 # define VSYSCALL_START (-10UL << 20)
2817 # define VSYSCALL_SIZE 1024
2818 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2819 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2820 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2821 retval = vgetcpu(&cpu, NULL, NULL);
2822 #endif
2823
2824 return (retval == -1) ? retval : cpu;
2825 }
2826
2827 // Something to do with the numa-aware allocator needs these symbols
2828 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2829 extern "C" JNIEXPORT void numa_error(char *where) { }
2830 extern "C" JNIEXPORT int fork1() { return fork(); }
2831
2832
2833 // If we are running with libnuma version > 2, then we should
2834 // be trying to use symbols with versions 1.1
2835 // If we are running with earlier version, which did not have symbol versions,
2836 // we should use the base version.
2837 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2838 void *f = dlvsym(handle, name, "libnuma_1.1");
2839 if (f == NULL) {
2840 f = dlsym(handle, name);
2841 }
2842 return f;
2843 }
2844
2845 bool os::Linux::libnuma_init() {
2846 // sched_getcpu() should be in libc.
2847 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2848 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2849
2850 // If it's not, try a direct syscall.
2851 if (sched_getcpu() == -1)
2852 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2853
2854 if (sched_getcpu() != -1) { // Does it work?
2855 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2856 if (handle != NULL) {
2857 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2858 libnuma_dlsym(handle, "numa_node_to_cpus")));
2859 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2860 libnuma_dlsym(handle, "numa_max_node")));
2861 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2862 libnuma_dlsym(handle, "numa_available")));
2863 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2864 libnuma_dlsym(handle, "numa_tonode_memory")));
2865 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2866 libnuma_dlsym(handle, "numa_interleave_memory")));
2867 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2868 libnuma_dlsym(handle, "numa_set_bind_policy")));
2869
2870
2871 if (numa_available() != -1) {
2872 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2873 // Create a cpu -> node mapping
2874 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2875 rebuild_cpu_to_node_map();
2876 return true;
2877 }
2878 }
2879 }
2880 return false;
2881 }
2882
2883 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2884 // The table is later used in get_node_by_cpu().
2885 void os::Linux::rebuild_cpu_to_node_map() {
2886 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2887 // in libnuma (possible values are starting from 16,
2888 // and continuing up with every other power of 2, but less
2889 // than the maximum number of CPUs supported by kernel), and
2890 // is a subject to change (in libnuma version 2 the requirements
2891 // are more reasonable) we'll just hardcode the number they use
2892 // in the library.
2893 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2894
2895 size_t cpu_num = os::active_processor_count();
2896 size_t cpu_map_size = NCPUS / BitsPerCLong;
2897 size_t cpu_map_valid_size =
2898 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2899
2900 cpu_to_node()->clear();
2901 cpu_to_node()->at_grow(cpu_num - 1);
2902 size_t node_num = numa_get_groups_num();
2903
2904 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2905 for (size_t i = 0; i < node_num; i++) {
2906 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2907 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2908 if (cpu_map[j] != 0) {
2909 for (size_t k = 0; k < BitsPerCLong; k++) {
2910 if (cpu_map[j] & (1UL << k)) {
2911 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2912 }
2913 }
2914 }
2915 }
2916 }
2917 }
2918 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2919 }
2920
2921 int os::Linux::get_node_by_cpu(int cpu_id) {
2922 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2923 return cpu_to_node()->at(cpu_id);
2924 }
2925 return -1;
2926 }
2927
2928 GrowableArray<int>* os::Linux::_cpu_to_node;
2929 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2930 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2931 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2932 os::Linux::numa_available_func_t os::Linux::_numa_available;
2933 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2934 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2935 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2936 unsigned long* os::Linux::_numa_all_nodes;
2937
2938 bool os::pd_uncommit_memory(char* addr, size_t size) {
2939 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2940 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2941 return res != (uintptr_t) MAP_FAILED;
2942 }
2943
2944 static
2945 address get_stack_commited_bottom(address bottom, size_t size) {
2946 address nbot = bottom;
2947 address ntop = bottom + size;
2948
2949 size_t page_sz = os::vm_page_size();
2950 unsigned pages = size / page_sz;
2951
2952 unsigned char vec[1];
2953 unsigned imin = 1, imax = pages + 1, imid;
2954 int mincore_return_value = 0;
2955
2956 assert(imin <= imax, "Unexpected page size");
2957
2958 while (imin < imax) {
2959 imid = (imax + imin) / 2;
2960 nbot = ntop - (imid * page_sz);
2961
2962 // Use a trick with mincore to check whether the page is mapped or not.
2963 // mincore sets vec to 1 if page resides in memory and to 0 if page
2964 // is swapped output but if page we are asking for is unmapped
2965 // it returns -1,ENOMEM
2966 mincore_return_value = mincore(nbot, page_sz, vec);
2967
2968 if (mincore_return_value == -1) {
2969 // Page is not mapped go up
2970 // to find first mapped page
2971 if (errno != EAGAIN) {
2972 assert(errno == ENOMEM, "Unexpected mincore errno");
2973 imax = imid;
2974 }
2975 } else {
2976 // Page is mapped go down
2977 // to find first not mapped page
2978 imin = imid + 1;
2979 }
2980 }
2981
2982 nbot = nbot + page_sz;
2983
2984 // Adjust stack bottom one page up if last checked page is not mapped
2985 if (mincore_return_value == -1) {
2986 nbot = nbot + page_sz;
2987 }
2988
2989 return nbot;
2990 }
2991
2992
2993 // Linux uses a growable mapping for the stack, and if the mapping for
2994 // the stack guard pages is not removed when we detach a thread the
2995 // stack cannot grow beyond the pages where the stack guard was
2996 // mapped. If at some point later in the process the stack expands to
2997 // that point, the Linux kernel cannot expand the stack any further
2998 // because the guard pages are in the way, and a segfault occurs.
2999 //
3000 // However, it's essential not to split the stack region by unmapping
3001 // a region (leaving a hole) that's already part of the stack mapping,
3002 // so if the stack mapping has already grown beyond the guard pages at
3003 // the time we create them, we have to truncate the stack mapping.
3004 // So, we need to know the extent of the stack mapping when
3005 // create_stack_guard_pages() is called.
3006
3007 // We only need this for stacks that are growable: at the time of
3008 // writing thread stacks don't use growable mappings (i.e. those
3009 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3010 // only applies to the main thread.
3011
3012 // If the (growable) stack mapping already extends beyond the point
3013 // where we're going to put our guard pages, truncate the mapping at
3014 // that point by munmap()ping it. This ensures that when we later
3015 // munmap() the guard pages we don't leave a hole in the stack
3016 // mapping. This only affects the main/initial thread
3017
3018 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3019
3020 if (os::Linux::is_initial_thread()) {
3021 // As we manually grow stack up to bottom inside create_attached_thread(),
3022 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3023 // we don't need to do anything special.
3024 // Check it first, before calling heavy function.
3025 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3026 unsigned char vec[1];
3027
3028 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3029 // Fallback to slow path on all errors, including EAGAIN
3030 stack_extent = (uintptr_t) get_stack_commited_bottom(
3031 os::Linux::initial_thread_stack_bottom(),
3032 (size_t)addr - stack_extent);
3033 }
3034
3035 if (stack_extent < (uintptr_t)addr) {
3036 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3037 }
3038 }
3039
3040 return os::commit_memory(addr, size, !ExecMem);
3041 }
3042
3043 // If this is a growable mapping, remove the guard pages entirely by
3044 // munmap()ping them. If not, just call uncommit_memory(). This only
3045 // affects the main/initial thread, but guard against future OS changes
3046 // It's safe to always unmap guard pages for initial thread because we
3047 // always place it right after end of the mapped region
3048
3049 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3050 uintptr_t stack_extent, stack_base;
3051
3052 if (os::Linux::is_initial_thread()) {
3053 return ::munmap(addr, size) == 0;
3054 }
3055
3056 return os::uncommit_memory(addr, size);
3057 }
3058
3059 static address _highest_vm_reserved_address = NULL;
3060
3061 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3062 // at 'requested_addr'. If there are existing memory mappings at the same
3063 // location, however, they will be overwritten. If 'fixed' is false,
3064 // 'requested_addr' is only treated as a hint, the return value may or
3065 // may not start from the requested address. Unlike Linux mmap(), this
3066 // function returns NULL to indicate failure.
3067 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3068 char * addr;
3069 int flags;
3070
3071 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3072 if (fixed) {
3073 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3074 flags |= MAP_FIXED;
3075 }
3076
3077 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3078 // touch an uncommitted page. Otherwise, the read/write might
3079 // succeed if we have enough swap space to back the physical page.
3080 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3081 flags, -1, 0);
3082
3083 if (addr != MAP_FAILED) {
3084 // anon_mmap() should only get called during VM initialization,
3085 // don't need lock (actually we can skip locking even it can be called
3086 // from multiple threads, because _highest_vm_reserved_address is just a
3087 // hint about the upper limit of non-stack memory regions.)
3088 if ((address)addr + bytes > _highest_vm_reserved_address) {
3089 _highest_vm_reserved_address = (address)addr + bytes;
3090 }
3091 }
3092
3093 return addr == MAP_FAILED ? NULL : addr;
3094 }
3095
3096 // Don't update _highest_vm_reserved_address, because there might be memory
3097 // regions above addr + size. If so, releasing a memory region only creates
3098 // a hole in the address space, it doesn't help prevent heap-stack collision.
3099 //
3100 static int anon_munmap(char * addr, size_t size) {
3101 return ::munmap(addr, size) == 0;
3102 }
3103
3104 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3105 size_t alignment_hint) {
3106 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3107 }
3108
3109 bool os::pd_release_memory(char* addr, size_t size) {
3110 return anon_munmap(addr, size);
3111 }
3112
3113 static address highest_vm_reserved_address() {
3114 return _highest_vm_reserved_address;
3115 }
3116
3117 static bool linux_mprotect(char* addr, size_t size, int prot) {
3118 // Linux wants the mprotect address argument to be page aligned.
3119 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3120
3121 // According to SUSv3, mprotect() should only be used with mappings
3122 // established by mmap(), and mmap() always maps whole pages. Unaligned
3123 // 'addr' likely indicates problem in the VM (e.g. trying to change
3124 // protection of malloc'ed or statically allocated memory). Check the
3125 // caller if you hit this assert.
3126 assert(addr == bottom, "sanity check");
3127
3128 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3129 return ::mprotect(bottom, size, prot) == 0;
3130 }
3131
3132 // Set protections specified
3133 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3134 bool is_committed) {
3135 unsigned int p = 0;
3136 switch (prot) {
3137 case MEM_PROT_NONE: p = PROT_NONE; break;
3138 case MEM_PROT_READ: p = PROT_READ; break;
3139 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3140 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3141 default:
3142 ShouldNotReachHere();
3143 }
3144 // is_committed is unused.
3145 return linux_mprotect(addr, bytes, p);
3146 }
3147
3148 bool os::guard_memory(char* addr, size_t size) {
3149 return linux_mprotect(addr, size, PROT_NONE);
3150 }
3151
3152 bool os::unguard_memory(char* addr, size_t size) {
3153 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3154 }
3155
3156 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3157 bool result = false;
3158 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3159 MAP_ANONYMOUS|MAP_PRIVATE,
3160 -1, 0);
3161 if (p != MAP_FAILED) {
3162 void *aligned_p = align_ptr_up(p, page_size);
3163
3164 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3165
3166 munmap(p, page_size * 2);
3167 }
3168
3169 if (warn && !result) {
3170 warning("TransparentHugePages is not supported by the operating system.");
3171 }
3172
3173 return result;
3174 }
3175
3176 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3177 bool result = false;
3178 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3179 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3180 -1, 0);
3181
3182 if (p != MAP_FAILED) {
3183 // We don't know if this really is a huge page or not.
3184 FILE *fp = fopen("/proc/self/maps", "r");
3185 if (fp) {
3186 while (!feof(fp)) {
3187 char chars[257];
3188 long x = 0;
3189 if (fgets(chars, sizeof(chars), fp)) {
3190 if (sscanf(chars, "%lx-%*x", &x) == 1
3191 && x == (long)p) {
3192 if (strstr (chars, "hugepage")) {
3193 result = true;
3194 break;
3195 }
3196 }
3197 }
3198 }
3199 fclose(fp);
3200 }
3201 munmap(p, page_size);
3202 }
3203
3204 if (warn && !result) {
3205 warning("HugeTLBFS is not supported by the operating system.");
3206 }
3207
3208 return result;
3209 }
3210
3211 /*
3212 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3213 *
3214 * From the coredump_filter documentation:
3215 *
3216 * - (bit 0) anonymous private memory
3217 * - (bit 1) anonymous shared memory
3218 * - (bit 2) file-backed private memory
3219 * - (bit 3) file-backed shared memory
3220 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3221 * effective only if the bit 2 is cleared)
3222 * - (bit 5) hugetlb private memory
3223 * - (bit 6) hugetlb shared memory
3224 */
3225 static void set_coredump_filter(void) {
3226 FILE *f;
3227 long cdm;
3228
3229 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3230 return;
3231 }
3232
3233 if (fscanf(f, "%lx", &cdm) != 1) {
3234 fclose(f);
3235 return;
3236 }
3237
3238 rewind(f);
3239
3240 if ((cdm & LARGEPAGES_BIT) == 0) {
3241 cdm |= LARGEPAGES_BIT;
3242 fprintf(f, "%#lx", cdm);
3243 }
3244
3245 fclose(f);
3246 }
3247
3248 // Large page support
3249
3250 static size_t _large_page_size = 0;
3251
3252 size_t os::Linux::find_large_page_size() {
3253 size_t large_page_size = 0;
3254
3255 // large_page_size on Linux is used to round up heap size. x86 uses either
3256 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3257 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3258 // page as large as 256M.
3259 //
3260 // Here we try to figure out page size by parsing /proc/meminfo and looking
3261 // for a line with the following format:
3262 // Hugepagesize: 2048 kB
3263 //
3264 // If we can't determine the value (e.g. /proc is not mounted, or the text
3265 // format has been changed), we'll use the largest page size supported by
3266 // the processor.
3267
3268 #ifndef ZERO
3269 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3270 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3271 #endif // ZERO
3272
3273 FILE *fp = fopen("/proc/meminfo", "r");
3274 if (fp) {
3275 while (!feof(fp)) {
3276 int x = 0;
3277 char buf[16];
3278 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3279 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3280 large_page_size = x * K;
3281 break;
3282 }
3283 } else {
3284 // skip to next line
3285 for (;;) {
3286 int ch = fgetc(fp);
3287 if (ch == EOF || ch == (int)'\n') break;
3288 }
3289 }
3290 }
3291 fclose(fp);
3292 }
3293
3294 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3295 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3296 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3297 proper_unit_for_byte_size(large_page_size));
3298 }
3299
3300 return large_page_size;
3301 }
3302
3303 size_t os::Linux::setup_large_page_size() {
3304 _large_page_size = Linux::find_large_page_size();
3305 const size_t default_page_size = (size_t)Linux::page_size();
3306 if (_large_page_size > default_page_size) {
3307 _page_sizes[0] = _large_page_size;
3308 _page_sizes[1] = default_page_size;
3309 _page_sizes[2] = 0;
3310 }
3311
3312 return _large_page_size;
3313 }
3314
3315 bool os::Linux::setup_large_page_type(size_t page_size) {
3316 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3317 FLAG_IS_DEFAULT(UseSHM) &&
3318 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3319
3320 // The type of large pages has not been specified by the user.
3321
3322 // Try UseHugeTLBFS and then UseSHM.
3323 UseHugeTLBFS = UseSHM = true;
3324
3325 // Don't try UseTransparentHugePages since there are known
3326 // performance issues with it turned on. This might change in the future.
3327 UseTransparentHugePages = false;
3328 }
3329
3330 if (UseTransparentHugePages) {
3331 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3332 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3333 UseHugeTLBFS = false;
3334 UseSHM = false;
3335 return true;
3336 }
3337 UseTransparentHugePages = false;
3338 }
3339
3340 if (UseHugeTLBFS) {
3341 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3342 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3343 UseSHM = false;
3344 return true;
3345 }
3346 UseHugeTLBFS = false;
3347 }
3348
3349 return UseSHM;
3350 }
3351
3352 void os::large_page_init() {
3353 if (!UseLargePages &&
3354 !UseTransparentHugePages &&
3355 !UseHugeTLBFS &&
3356 !UseSHM) {
3357 // Not using large pages.
3358 return;
3359 }
3360
3361 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3362 // The user explicitly turned off large pages.
3363 // Ignore the rest of the large pages flags.
3364 UseTransparentHugePages = false;
3365 UseHugeTLBFS = false;
3366 UseSHM = false;
3367 return;
3368 }
3369
3370 size_t large_page_size = Linux::setup_large_page_size();
3371 UseLargePages = Linux::setup_large_page_type(large_page_size);
3372
3373 set_coredump_filter();
3374 }
3375
3376 #ifndef SHM_HUGETLB
3377 #define SHM_HUGETLB 04000
3378 #endif
3379
3380 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3381 // "exec" is passed in but not used. Creating the shared image for
3382 // the code cache doesn't have an SHM_X executable permission to check.
3383 assert(UseLargePages && UseSHM, "only for SHM large pages");
3384 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3385
3386 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3387 return NULL; // Fallback to small pages.
3388 }
3389
3390 key_t key = IPC_PRIVATE;
3391 char *addr;
3392
3393 bool warn_on_failure = UseLargePages &&
3394 (!FLAG_IS_DEFAULT(UseLargePages) ||
3395 !FLAG_IS_DEFAULT(UseSHM) ||
3396 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3397 );
3398 char msg[128];
3399
3400 // Create a large shared memory region to attach to based on size.
3401 // Currently, size is the total size of the heap
3402 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3403 if (shmid == -1) {
3404 // Possible reasons for shmget failure:
3405 // 1. shmmax is too small for Java heap.
3406 // > check shmmax value: cat /proc/sys/kernel/shmmax
3407 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3408 // 2. not enough large page memory.
3409 // > check available large pages: cat /proc/meminfo
3410 // > increase amount of large pages:
3411 // echo new_value > /proc/sys/vm/nr_hugepages
3412 // Note 1: different Linux may use different name for this property,
3413 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3414 // Note 2: it's possible there's enough physical memory available but
3415 // they are so fragmented after a long run that they can't
3416 // coalesce into large pages. Try to reserve large pages when
3417 // the system is still "fresh".
3418 if (warn_on_failure) {
3419 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3420 warning("%s", msg);
3421 }
3422 return NULL;
3423 }
3424
3425 // attach to the region
3426 addr = (char*)shmat(shmid, req_addr, 0);
3427 int err = errno;
3428
3429 // Remove shmid. If shmat() is successful, the actual shared memory segment
3430 // will be deleted when it's detached by shmdt() or when the process
3431 // terminates. If shmat() is not successful this will remove the shared
3432 // segment immediately.
3433 shmctl(shmid, IPC_RMID, NULL);
3434
3435 if ((intptr_t)addr == -1) {
3436 if (warn_on_failure) {
3437 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3438 warning("%s", msg);
3439 }
3440 return NULL;
3441 }
3442
3443 return addr;
3444 }
3445
3446 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3447 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3448
3449 bool warn_on_failure = UseLargePages &&
3450 (!FLAG_IS_DEFAULT(UseLargePages) ||
3451 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3452 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3453
3454 if (warn_on_failure) {
3455 char msg[128];
3456 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3457 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3458 warning("%s", msg);
3459 }
3460 }
3461
3462 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3463 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3464 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3465 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3466
3467 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3468 char* addr = (char*)::mmap(req_addr, bytes, prot,
3469 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3470 -1, 0);
3471
3472 if (addr == MAP_FAILED) {
3473 warn_on_large_pages_failure(req_addr, bytes, errno);
3474 return NULL;
3475 }
3476
3477 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3478
3479 return addr;
3480 }
3481
3482 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3483 size_t large_page_size = os::large_page_size();
3484
3485 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3486
3487 // Allocate small pages.
3488
3489 char* start;
3490 if (req_addr != NULL) {
3491 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3492 assert(is_size_aligned(bytes, alignment), "Must be");
3493 start = os::reserve_memory(bytes, req_addr);
3494 assert(start == NULL || start == req_addr, "Must be");
3495 } else {
3496 start = os::reserve_memory_aligned(bytes, alignment);
3497 }
3498
3499 if (start == NULL) {
3500 return NULL;
3501 }
3502
3503 assert(is_ptr_aligned(start, alignment), "Must be");
3504
3505 if (MemTracker::tracking_level() > NMT_minimal) {
3506 // os::reserve_memory_special will record this memory area.
3507 // Need to release it here to prevent overlapping reservations.
3508 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3509 tkr.record((address)start, bytes);
3510 }
3511
3512 char* end = start + bytes;
3513
3514 // Find the regions of the allocated chunk that can be promoted to large pages.
3515 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3516 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3517
3518 size_t lp_bytes = lp_end - lp_start;
3519
3520 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3521
3522 if (lp_bytes == 0) {
3523 // The mapped region doesn't even span the start and the end of a large page.
3524 // Fall back to allocate a non-special area.
3525 ::munmap(start, end - start);
3526 return NULL;
3527 }
3528
3529 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3530
3531
3532 void* result;
3533
3534 if (start != lp_start) {
3535 result = ::mmap(start, lp_start - start, prot,
3536 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3537 -1, 0);
3538 if (result == MAP_FAILED) {
3539 ::munmap(lp_start, end - lp_start);
3540 return NULL;
3541 }
3542 }
3543
3544 result = ::mmap(lp_start, lp_bytes, prot,
3545 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3546 -1, 0);
3547 if (result == MAP_FAILED) {
3548 warn_on_large_pages_failure(req_addr, bytes, errno);
3549 // If the mmap above fails, the large pages region will be unmapped and we
3550 // have regions before and after with small pages. Release these regions.
3551 //
3552 // | mapped | unmapped | mapped |
3553 // ^ ^ ^ ^
3554 // start lp_start lp_end end
3555 //
3556 ::munmap(start, lp_start - start);
3557 ::munmap(lp_end, end - lp_end);
3558 return NULL;
3559 }
3560
3561 if (lp_end != end) {
3562 result = ::mmap(lp_end, end - lp_end, prot,
3563 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3564 -1, 0);
3565 if (result == MAP_FAILED) {
3566 ::munmap(start, lp_end - start);
3567 return NULL;
3568 }
3569 }
3570
3571 return start;
3572 }
3573
3574 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3575 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3576 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3577 assert(is_power_of_2(alignment), "Must be");
3578 assert(is_power_of_2(os::large_page_size()), "Must be");
3579 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3580
3581 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3582 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3583 } else {
3584 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3585 }
3586 }
3587
3588 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3589 assert(UseLargePages, "only for large pages");
3590
3591 char* addr;
3592 if (UseSHM) {
3593 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3594 } else {
3595 assert(UseHugeTLBFS, "must be");
3596 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3597 }
3598
3599 if (addr != NULL) {
3600 if (UseNUMAInterleaving) {
3601 numa_make_global(addr, bytes);
3602 }
3603
3604 // The memory is committed
3605 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3606 }
3607
3608 return addr;
3609 }
3610
3611 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3612 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3613 return shmdt(base) == 0;
3614 }
3615
3616 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3617 return pd_release_memory(base, bytes);
3618 }
3619
3620 bool os::release_memory_special(char* base, size_t bytes) {
3621 bool res;
3622 if (MemTracker::tracking_level() > NMT_minimal) {
3623 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3624 res = os::Linux::release_memory_special_impl(base, bytes);
3625 if (res) {
3626 tkr.record((address)base, bytes);
3627 }
3628
3629 } else {
3630 res = os::Linux::release_memory_special_impl(base, bytes);
3631 }
3632 return res;
3633 }
3634
3635 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3636 assert(UseLargePages, "only for large pages");
3637 bool res;
3638
3639 if (UseSHM) {
3640 res = os::Linux::release_memory_special_shm(base, bytes);
3641 } else {
3642 assert(UseHugeTLBFS, "must be");
3643 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3644 }
3645 return res;
3646 }
3647
3648 size_t os::large_page_size() {
3649 return _large_page_size;
3650 }
3651
3652 // With SysV SHM the entire memory region must be allocated as shared
3653 // memory.
3654 // HugeTLBFS allows application to commit large page memory on demand.
3655 // However, when committing memory with HugeTLBFS fails, the region
3656 // that was supposed to be committed will lose the old reservation
3657 // and allow other threads to steal that memory region. Because of this
3658 // behavior we can't commit HugeTLBFS memory.
3659 bool os::can_commit_large_page_memory() {
3660 return UseTransparentHugePages;
3661 }
3662
3663 bool os::can_execute_large_page_memory() {
3664 return UseTransparentHugePages || UseHugeTLBFS;
3665 }
3666
3667 // Reserve memory at an arbitrary address, only if that area is
3668 // available (and not reserved for something else).
3669
3670 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3671 const int max_tries = 10;
3672 char* base[max_tries];
3673 size_t size[max_tries];
3674 const size_t gap = 0x000000;
3675
3676 // Assert only that the size is a multiple of the page size, since
3677 // that's all that mmap requires, and since that's all we really know
3678 // about at this low abstraction level. If we need higher alignment,
3679 // we can either pass an alignment to this method or verify alignment
3680 // in one of the methods further up the call chain. See bug 5044738.
3681 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3682
3683 // Repeatedly allocate blocks until the block is allocated at the
3684 // right spot. Give up after max_tries. Note that reserve_memory() will
3685 // automatically update _highest_vm_reserved_address if the call is
3686 // successful. The variable tracks the highest memory address every reserved
3687 // by JVM. It is used to detect heap-stack collision if running with
3688 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3689 // space than needed, it could confuse the collision detecting code. To
3690 // solve the problem, save current _highest_vm_reserved_address and
3691 // calculate the correct value before return.
3692 address old_highest = _highest_vm_reserved_address;
3693
3694 // Linux mmap allows caller to pass an address as hint; give it a try first,
3695 // if kernel honors the hint then we can return immediately.
3696 char * addr = anon_mmap(requested_addr, bytes, false);
3697 if (addr == requested_addr) {
3698 return requested_addr;
3699 }
3700
3701 if (addr != NULL) {
3702 // mmap() is successful but it fails to reserve at the requested address
3703 anon_munmap(addr, bytes);
3704 }
3705
3706 int i;
3707 for (i = 0; i < max_tries; ++i) {
3708 base[i] = reserve_memory(bytes);
3709
3710 if (base[i] != NULL) {
3711 // Is this the block we wanted?
3712 if (base[i] == requested_addr) {
3713 size[i] = bytes;
3714 break;
3715 }
3716
3717 // Does this overlap the block we wanted? Give back the overlapped
3718 // parts and try again.
3719
3720 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3721 if (top_overlap >= 0 && top_overlap < bytes) {
3722 unmap_memory(base[i], top_overlap);
3723 base[i] += top_overlap;
3724 size[i] = bytes - top_overlap;
3725 } else {
3726 size_t bottom_overlap = base[i] + bytes - requested_addr;
3727 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3728 unmap_memory(requested_addr, bottom_overlap);
3729 size[i] = bytes - bottom_overlap;
3730 } else {
3731 size[i] = bytes;
3732 }
3733 }
3734 }
3735 }
3736
3737 // Give back the unused reserved pieces.
3738
3739 for (int j = 0; j < i; ++j) {
3740 if (base[j] != NULL) {
3741 unmap_memory(base[j], size[j]);
3742 }
3743 }
3744
3745 if (i < max_tries) {
3746 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3747 return requested_addr;
3748 } else {
3749 _highest_vm_reserved_address = old_highest;
3750 return NULL;
3751 }
3752 }
3753
3754 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3755 return ::read(fd, buf, nBytes);
3756 }
3757
3758 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3759 // Solaris uses poll(), linux uses park().
3760 // Poll() is likely a better choice, assuming that Thread.interrupt()
3761 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3762 // SIGSEGV, see 4355769.
3763
3764 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3765 assert(thread == Thread::current(), "thread consistency check");
3766
3767 ParkEvent * const slp = thread->_SleepEvent ;
3768 slp->reset() ;
3769 OrderAccess::fence() ;
3770
3771 if (interruptible) {
3772 jlong prevtime = javaTimeNanos();
3773
3774 for (;;) {
3775 if (os::is_interrupted(thread, true)) {
3776 return OS_INTRPT;
3777 }
3778
3779 jlong newtime = javaTimeNanos();
3780
3781 if (newtime - prevtime < 0) {
3782 // time moving backwards, should only happen if no monotonic clock
3783 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3784 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3785 } else {
3786 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3787 }
3788
3789 if(millis <= 0) {
3790 return OS_OK;
3791 }
3792
3793 prevtime = newtime;
3794
3795 {
3796 assert(thread->is_Java_thread(), "sanity check");
3797 JavaThread *jt = (JavaThread *) thread;
3798 ThreadBlockInVM tbivm(jt);
3799 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3800
3801 jt->set_suspend_equivalent();
3802 // cleared by handle_special_suspend_equivalent_condition() or
3803 // java_suspend_self() via check_and_wait_while_suspended()
3804
3805 slp->park(millis);
3806
3807 // were we externally suspended while we were waiting?
3808 jt->check_and_wait_while_suspended();
3809 }
3810 }
3811 } else {
3812 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3813 jlong prevtime = javaTimeNanos();
3814
3815 for (;;) {
3816 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3817 // the 1st iteration ...
3818 jlong newtime = javaTimeNanos();
3819
3820 if (newtime - prevtime < 0) {
3821 // time moving backwards, should only happen if no monotonic clock
3822 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3823 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3824 } else {
3825 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3826 }
3827
3828 if(millis <= 0) break ;
3829
3830 prevtime = newtime;
3831 slp->park(millis);
3832 }
3833 return OS_OK ;
3834 }
3835 }
3836
3837 //
3838 // Short sleep, direct OS call.
3839 //
3840 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3841 // sched_yield(2) will actually give up the CPU:
3842 //
3843 // * Alone on this pariticular CPU, keeps running.
3844 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3845 // (pre 2.6.39).
3846 //
3847 // So calling this with 0 is an alternative.
3848 //
3849 void os::naked_short_sleep(jlong ms) {
3850 struct timespec req;
3851
3852 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3853 req.tv_sec = 0;
3854 if (ms > 0) {
3855 req.tv_nsec = (ms % 1000) * 1000000;
3856 }
3857 else {
3858 req.tv_nsec = 1;
3859 }
3860
3861 nanosleep(&req, NULL);
3862
3863 return;
3864 }
3865
3866 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3867 void os::infinite_sleep() {
3868 while (true) { // sleep forever ...
3869 ::sleep(100); // ... 100 seconds at a time
3870 }
3871 }
3872
3873 // Used to convert frequent JVM_Yield() to nops
3874 bool os::dont_yield() {
3875 return DontYieldALot;
3876 }
3877
3878 void os::yield() {
3879 sched_yield();
3880 }
3881
3882 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3883
3884 void os::yield_all(int attempts) {
3885 // Yields to all threads, including threads with lower priorities
3886 // Threads on Linux are all with same priority. The Solaris style
3887 // os::yield_all() with nanosleep(1ms) is not necessary.
3888 sched_yield();
3889 }
3890
3891 // Called from the tight loops to possibly influence time-sharing heuristics
3892 void os::loop_breaker(int attempts) {
3893 os::yield_all(attempts);
3894 }
3895
3896 ////////////////////////////////////////////////////////////////////////////////
3897 // thread priority support
3898
3899 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3900 // only supports dynamic priority, static priority must be zero. For real-time
3901 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3902 // However, for large multi-threaded applications, SCHED_RR is not only slower
3903 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3904 // of 5 runs - Sep 2005).
3905 //
3906 // The following code actually changes the niceness of kernel-thread/LWP. It
3907 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3908 // not the entire user process, and user level threads are 1:1 mapped to kernel
3909 // threads. It has always been the case, but could change in the future. For
3910 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3911 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3912
3913 int os::java_to_os_priority[CriticalPriority + 1] = {
3914 19, // 0 Entry should never be used
3915
3916 4, // 1 MinPriority
3917 3, // 2
3918 2, // 3
3919
3920 1, // 4
3921 0, // 5 NormPriority
3922 -1, // 6
3923
3924 -2, // 7
3925 -3, // 8
3926 -4, // 9 NearMaxPriority
3927
3928 -5, // 10 MaxPriority
3929
3930 -5 // 11 CriticalPriority
3931 };
3932
3933 static int prio_init() {
3934 if (ThreadPriorityPolicy == 1) {
3935 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3936 // if effective uid is not root. Perhaps, a more elegant way of doing
3937 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3938 if (geteuid() != 0) {
3939 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3940 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3941 }
3942 ThreadPriorityPolicy = 0;
3943 }
3944 }
3945 if (UseCriticalJavaThreadPriority) {
3946 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3947 }
3948 return 0;
3949 }
3950
3951 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3952 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3953
3954 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3955 return (ret == 0) ? OS_OK : OS_ERR;
3956 }
3957
3958 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3959 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3960 *priority_ptr = java_to_os_priority[NormPriority];
3961 return OS_OK;
3962 }
3963
3964 errno = 0;
3965 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3966 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3967 }
3968
3969 // Hint to the underlying OS that a task switch would not be good.
3970 // Void return because it's a hint and can fail.
3971 void os::hint_no_preempt() {}
3972
3973 ////////////////////////////////////////////////////////////////////////////////
3974 // suspend/resume support
3975
3976 // the low-level signal-based suspend/resume support is a remnant from the
3977 // old VM-suspension that used to be for java-suspension, safepoints etc,
3978 // within hotspot. Now there is a single use-case for this:
3979 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3980 // that runs in the watcher thread.
3981 // The remaining code is greatly simplified from the more general suspension
3982 // code that used to be used.
3983 //
3984 // The protocol is quite simple:
3985 // - suspend:
3986 // - sends a signal to the target thread
3987 // - polls the suspend state of the osthread using a yield loop
3988 // - target thread signal handler (SR_handler) sets suspend state
3989 // and blocks in sigsuspend until continued
3990 // - resume:
3991 // - sets target osthread state to continue
3992 // - sends signal to end the sigsuspend loop in the SR_handler
3993 //
3994 // Note that the SR_lock plays no role in this suspend/resume protocol.
3995 //
3996
3997 static void resume_clear_context(OSThread *osthread) {
3998 osthread->set_ucontext(NULL);
3999 osthread->set_siginfo(NULL);
4000 }
4001
4002 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4003 osthread->set_ucontext(context);
4004 osthread->set_siginfo(siginfo);
4005 }
4006
4007 //
4008 // Handler function invoked when a thread's execution is suspended or
4009 // resumed. We have to be careful that only async-safe functions are
4010 // called here (Note: most pthread functions are not async safe and
4011 // should be avoided.)
4012 //
4013 // Note: sigwait() is a more natural fit than sigsuspend() from an
4014 // interface point of view, but sigwait() prevents the signal hander
4015 // from being run. libpthread would get very confused by not having
4016 // its signal handlers run and prevents sigwait()'s use with the
4017 // mutex granting granting signal.
4018 //
4019 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4020 //
4021 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4022 // Save and restore errno to avoid confusing native code with EINTR
4023 // after sigsuspend.
4024 int old_errno = errno;
4025
4026 Thread* thread = Thread::current();
4027 OSThread* osthread = thread->osthread();
4028 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4029
4030 os::SuspendResume::State current = osthread->sr.state();
4031 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4032 suspend_save_context(osthread, siginfo, context);
4033
4034 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4035 os::SuspendResume::State state = osthread->sr.suspended();
4036 if (state == os::SuspendResume::SR_SUSPENDED) {
4037 sigset_t suspend_set; // signals for sigsuspend()
4038
4039 // get current set of blocked signals and unblock resume signal
4040 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4041 sigdelset(&suspend_set, SR_signum);
4042
4043 sr_semaphore.signal();
4044 // wait here until we are resumed
4045 while (1) {
4046 sigsuspend(&suspend_set);
4047
4048 os::SuspendResume::State result = osthread->sr.running();
4049 if (result == os::SuspendResume::SR_RUNNING) {
4050 sr_semaphore.signal();
4051 break;
4052 }
4053 }
4054
4055 } else if (state == os::SuspendResume::SR_RUNNING) {
4056 // request was cancelled, continue
4057 } else {
4058 ShouldNotReachHere();
4059 }
4060
4061 resume_clear_context(osthread);
4062 } else if (current == os::SuspendResume::SR_RUNNING) {
4063 // request was cancelled, continue
4064 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4065 // ignore
4066 } else {
4067 // ignore
4068 }
4069
4070 errno = old_errno;
4071 }
4072
4073
4074 static int SR_initialize() {
4075 struct sigaction act;
4076 char *s;
4077 /* Get signal number to use for suspend/resume */
4078 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4079 int sig = ::strtol(s, 0, 10);
4080 if (sig > 0 || sig < _NSIG) {
4081 SR_signum = sig;
4082 }
4083 }
4084
4085 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4086 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4087
4088 sigemptyset(&SR_sigset);
4089 sigaddset(&SR_sigset, SR_signum);
4090
4091 /* Set up signal handler for suspend/resume */
4092 act.sa_flags = SA_RESTART|SA_SIGINFO;
4093 act.sa_handler = (void (*)(int)) SR_handler;
4094
4095 // SR_signum is blocked by default.
4096 // 4528190 - We also need to block pthread restart signal (32 on all
4097 // supported Linux platforms). Note that LinuxThreads need to block
4098 // this signal for all threads to work properly. So we don't have
4099 // to use hard-coded signal number when setting up the mask.
4100 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4101
4102 if (sigaction(SR_signum, &act, 0) == -1) {
4103 return -1;
4104 }
4105
4106 // Save signal flag
4107 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4108 return 0;
4109 }
4110
4111 static int sr_notify(OSThread* osthread) {
4112 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4113 assert_status(status == 0, status, "pthread_kill");
4114 return status;
4115 }
4116
4117 // "Randomly" selected value for how long we want to spin
4118 // before bailing out on suspending a thread, also how often
4119 // we send a signal to a thread we want to resume
4120 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4121 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4122
4123 // returns true on success and false on error - really an error is fatal
4124 // but this seems the normal response to library errors
4125 static bool do_suspend(OSThread* osthread) {
4126 assert(osthread->sr.is_running(), "thread should be running");
4127 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4128
4129 // mark as suspended and send signal
4130 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4131 // failed to switch, state wasn't running?
4132 ShouldNotReachHere();
4133 return false;
4134 }
4135
4136 if (sr_notify(osthread) != 0) {
4137 ShouldNotReachHere();
4138 }
4139
4140 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4141 while (true) {
4142 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4143 break;
4144 } else {
4145 // timeout
4146 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4147 if (cancelled == os::SuspendResume::SR_RUNNING) {
4148 return false;
4149 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4150 // make sure that we consume the signal on the semaphore as well
4151 sr_semaphore.wait();
4152 break;
4153 } else {
4154 ShouldNotReachHere();
4155 return false;
4156 }
4157 }
4158 }
4159
4160 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4161 return true;
4162 }
4163
4164 static void do_resume(OSThread* osthread) {
4165 assert(osthread->sr.is_suspended(), "thread should be suspended");
4166 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4167
4168 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4169 // failed to switch to WAKEUP_REQUEST
4170 ShouldNotReachHere();
4171 return;
4172 }
4173
4174 while (true) {
4175 if (sr_notify(osthread) == 0) {
4176 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4177 if (osthread->sr.is_running()) {
4178 return;
4179 }
4180 }
4181 } else {
4182 ShouldNotReachHere();
4183 }
4184 }
4185
4186 guarantee(osthread->sr.is_running(), "Must be running!");
4187 }
4188
4189 ////////////////////////////////////////////////////////////////////////////////
4190 // interrupt support
4191
4192 void os::interrupt(Thread* thread) {
4193 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4194 "possibility of dangling Thread pointer");
4195
4196 OSThread* osthread = thread->osthread();
4197
4198 if (!osthread->interrupted()) {
4199 osthread->set_interrupted(true);
4200 // More than one thread can get here with the same value of osthread,
4201 // resulting in multiple notifications. We do, however, want the store
4202 // to interrupted() to be visible to other threads before we execute unpark().
4203 OrderAccess::fence();
4204 ParkEvent * const slp = thread->_SleepEvent ;
4205 if (slp != NULL) slp->unpark() ;
4206 }
4207
4208 // For JSR166. Unpark even if interrupt status already was set
4209 if (thread->is_Java_thread())
4210 ((JavaThread*)thread)->parker()->unpark();
4211
4212 ParkEvent * ev = thread->_ParkEvent ;
4213 if (ev != NULL) ev->unpark() ;
4214
4215 }
4216
4217 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4218 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4219 "possibility of dangling Thread pointer");
4220
4221 OSThread* osthread = thread->osthread();
4222
4223 bool interrupted = osthread->interrupted();
4224
4225 if (interrupted && clear_interrupted) {
4226 osthread->set_interrupted(false);
4227 // consider thread->_SleepEvent->reset() ... optional optimization
4228 }
4229
4230 return interrupted;
4231 }
4232
4233 ///////////////////////////////////////////////////////////////////////////////////
4234 // signal handling (except suspend/resume)
4235
4236 // This routine may be used by user applications as a "hook" to catch signals.
4237 // The user-defined signal handler must pass unrecognized signals to this
4238 // routine, and if it returns true (non-zero), then the signal handler must
4239 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4240 // routine will never retun false (zero), but instead will execute a VM panic
4241 // routine kill the process.
4242 //
4243 // If this routine returns false, it is OK to call it again. This allows
4244 // the user-defined signal handler to perform checks either before or after
4245 // the VM performs its own checks. Naturally, the user code would be making
4246 // a serious error if it tried to handle an exception (such as a null check
4247 // or breakpoint) that the VM was generating for its own correct operation.
4248 //
4249 // This routine may recognize any of the following kinds of signals:
4250 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4251 // It should be consulted by handlers for any of those signals.
4252 //
4253 // The caller of this routine must pass in the three arguments supplied
4254 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4255 // field of the structure passed to sigaction(). This routine assumes that
4256 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4257 //
4258 // Note that the VM will print warnings if it detects conflicting signal
4259 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4260 //
4261 extern "C" JNIEXPORT int
4262 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4263 void* ucontext, int abort_if_unrecognized);
4264
4265 void signalHandler(int sig, siginfo_t* info, void* uc) {
4266 assert(info != NULL && uc != NULL, "it must be old kernel");
4267 int orig_errno = errno; // Preserve errno value over signal handler.
4268 JVM_handle_linux_signal(sig, info, uc, true);
4269 errno = orig_errno;
4270 }
4271
4272
4273 // This boolean allows users to forward their own non-matching signals
4274 // to JVM_handle_linux_signal, harmlessly.
4275 bool os::Linux::signal_handlers_are_installed = false;
4276
4277 // For signal-chaining
4278 struct sigaction os::Linux::sigact[MAXSIGNUM];
4279 unsigned int os::Linux::sigs = 0;
4280 bool os::Linux::libjsig_is_loaded = false;
4281 typedef struct sigaction *(*get_signal_t)(int);
4282 get_signal_t os::Linux::get_signal_action = NULL;
4283
4284 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4285 struct sigaction *actp = NULL;
4286
4287 if (libjsig_is_loaded) {
4288 // Retrieve the old signal handler from libjsig
4289 actp = (*get_signal_action)(sig);
4290 }
4291 if (actp == NULL) {
4292 // Retrieve the preinstalled signal handler from jvm
4293 actp = get_preinstalled_handler(sig);
4294 }
4295
4296 return actp;
4297 }
4298
4299 static bool call_chained_handler(struct sigaction *actp, int sig,
4300 siginfo_t *siginfo, void *context) {
4301 // Call the old signal handler
4302 if (actp->sa_handler == SIG_DFL) {
4303 // It's more reasonable to let jvm treat it as an unexpected exception
4304 // instead of taking the default action.
4305 return false;
4306 } else if (actp->sa_handler != SIG_IGN) {
4307 if ((actp->sa_flags & SA_NODEFER) == 0) {
4308 // automaticlly block the signal
4309 sigaddset(&(actp->sa_mask), sig);
4310 }
4311
4312 sa_handler_t hand;
4313 sa_sigaction_t sa;
4314 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4315 // retrieve the chained handler
4316 if (siginfo_flag_set) {
4317 sa = actp->sa_sigaction;
4318 } else {
4319 hand = actp->sa_handler;
4320 }
4321
4322 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4323 actp->sa_handler = SIG_DFL;
4324 }
4325
4326 // try to honor the signal mask
4327 sigset_t oset;
4328 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4329
4330 // call into the chained handler
4331 if (siginfo_flag_set) {
4332 (*sa)(sig, siginfo, context);
4333 } else {
4334 (*hand)(sig);
4335 }
4336
4337 // restore the signal mask
4338 pthread_sigmask(SIG_SETMASK, &oset, 0);
4339 }
4340 // Tell jvm's signal handler the signal is taken care of.
4341 return true;
4342 }
4343
4344 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4345 bool chained = false;
4346 // signal-chaining
4347 if (UseSignalChaining) {
4348 struct sigaction *actp = get_chained_signal_action(sig);
4349 if (actp != NULL) {
4350 chained = call_chained_handler(actp, sig, siginfo, context);
4351 }
4352 }
4353 return chained;
4354 }
4355
4356 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4357 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4358 return &sigact[sig];
4359 }
4360 return NULL;
4361 }
4362
4363 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4364 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4365 sigact[sig] = oldAct;
4366 sigs |= (unsigned int)1 << sig;
4367 }
4368
4369 // for diagnostic
4370 int os::Linux::sigflags[MAXSIGNUM];
4371
4372 int os::Linux::get_our_sigflags(int sig) {
4373 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4374 return sigflags[sig];
4375 }
4376
4377 void os::Linux::set_our_sigflags(int sig, int flags) {
4378 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4379 sigflags[sig] = flags;
4380 }
4381
4382 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4383 // Check for overwrite.
4384 struct sigaction oldAct;
4385 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4386
4387 void* oldhand = oldAct.sa_sigaction
4388 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4389 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4390 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4391 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4392 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4393 if (AllowUserSignalHandlers || !set_installed) {
4394 // Do not overwrite; user takes responsibility to forward to us.
4395 return;
4396 } else if (UseSignalChaining) {
4397 // save the old handler in jvm
4398 save_preinstalled_handler(sig, oldAct);
4399 // libjsig also interposes the sigaction() call below and saves the
4400 // old sigaction on it own.
4401 } else {
4402 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4403 "%#lx for signal %d.", (long)oldhand, sig));
4404 }
4405 }
4406
4407 struct sigaction sigAct;
4408 sigfillset(&(sigAct.sa_mask));
4409 sigAct.sa_handler = SIG_DFL;
4410 if (!set_installed) {
4411 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4412 } else {
4413 sigAct.sa_sigaction = signalHandler;
4414 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4415 }
4416 // Save flags, which are set by ours
4417 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4418 sigflags[sig] = sigAct.sa_flags;
4419
4420 int ret = sigaction(sig, &sigAct, &oldAct);
4421 assert(ret == 0, "check");
4422
4423 void* oldhand2 = oldAct.sa_sigaction
4424 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4425 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4426 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4427 }
4428
4429 // install signal handlers for signals that HotSpot needs to
4430 // handle in order to support Java-level exception handling.
4431
4432 void os::Linux::install_signal_handlers() {
4433 if (!signal_handlers_are_installed) {
4434 signal_handlers_are_installed = true;
4435
4436 // signal-chaining
4437 typedef void (*signal_setting_t)();
4438 signal_setting_t begin_signal_setting = NULL;
4439 signal_setting_t end_signal_setting = NULL;
4440 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4441 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4442 if (begin_signal_setting != NULL) {
4443 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4444 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4445 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4446 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4447 libjsig_is_loaded = true;
4448 assert(UseSignalChaining, "should enable signal-chaining");
4449 }
4450 if (libjsig_is_loaded) {
4451 // Tell libjsig jvm is setting signal handlers
4452 (*begin_signal_setting)();
4453 }
4454
4455 set_signal_handler(SIGSEGV, true);
4456 set_signal_handler(SIGPIPE, true);
4457 set_signal_handler(SIGBUS, true);
4458 set_signal_handler(SIGILL, true);
4459 set_signal_handler(SIGFPE, true);
4460 #if defined(PPC64)
4461 set_signal_handler(SIGTRAP, true);
4462 #endif
4463 set_signal_handler(SIGXFSZ, true);
4464
4465 if (libjsig_is_loaded) {
4466 // Tell libjsig jvm finishes setting signal handlers
4467 (*end_signal_setting)();
4468 }
4469
4470 // We don't activate signal checker if libjsig is in place, we trust ourselves
4471 // and if UserSignalHandler is installed all bets are off.
4472 // Log that signal checking is off only if -verbose:jni is specified.
4473 if (CheckJNICalls) {
4474 if (libjsig_is_loaded) {
4475 if (PrintJNIResolving) {
4476 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4477 }
4478 check_signals = false;
4479 }
4480 if (AllowUserSignalHandlers) {
4481 if (PrintJNIResolving) {
4482 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4483 }
4484 check_signals = false;
4485 }
4486 }
4487 }
4488 }
4489
4490 // This is the fastest way to get thread cpu time on Linux.
4491 // Returns cpu time (user+sys) for any thread, not only for current.
4492 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4493 // It might work on 2.6.10+ with a special kernel/glibc patch.
4494 // For reference, please, see IEEE Std 1003.1-2004:
4495 // http://www.unix.org/single_unix_specification
4496
4497 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4498 struct timespec tp;
4499 int rc = os::Linux::clock_gettime(clockid, &tp);
4500 assert(rc == 0, "clock_gettime is expected to return 0 code");
4501
4502 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4503 }
4504
4505 /////
4506 // glibc on Linux platform uses non-documented flag
4507 // to indicate, that some special sort of signal
4508 // trampoline is used.
4509 // We will never set this flag, and we should
4510 // ignore this flag in our diagnostic
4511 #ifdef SIGNIFICANT_SIGNAL_MASK
4512 #undef SIGNIFICANT_SIGNAL_MASK
4513 #endif
4514 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4515
4516 static const char* get_signal_handler_name(address handler,
4517 char* buf, int buflen) {
4518 int offset;
4519 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4520 if (found) {
4521 // skip directory names
4522 const char *p1, *p2;
4523 p1 = buf;
4524 size_t len = strlen(os::file_separator());
4525 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4526 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4527 } else {
4528 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4529 }
4530 return buf;
4531 }
4532
4533 static void print_signal_handler(outputStream* st, int sig,
4534 char* buf, size_t buflen) {
4535 struct sigaction sa;
4536
4537 sigaction(sig, NULL, &sa);
4538
4539 // See comment for SIGNIFICANT_SIGNAL_MASK define
4540 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4541
4542 st->print("%s: ", os::exception_name(sig, buf, buflen));
4543
4544 address handler = (sa.sa_flags & SA_SIGINFO)
4545 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4546 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4547
4548 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4549 st->print("SIG_DFL");
4550 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4551 st->print("SIG_IGN");
4552 } else {
4553 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4554 }
4555
4556 st->print(", sa_mask[0]=");
4557 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4558
4559 address rh = VMError::get_resetted_sighandler(sig);
4560 // May be, handler was resetted by VMError?
4561 if(rh != NULL) {
4562 handler = rh;
4563 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4564 }
4565
4566 st->print(", sa_flags=");
4567 os::Posix::print_sa_flags(st, sa.sa_flags);
4568
4569 // Check: is it our handler?
4570 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4571 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4572 // It is our signal handler
4573 // check for flags, reset system-used one!
4574 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4575 st->print(
4576 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4577 os::Linux::get_our_sigflags(sig));
4578 }
4579 }
4580 st->cr();
4581 }
4582
4583
4584 #define DO_SIGNAL_CHECK(sig) \
4585 if (!sigismember(&check_signal_done, sig)) \
4586 os::Linux::check_signal_handler(sig)
4587
4588 // This method is a periodic task to check for misbehaving JNI applications
4589 // under CheckJNI, we can add any periodic checks here
4590
4591 void os::run_periodic_checks() {
4592
4593 if (check_signals == false) return;
4594
4595 // SEGV and BUS if overridden could potentially prevent
4596 // generation of hs*.log in the event of a crash, debugging
4597 // such a case can be very challenging, so we absolutely
4598 // check the following for a good measure:
4599 DO_SIGNAL_CHECK(SIGSEGV);
4600 DO_SIGNAL_CHECK(SIGILL);
4601 DO_SIGNAL_CHECK(SIGFPE);
4602 DO_SIGNAL_CHECK(SIGBUS);
4603 DO_SIGNAL_CHECK(SIGPIPE);
4604 DO_SIGNAL_CHECK(SIGXFSZ);
4605 #if defined(PPC64)
4606 DO_SIGNAL_CHECK(SIGTRAP);
4607 #endif
4608
4609 // ReduceSignalUsage allows the user to override these handlers
4610 // see comments at the very top and jvm_solaris.h
4611 if (!ReduceSignalUsage) {
4612 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4613 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4614 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4615 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4616 }
4617
4618 DO_SIGNAL_CHECK(SR_signum);
4619 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4620 }
4621
4622 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4623
4624 static os_sigaction_t os_sigaction = NULL;
4625
4626 void os::Linux::check_signal_handler(int sig) {
4627 char buf[O_BUFLEN];
4628 address jvmHandler = NULL;
4629
4630
4631 struct sigaction act;
4632 if (os_sigaction == NULL) {
4633 // only trust the default sigaction, in case it has been interposed
4634 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4635 if (os_sigaction == NULL) return;
4636 }
4637
4638 os_sigaction(sig, (struct sigaction*)NULL, &act);
4639
4640
4641 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4642
4643 address thisHandler = (act.sa_flags & SA_SIGINFO)
4644 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4645 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4646
4647
4648 switch(sig) {
4649 case SIGSEGV:
4650 case SIGBUS:
4651 case SIGFPE:
4652 case SIGPIPE:
4653 case SIGILL:
4654 case SIGXFSZ:
4655 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4656 break;
4657
4658 case SHUTDOWN1_SIGNAL:
4659 case SHUTDOWN2_SIGNAL:
4660 case SHUTDOWN3_SIGNAL:
4661 case BREAK_SIGNAL:
4662 jvmHandler = (address)user_handler();
4663 break;
4664
4665 case INTERRUPT_SIGNAL:
4666 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4667 break;
4668
4669 default:
4670 if (sig == SR_signum) {
4671 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4672 } else {
4673 return;
4674 }
4675 break;
4676 }
4677
4678 if (thisHandler != jvmHandler) {
4679 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4680 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4681 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4682 // No need to check this sig any longer
4683 sigaddset(&check_signal_done, sig);
4684 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4685 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4686 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4687 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4688 // No need to check this sig any longer
4689 sigaddset(&check_signal_done, sig);
4690 }
4691
4692 // Dump all the signal
4693 if (sigismember(&check_signal_done, sig)) {
4694 print_signal_handlers(tty, buf, O_BUFLEN);
4695 }
4696 }
4697
4698 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4699
4700 extern bool signal_name(int signo, char* buf, size_t len);
4701
4702 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4703 if (0 < exception_code && exception_code <= SIGRTMAX) {
4704 // signal
4705 if (!signal_name(exception_code, buf, size)) {
4706 jio_snprintf(buf, size, "SIG%d", exception_code);
4707 }
4708 return buf;
4709 } else {
4710 return NULL;
4711 }
4712 }
4713
4714 // this is called _before_ the most of global arguments have been parsed
4715 void os::init(void) {
4716 char dummy; /* used to get a guess on initial stack address */
4717 // first_hrtime = gethrtime();
4718
4719 // With LinuxThreads the JavaMain thread pid (primordial thread)
4720 // is different than the pid of the java launcher thread.
4721 // So, on Linux, the launcher thread pid is passed to the VM
4722 // via the sun.java.launcher.pid property.
4723 // Use this property instead of getpid() if it was correctly passed.
4724 // See bug 6351349.
4725 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4726
4727 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4728
4729 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4730
4731 init_random(1234567);
4732
4733 ThreadCritical::initialize();
4734
4735 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4736 if (Linux::page_size() == -1) {
4737 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4738 strerror(errno)));
4739 }
4740 init_page_sizes((size_t) Linux::page_size());
4741
4742 Linux::initialize_system_info();
4743
4744 // main_thread points to the aboriginal thread
4745 Linux::_main_thread = pthread_self();
4746
4747 Linux::clock_init();
4748 initial_time_count = javaTimeNanos();
4749
4750 // pthread_condattr initialization for monotonic clock
4751 int status;
4752 pthread_condattr_t* _condattr = os::Linux::condAttr();
4753 if ((status = pthread_condattr_init(_condattr)) != 0) {
4754 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4755 }
4756 // Only set the clock if CLOCK_MONOTONIC is available
4757 if (Linux::supports_monotonic_clock()) {
4758 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4759 if (status == EINVAL) {
4760 warning("Unable to use monotonic clock with relative timed-waits" \
4761 " - changes to the time-of-day clock may have adverse affects");
4762 } else {
4763 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4764 }
4765 }
4766 }
4767 // else it defaults to CLOCK_REALTIME
4768
4769 pthread_mutex_init(&dl_mutex, NULL);
4770
4771 // If the pagesize of the VM is greater than 8K determine the appropriate
4772 // number of initial guard pages. The user can change this with the
4773 // command line arguments, if needed.
4774 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4775 StackYellowPages = 1;
4776 StackRedPages = 1;
4777 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4778 }
4779 }
4780
4781 // To install functions for atexit system call
4782 extern "C" {
4783 static void perfMemory_exit_helper() {
4784 perfMemory_exit();
4785 }
4786 }
4787
4788 // this is called _after_ the global arguments have been parsed
4789 jint os::init_2(void)
4790 {
4791 Linux::fast_thread_clock_init();
4792
4793 // Allocate a single page and mark it as readable for safepoint polling
4794 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4795 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4796
4797 os::set_polling_page( polling_page );
4798
4799 #ifndef PRODUCT
4800 if(Verbose && PrintMiscellaneous)
4801 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4802 #endif
4803
4804 if (!UseMembar) {
4805 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4806 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4807 os::set_memory_serialize_page( mem_serialize_page );
4808
4809 #ifndef PRODUCT
4810 if(Verbose && PrintMiscellaneous)
4811 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4812 #endif
4813 }
4814
4815 // initialize suspend/resume support - must do this before signal_sets_init()
4816 if (SR_initialize() != 0) {
4817 perror("SR_initialize failed");
4818 return JNI_ERR;
4819 }
4820
4821 Linux::signal_sets_init();
4822 Linux::install_signal_handlers();
4823
4824 // Check minimum allowable stack size for thread creation and to initialize
4825 // the java system classes, including StackOverflowError - depends on page
4826 // size. Add a page for compiler2 recursion in main thread.
4827 // Add in 2*BytesPerWord times page size to account for VM stack during
4828 // class initialization depending on 32 or 64 bit VM.
4829 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4830 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4831 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4832
4833 size_t threadStackSizeInBytes = ThreadStackSize * K;
4834 if (threadStackSizeInBytes != 0 &&
4835 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4836 tty->print_cr("\nThe stack size specified is too small, "
4837 "Specify at least %dk",
4838 os::Linux::min_stack_allowed/ K);
4839 return JNI_ERR;
4840 }
4841
4842 // Make the stack size a multiple of the page size so that
4843 // the yellow/red zones can be guarded.
4844 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4845 vm_page_size()));
4846
4847 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4848
4849 #if defined(IA32)
4850 workaround_expand_exec_shield_cs_limit();
4851 #endif
4852
4853 Linux::libpthread_init();
4854 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4855 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4856 Linux::glibc_version(), Linux::libpthread_version(),
4857 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4858 }
4859
4860 if (UseNUMA) {
4861 if (!Linux::libnuma_init()) {
4862 UseNUMA = false;
4863 } else {
4864 if ((Linux::numa_max_node() < 1)) {
4865 // There's only one node(they start from 0), disable NUMA.
4866 UseNUMA = false;
4867 }
4868 }
4869 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4870 // we can make the adaptive lgrp chunk resizing work. If the user specified
4871 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4872 // disable adaptive resizing.
4873 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4874 if (FLAG_IS_DEFAULT(UseNUMA)) {
4875 UseNUMA = false;
4876 } else {
4877 if (FLAG_IS_DEFAULT(UseLargePages) &&
4878 FLAG_IS_DEFAULT(UseSHM) &&
4879 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4880 UseLargePages = false;
4881 } else {
4882 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4883 UseAdaptiveSizePolicy = false;
4884 UseAdaptiveNUMAChunkSizing = false;
4885 }
4886 }
4887 }
4888 if (!UseNUMA && ForceNUMA) {
4889 UseNUMA = true;
4890 }
4891 }
4892
4893 if (MaxFDLimit) {
4894 // set the number of file descriptors to max. print out error
4895 // if getrlimit/setrlimit fails but continue regardless.
4896 struct rlimit nbr_files;
4897 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4898 if (status != 0) {
4899 if (PrintMiscellaneous && (Verbose || WizardMode))
4900 perror("os::init_2 getrlimit failed");
4901 } else {
4902 nbr_files.rlim_cur = nbr_files.rlim_max;
4903 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4904 if (status != 0) {
4905 if (PrintMiscellaneous && (Verbose || WizardMode))
4906 perror("os::init_2 setrlimit failed");
4907 }
4908 }
4909 }
4910
4911 // Initialize lock used to serialize thread creation (see os::create_thread)
4912 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4913
4914 // at-exit methods are called in the reverse order of their registration.
4915 // atexit functions are called on return from main or as a result of a
4916 // call to exit(3C). There can be only 32 of these functions registered
4917 // and atexit() does not set errno.
4918
4919 if (PerfAllowAtExitRegistration) {
4920 // only register atexit functions if PerfAllowAtExitRegistration is set.
4921 // atexit functions can be delayed until process exit time, which
4922 // can be problematic for embedded VM situations. Embedded VMs should
4923 // call DestroyJavaVM() to assure that VM resources are released.
4924
4925 // note: perfMemory_exit_helper atexit function may be removed in
4926 // the future if the appropriate cleanup code can be added to the
4927 // VM_Exit VMOperation's doit method.
4928 if (atexit(perfMemory_exit_helper) != 0) {
4929 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4930 }
4931 }
4932
4933 // initialize thread priority policy
4934 prio_init();
4935
4936 return JNI_OK;
4937 }
4938
4939 // this is called at the end of vm_initialization
4940 void os::init_3(void) {
4941 #ifdef JAVASE_EMBEDDED
4942 // Start the MemNotifyThread
4943 if (LowMemoryProtection) {
4944 MemNotifyThread::start();
4945 }
4946 return;
4947 #endif
4948 }
4949
4950 // Mark the polling page as unreadable
4951 void os::make_polling_page_unreadable(void) {
4952 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4953 fatal("Could not disable polling page");
4954 };
4955
4956 // Mark the polling page as readable
4957 void os::make_polling_page_readable(void) {
4958 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4959 fatal("Could not enable polling page");
4960 }
4961 };
4962
4963 int os::active_processor_count() {
4964 // Linux doesn't yet have a (official) notion of processor sets,
4965 // so just return the number of online processors.
4966 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4967 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4968 return online_cpus;
4969 }
4970
4971 void os::set_native_thread_name(const char *name) {
4972 // Not yet implemented.
4973 return;
4974 }
4975
4976 bool os::distribute_processes(uint length, uint* distribution) {
4977 // Not yet implemented.
4978 return false;
4979 }
4980
4981 bool os::bind_to_processor(uint processor_id) {
4982 // Not yet implemented.
4983 return false;
4984 }
4985
4986 ///
4987
4988 void os::SuspendedThreadTask::internal_do_task() {
4989 if (do_suspend(_thread->osthread())) {
4990 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4991 do_task(context);
4992 do_resume(_thread->osthread());
4993 }
4994 }
4995
4996 class PcFetcher : public os::SuspendedThreadTask {
4997 public:
4998 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4999 ExtendedPC result();
5000 protected:
5001 void do_task(const os::SuspendedThreadTaskContext& context);
5002 private:
5003 ExtendedPC _epc;
5004 };
5005
5006 ExtendedPC PcFetcher::result() {
5007 guarantee(is_done(), "task is not done yet.");
5008 return _epc;
5009 }
5010
5011 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5012 Thread* thread = context.thread();
5013 OSThread* osthread = thread->osthread();
5014 if (osthread->ucontext() != NULL) {
5015 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5016 } else {
5017 // NULL context is unexpected, double-check this is the VMThread
5018 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5019 }
5020 }
5021
5022 // Suspends the target using the signal mechanism and then grabs the PC before
5023 // resuming the target. Used by the flat-profiler only
5024 ExtendedPC os::get_thread_pc(Thread* thread) {
5025 // Make sure that it is called by the watcher for the VMThread
5026 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5027 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5028
5029 PcFetcher fetcher(thread);
5030 fetcher.run();
5031 return fetcher.result();
5032 }
5033
5034 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5035 {
5036 if (is_NPTL()) {
5037 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5038 } else {
5039 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5040 // word back to default 64bit precision if condvar is signaled. Java
5041 // wants 53bit precision. Save and restore current value.
5042 int fpu = get_fpu_control_word();
5043 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5044 set_fpu_control_word(fpu);
5045 return status;
5046 }
5047 }
5048
5049 ////////////////////////////////////////////////////////////////////////////////
5050 // debug support
5051
5052 bool os::find(address addr, outputStream* st) {
5053 Dl_info dlinfo;
5054 memset(&dlinfo, 0, sizeof(dlinfo));
5055 if (dladdr(addr, &dlinfo) != 0) {
5056 st->print(PTR_FORMAT ": ", addr);
5057 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5058 st->print("%s+%#x", dlinfo.dli_sname,
5059 addr - (intptr_t)dlinfo.dli_saddr);
5060 } else if (dlinfo.dli_fbase != NULL) {
5061 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5062 } else {
5063 st->print("<absolute address>");
5064 }
5065 if (dlinfo.dli_fname != NULL) {
5066 st->print(" in %s", dlinfo.dli_fname);
5067 }
5068 if (dlinfo.dli_fbase != NULL) {
5069 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5070 }
5071 st->cr();
5072
5073 if (Verbose) {
5074 // decode some bytes around the PC
5075 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5076 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5077 address lowest = (address) dlinfo.dli_sname;
5078 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5079 if (begin < lowest) begin = lowest;
5080 Dl_info dlinfo2;
5081 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5082 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5083 end = (address) dlinfo2.dli_saddr;
5084 Disassembler::decode(begin, end, st);
5085 }
5086 return true;
5087 }
5088 return false;
5089 }
5090
5091 ////////////////////////////////////////////////////////////////////////////////
5092 // misc
5093
5094 // This does not do anything on Linux. This is basically a hook for being
5095 // able to use structured exception handling (thread-local exception filters)
5096 // on, e.g., Win32.
5097 void
5098 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5099 JavaCallArguments* args, Thread* thread) {
5100 f(value, method, args, thread);
5101 }
5102
5103 void os::print_statistics() {
5104 }
5105
5106 int os::message_box(const char* title, const char* message) {
5107 int i;
5108 fdStream err(defaultStream::error_fd());
5109 for (i = 0; i < 78; i++) err.print_raw("=");
5110 err.cr();
5111 err.print_raw_cr(title);
5112 for (i = 0; i < 78; i++) err.print_raw("-");
5113 err.cr();
5114 err.print_raw_cr(message);
5115 for (i = 0; i < 78; i++) err.print_raw("=");
5116 err.cr();
5117
5118 char buf[16];
5119 // Prevent process from exiting upon "read error" without consuming all CPU
5120 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5121
5122 return buf[0] == 'y' || buf[0] == 'Y';
5123 }
5124
5125 int os::stat(const char *path, struct stat *sbuf) {
5126 char pathbuf[MAX_PATH];
5127 if (strlen(path) > MAX_PATH - 1) {
5128 errno = ENAMETOOLONG;
5129 return -1;
5130 }
5131 os::native_path(strcpy(pathbuf, path));
5132 return ::stat(pathbuf, sbuf);
5133 }
5134
5135 bool os::check_heap(bool force) {
5136 return true;
5137 }
5138
5139 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5140 return ::vsnprintf(buf, count, format, args);
5141 }
5142
5143 // Is a (classpath) directory empty?
5144 bool os::dir_is_empty(const char* path) {
5145 DIR *dir = NULL;
5146 struct dirent *ptr;
5147
5148 dir = opendir(path);
5149 if (dir == NULL) return true;
5150
5151 /* Scan the directory */
5152 bool result = true;
5153 char buf[sizeof(struct dirent) + MAX_PATH];
5154 while (result && (ptr = ::readdir(dir)) != NULL) {
5155 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5156 result = false;
5157 }
5158 }
5159 closedir(dir);
5160 return result;
5161 }
5162
5163 // This code originates from JDK's sysOpen and open64_w
5164 // from src/solaris/hpi/src/system_md.c
5165
5166 #ifndef O_DELETE
5167 #define O_DELETE 0x10000
5168 #endif
5169
5170 // Open a file. Unlink the file immediately after open returns
5171 // if the specified oflag has the O_DELETE flag set.
5172 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5173
5174 int os::open(const char *path, int oflag, int mode) {
5175
5176 if (strlen(path) > MAX_PATH - 1) {
5177 errno = ENAMETOOLONG;
5178 return -1;
5179 }
5180 int fd;
5181 int o_delete = (oflag & O_DELETE);
5182 oflag = oflag & ~O_DELETE;
5183
5184 fd = ::open64(path, oflag, mode);
5185 if (fd == -1) return -1;
5186
5187 //If the open succeeded, the file might still be a directory
5188 {
5189 struct stat64 buf64;
5190 int ret = ::fstat64(fd, &buf64);
5191 int st_mode = buf64.st_mode;
5192
5193 if (ret != -1) {
5194 if ((st_mode & S_IFMT) == S_IFDIR) {
5195 errno = EISDIR;
5196 ::close(fd);
5197 return -1;
5198 }
5199 } else {
5200 ::close(fd);
5201 return -1;
5202 }
5203 }
5204
5205 /*
5206 * All file descriptors that are opened in the JVM and not
5207 * specifically destined for a subprocess should have the
5208 * close-on-exec flag set. If we don't set it, then careless 3rd
5209 * party native code might fork and exec without closing all
5210 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5211 * UNIXProcess.c), and this in turn might:
5212 *
5213 * - cause end-of-file to fail to be detected on some file
5214 * descriptors, resulting in mysterious hangs, or
5215 *
5216 * - might cause an fopen in the subprocess to fail on a system
5217 * suffering from bug 1085341.
5218 *
5219 * (Yes, the default setting of the close-on-exec flag is a Unix
5220 * design flaw)
5221 *
5222 * See:
5223 * 1085341: 32-bit stdio routines should support file descriptors >255
5224 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5225 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5226 */
5227 #ifdef FD_CLOEXEC
5228 {
5229 int flags = ::fcntl(fd, F_GETFD);
5230 if (flags != -1)
5231 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5232 }
5233 #endif
5234
5235 if (o_delete != 0) {
5236 ::unlink(path);
5237 }
5238 return fd;
5239 }
5240
5241
5242 // create binary file, rewriting existing file if required
5243 int os::create_binary_file(const char* path, bool rewrite_existing) {
5244 int oflags = O_WRONLY | O_CREAT;
5245 if (!rewrite_existing) {
5246 oflags |= O_EXCL;
5247 }
5248 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5249 }
5250
5251 // return current position of file pointer
5252 jlong os::current_file_offset(int fd) {
5253 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5254 }
5255
5256 // move file pointer to the specified offset
5257 jlong os::seek_to_file_offset(int fd, jlong offset) {
5258 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5259 }
5260
5261 // This code originates from JDK's sysAvailable
5262 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5263
5264 int os::available(int fd, jlong *bytes) {
5265 jlong cur, end;
5266 int mode;
5267 struct stat64 buf64;
5268
5269 if (::fstat64(fd, &buf64) >= 0) {
5270 mode = buf64.st_mode;
5271 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5272 /*
5273 * XXX: is the following call interruptible? If so, this might
5274 * need to go through the INTERRUPT_IO() wrapper as for other
5275 * blocking, interruptible calls in this file.
5276 */
5277 int n;
5278 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5279 *bytes = n;
5280 return 1;
5281 }
5282 }
5283 }
5284 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5285 return 0;
5286 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5287 return 0;
5288 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5289 return 0;
5290 }
5291 *bytes = end - cur;
5292 return 1;
5293 }
5294
5295 int os::socket_available(int fd, jint *pbytes) {
5296 // Linux doc says EINTR not returned, unlike Solaris
5297 int ret = ::ioctl(fd, FIONREAD, pbytes);
5298
5299 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5300 // is expected to return 0 on failure and 1 on success to the jdk.
5301 return (ret < 0) ? 0 : 1;
5302 }
5303
5304 // Map a block of memory.
5305 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5306 char *addr, size_t bytes, bool read_only,
5307 bool allow_exec) {
5308 int prot;
5309 int flags = MAP_PRIVATE;
5310
5311 if (read_only) {
5312 prot = PROT_READ;
5313 } else {
5314 prot = PROT_READ | PROT_WRITE;
5315 }
5316
5317 if (allow_exec) {
5318 prot |= PROT_EXEC;
5319 }
5320
5321 if (addr != NULL) {
5322 flags |= MAP_FIXED;
5323 }
5324
5325 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5326 fd, file_offset);
5327 if (mapped_address == MAP_FAILED) {
5328 return NULL;
5329 }
5330 return mapped_address;
5331 }
5332
5333
5334 // Remap a block of memory.
5335 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5336 char *addr, size_t bytes, bool read_only,
5337 bool allow_exec) {
5338 // same as map_memory() on this OS
5339 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5340 allow_exec);
5341 }
5342
5343
5344 // Unmap a block of memory.
5345 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5346 return munmap(addr, bytes) == 0;
5347 }
5348
5349 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5350
5351 static clockid_t thread_cpu_clockid(Thread* thread) {
5352 pthread_t tid = thread->osthread()->pthread_id();
5353 clockid_t clockid;
5354
5355 // Get thread clockid
5356 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5357 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5358 return clockid;
5359 }
5360
5361 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5362 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5363 // of a thread.
5364 //
5365 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5366 // the fast estimate available on the platform.
5367
5368 jlong os::current_thread_cpu_time() {
5369 if (os::Linux::supports_fast_thread_cpu_time()) {
5370 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5371 } else {
5372 // return user + sys since the cost is the same
5373 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5374 }
5375 }
5376
5377 jlong os::thread_cpu_time(Thread* thread) {
5378 // consistent with what current_thread_cpu_time() returns
5379 if (os::Linux::supports_fast_thread_cpu_time()) {
5380 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5381 } else {
5382 return slow_thread_cpu_time(thread, true /* user + sys */);
5383 }
5384 }
5385
5386 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5387 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5388 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5389 } else {
5390 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5391 }
5392 }
5393
5394 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5395 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5396 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5397 } else {
5398 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5399 }
5400 }
5401
5402 //
5403 // -1 on error.
5404 //
5405
5406 PRAGMA_DIAG_PUSH
5407 PRAGMA_FORMAT_NONLITERAL_IGNORED
5408 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5409 static bool proc_task_unchecked = true;
5410 static const char *proc_stat_path = "/proc/%d/stat";
5411 pid_t tid = thread->osthread()->thread_id();
5412 char *s;
5413 char stat[2048];
5414 int statlen;
5415 char proc_name[64];
5416 int count;
5417 long sys_time, user_time;
5418 char cdummy;
5419 int idummy;
5420 long ldummy;
5421 FILE *fp;
5422
5423 // The /proc/<tid>/stat aggregates per-process usage on
5424 // new Linux kernels 2.6+ where NPTL is supported.
5425 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5426 // See bug 6328462.
5427 // There possibly can be cases where there is no directory
5428 // /proc/self/task, so we check its availability.
5429 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5430 // This is executed only once
5431 proc_task_unchecked = false;
5432 fp = fopen("/proc/self/task", "r");
5433 if (fp != NULL) {
5434 proc_stat_path = "/proc/self/task/%d/stat";
5435 fclose(fp);
5436 }
5437 }
5438
5439 sprintf(proc_name, proc_stat_path, tid);
5440 fp = fopen(proc_name, "r");
5441 if ( fp == NULL ) return -1;
5442 statlen = fread(stat, 1, 2047, fp);
5443 stat[statlen] = '\0';
5444 fclose(fp);
5445
5446 // Skip pid and the command string. Note that we could be dealing with
5447 // weird command names, e.g. user could decide to rename java launcher
5448 // to "java 1.4.2 :)", then the stat file would look like
5449 // 1234 (java 1.4.2 :)) R ... ...
5450 // We don't really need to know the command string, just find the last
5451 // occurrence of ")" and then start parsing from there. See bug 4726580.
5452 s = strrchr(stat, ')');
5453 if (s == NULL ) return -1;
5454
5455 // Skip blank chars
5456 do s++; while (isspace(*s));
5457
5458 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5459 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5460 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5461 &user_time, &sys_time);
5462 if ( count != 13 ) return -1;
5463 if (user_sys_cpu_time) {
5464 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5465 } else {
5466 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5467 }
5468 }
5469 PRAGMA_DIAG_POP
5470
5471 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5472 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5473 info_ptr->may_skip_backward = false; // elapsed time not wall time
5474 info_ptr->may_skip_forward = false; // elapsed time not wall time
5475 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5476 }
5477
5478 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5479 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5480 info_ptr->may_skip_backward = false; // elapsed time not wall time
5481 info_ptr->may_skip_forward = false; // elapsed time not wall time
5482 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5483 }
5484
5485 bool os::is_thread_cpu_time_supported() {
5486 return true;
5487 }
5488
5489 // System loadavg support. Returns -1 if load average cannot be obtained.
5490 // Linux doesn't yet have a (official) notion of processor sets,
5491 // so just return the system wide load average.
5492 int os::loadavg(double loadavg[], int nelem) {
5493 return ::getloadavg(loadavg, nelem);
5494 }
5495
5496 void os::pause() {
5497 char filename[MAX_PATH];
5498 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5499 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5500 } else {
5501 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5502 }
5503
5504 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5505 if (fd != -1) {
5506 struct stat buf;
5507 ::close(fd);
5508 while (::stat(filename, &buf) == 0) {
5509 (void)::poll(NULL, 0, 100);
5510 }
5511 } else {
5512 jio_fprintf(stderr,
5513 "Could not open pause file '%s', continuing immediately.\n", filename);
5514 }
5515 }
5516
5517
5518 // Refer to the comments in os_solaris.cpp park-unpark.
5519 //
5520 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5521 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5522 // For specifics regarding the bug see GLIBC BUGID 261237 :
5523 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5524 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5525 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5526 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5527 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5528 // and monitorenter when we're using 1-0 locking. All those operations may result in
5529 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5530 // of libpthread avoids the problem, but isn't practical.
5531 //
5532 // Possible remedies:
5533 //
5534 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5535 // This is palliative and probabilistic, however. If the thread is preempted
5536 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5537 // than the minimum period may have passed, and the abstime may be stale (in the
5538 // past) resultin in a hang. Using this technique reduces the odds of a hang
5539 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5540 //
5541 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5542 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5543 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5544 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5545 // thread.
5546 //
5547 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5548 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5549 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5550 // This also works well. In fact it avoids kernel-level scalability impediments
5551 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5552 // timers in a graceful fashion.
5553 //
5554 // 4. When the abstime value is in the past it appears that control returns
5555 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5556 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5557 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5558 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5559 // It may be possible to avoid reinitialization by checking the return
5560 // value from pthread_cond_timedwait(). In addition to reinitializing the
5561 // condvar we must establish the invariant that cond_signal() is only called
5562 // within critical sections protected by the adjunct mutex. This prevents
5563 // cond_signal() from "seeing" a condvar that's in the midst of being
5564 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5565 // desirable signal-after-unlock optimization that avoids futile context switching.
5566 //
5567 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5568 // structure when a condvar is used or initialized. cond_destroy() would
5569 // release the helper structure. Our reinitialize-after-timedwait fix
5570 // put excessive stress on malloc/free and locks protecting the c-heap.
5571 //
5572 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5573 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5574 // and only enabling the work-around for vulnerable environments.
5575
5576 // utility to compute the abstime argument to timedwait:
5577 // millis is the relative timeout time
5578 // abstime will be the absolute timeout time
5579 // TODO: replace compute_abstime() with unpackTime()
5580
5581 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5582 if (millis < 0) millis = 0;
5583
5584 jlong seconds = millis / 1000;
5585 millis %= 1000;
5586 if (seconds > 50000000) { // see man cond_timedwait(3T)
5587 seconds = 50000000;
5588 }
5589
5590 if (os::Linux::supports_monotonic_clock()) {
5591 struct timespec now;
5592 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5593 assert_status(status == 0, status, "clock_gettime");
5594 abstime->tv_sec = now.tv_sec + seconds;
5595 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5596 if (nanos >= NANOSECS_PER_SEC) {
5597 abstime->tv_sec += 1;
5598 nanos -= NANOSECS_PER_SEC;
5599 }
5600 abstime->tv_nsec = nanos;
5601 } else {
5602 struct timeval now;
5603 int status = gettimeofday(&now, NULL);
5604 assert(status == 0, "gettimeofday");
5605 abstime->tv_sec = now.tv_sec + seconds;
5606 long usec = now.tv_usec + millis * 1000;
5607 if (usec >= 1000000) {
5608 abstime->tv_sec += 1;
5609 usec -= 1000000;
5610 }
5611 abstime->tv_nsec = usec * 1000;
5612 }
5613 return abstime;
5614 }
5615
5616
5617 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5618 // Conceptually TryPark() should be equivalent to park(0).
5619
5620 int os::PlatformEvent::TryPark() {
5621 for (;;) {
5622 const int v = _Event ;
5623 guarantee ((v == 0) || (v == 1), "invariant") ;
5624 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5625 }
5626 }
5627
5628 void os::PlatformEvent::park() { // AKA "down()"
5629 // Invariant: Only the thread associated with the Event/PlatformEvent
5630 // may call park().
5631 // TODO: assert that _Assoc != NULL or _Assoc == Self
5632 int v ;
5633 for (;;) {
5634 v = _Event ;
5635 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5636 }
5637 guarantee (v >= 0, "invariant") ;
5638 if (v == 0) {
5639 // Do this the hard way by blocking ...
5640 int status = pthread_mutex_lock(_mutex);
5641 assert_status(status == 0, status, "mutex_lock");
5642 guarantee (_nParked == 0, "invariant") ;
5643 ++ _nParked ;
5644 while (_Event < 0) {
5645 status = pthread_cond_wait(_cond, _mutex);
5646 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5647 // Treat this the same as if the wait was interrupted
5648 if (status == ETIME) { status = EINTR; }
5649 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5650 }
5651 -- _nParked ;
5652
5653 _Event = 0 ;
5654 status = pthread_mutex_unlock(_mutex);
5655 assert_status(status == 0, status, "mutex_unlock");
5656 // Paranoia to ensure our locked and lock-free paths interact
5657 // correctly with each other.
5658 OrderAccess::fence();
5659 }
5660 guarantee (_Event >= 0, "invariant") ;
5661 }
5662
5663 int os::PlatformEvent::park(jlong millis) {
5664 guarantee (_nParked == 0, "invariant") ;
5665
5666 int v ;
5667 for (;;) {
5668 v = _Event ;
5669 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5670 }
5671 guarantee (v >= 0, "invariant") ;
5672 if (v != 0) return OS_OK ;
5673
5674 // We do this the hard way, by blocking the thread.
5675 // Consider enforcing a minimum timeout value.
5676 struct timespec abst;
5677 compute_abstime(&abst, millis);
5678
5679 int ret = OS_TIMEOUT;
5680 int status = pthread_mutex_lock(_mutex);
5681 assert_status(status == 0, status, "mutex_lock");
5682 guarantee (_nParked == 0, "invariant") ;
5683 ++_nParked ;
5684
5685 // Object.wait(timo) will return because of
5686 // (a) notification
5687 // (b) timeout
5688 // (c) thread.interrupt
5689 //
5690 // Thread.interrupt and object.notify{All} both call Event::set.
5691 // That is, we treat thread.interrupt as a special case of notification.
5692 // The underlying Solaris implementation, cond_timedwait, admits
5693 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5694 // JVM from making those visible to Java code. As such, we must
5695 // filter out spurious wakeups. We assume all ETIME returns are valid.
5696 //
5697 // TODO: properly differentiate simultaneous notify+interrupt.
5698 // In that case, we should propagate the notify to another waiter.
5699
5700 while (_Event < 0) {
5701 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5702 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5703 pthread_cond_destroy (_cond);
5704 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5705 }
5706 assert_status(status == 0 || status == EINTR ||
5707 status == ETIME || status == ETIMEDOUT,
5708 status, "cond_timedwait");
5709 if (!FilterSpuriousWakeups) break ; // previous semantics
5710 if (status == ETIME || status == ETIMEDOUT) break ;
5711 // We consume and ignore EINTR and spurious wakeups.
5712 }
5713 --_nParked ;
5714 if (_Event >= 0) {
5715 ret = OS_OK;
5716 }
5717 _Event = 0 ;
5718 status = pthread_mutex_unlock(_mutex);
5719 assert_status(status == 0, status, "mutex_unlock");
5720 assert (_nParked == 0, "invariant") ;
5721 // Paranoia to ensure our locked and lock-free paths interact
5722 // correctly with each other.
5723 OrderAccess::fence();
5724 return ret;
5725 }
5726
5727 void os::PlatformEvent::unpark() {
5728 // Transitions for _Event:
5729 // 0 :=> 1
5730 // 1 :=> 1
5731 // -1 :=> either 0 or 1; must signal target thread
5732 // That is, we can safely transition _Event from -1 to either
5733 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5734 // unpark() calls.
5735 // See also: "Semaphores in Plan 9" by Mullender & Cox
5736 //
5737 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5738 // that it will take two back-to-back park() calls for the owning
5739 // thread to block. This has the benefit of forcing a spurious return
5740 // from the first park() call after an unpark() call which will help
5741 // shake out uses of park() and unpark() without condition variables.
5742
5743 if (Atomic::xchg(1, &_Event) >= 0) return;
5744
5745 // Wait for the thread associated with the event to vacate
5746 int status = pthread_mutex_lock(_mutex);
5747 assert_status(status == 0, status, "mutex_lock");
5748 int AnyWaiters = _nParked;
5749 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5750 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5751 AnyWaiters = 0;
5752 pthread_cond_signal(_cond);
5753 }
5754 status = pthread_mutex_unlock(_mutex);
5755 assert_status(status == 0, status, "mutex_unlock");
5756 if (AnyWaiters != 0) {
5757 status = pthread_cond_signal(_cond);
5758 assert_status(status == 0, status, "cond_signal");
5759 }
5760
5761 // Note that we signal() _after dropping the lock for "immortal" Events.
5762 // This is safe and avoids a common class of futile wakeups. In rare
5763 // circumstances this can cause a thread to return prematurely from
5764 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5765 // simply re-test the condition and re-park itself.
5766 }
5767
5768
5769 // JSR166
5770 // -------------------------------------------------------
5771
5772 /*
5773 * The solaris and linux implementations of park/unpark are fairly
5774 * conservative for now, but can be improved. They currently use a
5775 * mutex/condvar pair, plus a a count.
5776 * Park decrements count if > 0, else does a condvar wait. Unpark
5777 * sets count to 1 and signals condvar. Only one thread ever waits
5778 * on the condvar. Contention seen when trying to park implies that someone
5779 * is unparking you, so don't wait. And spurious returns are fine, so there
5780 * is no need to track notifications.
5781 */
5782
5783 /*
5784 * This code is common to linux and solaris and will be moved to a
5785 * common place in dolphin.
5786 *
5787 * The passed in time value is either a relative time in nanoseconds
5788 * or an absolute time in milliseconds. Either way it has to be unpacked
5789 * into suitable seconds and nanoseconds components and stored in the
5790 * given timespec structure.
5791 * Given time is a 64-bit value and the time_t used in the timespec is only
5792 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5793 * overflow if times way in the future are given. Further on Solaris versions
5794 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5795 * number of seconds, in abstime, is less than current_time + 100,000,000.
5796 * As it will be 28 years before "now + 100000000" will overflow we can
5797 * ignore overflow and just impose a hard-limit on seconds using the value
5798 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5799 * years from "now".
5800 */
5801
5802 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5803 assert (time > 0, "convertTime");
5804 time_t max_secs = 0;
5805
5806 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5807 struct timeval now;
5808 int status = gettimeofday(&now, NULL);
5809 assert(status == 0, "gettimeofday");
5810
5811 max_secs = now.tv_sec + MAX_SECS;
5812
5813 if (isAbsolute) {
5814 jlong secs = time / 1000;
5815 if (secs > max_secs) {
5816 absTime->tv_sec = max_secs;
5817 } else {
5818 absTime->tv_sec = secs;
5819 }
5820 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5821 } else {
5822 jlong secs = time / NANOSECS_PER_SEC;
5823 if (secs >= MAX_SECS) {
5824 absTime->tv_sec = max_secs;
5825 absTime->tv_nsec = 0;
5826 } else {
5827 absTime->tv_sec = now.tv_sec + secs;
5828 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5829 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5830 absTime->tv_nsec -= NANOSECS_PER_SEC;
5831 ++absTime->tv_sec; // note: this must be <= max_secs
5832 }
5833 }
5834 }
5835 } else {
5836 // must be relative using monotonic clock
5837 struct timespec now;
5838 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5839 assert_status(status == 0, status, "clock_gettime");
5840 max_secs = now.tv_sec + MAX_SECS;
5841 jlong secs = time / NANOSECS_PER_SEC;
5842 if (secs >= MAX_SECS) {
5843 absTime->tv_sec = max_secs;
5844 absTime->tv_nsec = 0;
5845 } else {
5846 absTime->tv_sec = now.tv_sec + secs;
5847 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5848 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5849 absTime->tv_nsec -= NANOSECS_PER_SEC;
5850 ++absTime->tv_sec; // note: this must be <= max_secs
5851 }
5852 }
5853 }
5854 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5855 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5856 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5857 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5858 }
5859
5860 void Parker::park(bool isAbsolute, jlong time) {
5861 // Ideally we'd do something useful while spinning, such
5862 // as calling unpackTime().
5863
5864 // Optional fast-path check:
5865 // Return immediately if a permit is available.
5866 // We depend on Atomic::xchg() having full barrier semantics
5867 // since we are doing a lock-free update to _counter.
5868 if (Atomic::xchg(0, &_counter) > 0) return;
5869
5870 Thread* thread = Thread::current();
5871 assert(thread->is_Java_thread(), "Must be JavaThread");
5872 JavaThread *jt = (JavaThread *)thread;
5873
5874 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5875 // Check interrupt before trying to wait
5876 if (Thread::is_interrupted(thread, false)) {
5877 return;
5878 }
5879
5880 // Next, demultiplex/decode time arguments
5881 timespec absTime;
5882 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5883 return;
5884 }
5885 if (time > 0) {
5886 unpackTime(&absTime, isAbsolute, time);
5887 }
5888
5889
5890 // Enter safepoint region
5891 // Beware of deadlocks such as 6317397.
5892 // The per-thread Parker:: mutex is a classic leaf-lock.
5893 // In particular a thread must never block on the Threads_lock while
5894 // holding the Parker:: mutex. If safepoints are pending both the
5895 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5896 ThreadBlockInVM tbivm(jt);
5897
5898 // Don't wait if cannot get lock since interference arises from
5899 // unblocking. Also. check interrupt before trying wait
5900 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5901 return;
5902 }
5903
5904 int status ;
5905 if (_counter > 0) { // no wait needed
5906 _counter = 0;
5907 status = pthread_mutex_unlock(_mutex);
5908 assert (status == 0, "invariant") ;
5909 // Paranoia to ensure our locked and lock-free paths interact
5910 // correctly with each other and Java-level accesses.
5911 OrderAccess::fence();
5912 return;
5913 }
5914
5915 #ifdef ASSERT
5916 // Don't catch signals while blocked; let the running threads have the signals.
5917 // (This allows a debugger to break into the running thread.)
5918 sigset_t oldsigs;
5919 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5920 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5921 #endif
5922
5923 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5924 jt->set_suspend_equivalent();
5925 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5926
5927 assert(_cur_index == -1, "invariant");
5928 if (time == 0) {
5929 _cur_index = REL_INDEX; // arbitrary choice when not timed
5930 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5931 } else {
5932 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5933 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5934 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5935 pthread_cond_destroy (&_cond[_cur_index]) ;
5936 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5937 }
5938 }
5939 _cur_index = -1;
5940 assert_status(status == 0 || status == EINTR ||
5941 status == ETIME || status == ETIMEDOUT,
5942 status, "cond_timedwait");
5943
5944 #ifdef ASSERT
5945 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5946 #endif
5947
5948 _counter = 0 ;
5949 status = pthread_mutex_unlock(_mutex) ;
5950 assert_status(status == 0, status, "invariant") ;
5951 // Paranoia to ensure our locked and lock-free paths interact
5952 // correctly with each other and Java-level accesses.
5953 OrderAccess::fence();
5954
5955 // If externally suspended while waiting, re-suspend
5956 if (jt->handle_special_suspend_equivalent_condition()) {
5957 jt->java_suspend_self();
5958 }
5959 }
5960
5961 void Parker::unpark() {
5962 int s, status ;
5963 status = pthread_mutex_lock(_mutex);
5964 assert (status == 0, "invariant") ;
5965 s = _counter;
5966 _counter = 1;
5967 if (s < 1) {
5968 // thread might be parked
5969 if (_cur_index != -1) {
5970 // thread is definitely parked
5971 if (WorkAroundNPTLTimedWaitHang) {
5972 status = pthread_cond_signal (&_cond[_cur_index]);
5973 assert (status == 0, "invariant");
5974 status = pthread_mutex_unlock(_mutex);
5975 assert (status == 0, "invariant");
5976 } else {
5977 status = pthread_mutex_unlock(_mutex);
5978 assert (status == 0, "invariant");
5979 status = pthread_cond_signal (&_cond[_cur_index]);
5980 assert (status == 0, "invariant");
5981 }
5982 } else {
5983 pthread_mutex_unlock(_mutex);
5984 assert (status == 0, "invariant") ;
5985 }
5986 } else {
5987 pthread_mutex_unlock(_mutex);
5988 assert (status == 0, "invariant") ;
5989 }
5990 }
5991
5992
5993 extern char** environ;
5994
5995 #ifndef __NR_fork
5996 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5997 #endif
5998
5999 #ifndef __NR_execve
6000 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
6001 #endif
6002
6003 // Run the specified command in a separate process. Return its exit value,
6004 // or -1 on failure (e.g. can't fork a new process).
6005 // Unlike system(), this function can be called from signal handler. It
6006 // doesn't block SIGINT et al.
6007 int os::fork_and_exec(char* cmd) {
6008 const char * argv[4] = {"sh", "-c", cmd, NULL};
6009
6010 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
6011 // pthread_atfork handlers and reset pthread library. All we need is a
6012 // separate process to execve. Make a direct syscall to fork process.
6013 // On IA64 there's no fork syscall, we have to use fork() and hope for
6014 // the best...
6015 pid_t pid = NOT_IA64(syscall(__NR_fork);)
6016 IA64_ONLY(fork();)
6017
6018 if (pid < 0) {
6019 // fork failed
6020 return -1;
6021
6022 } else if (pid == 0) {
6023 // child process
6024
6025 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6026 // first to kill every thread on the thread list. Because this list is
6027 // not reset by fork() (see notes above), execve() will instead kill
6028 // every thread in the parent process. We know this is the only thread
6029 // in the new process, so make a system call directly.
6030 // IA64 should use normal execve() from glibc to match the glibc fork()
6031 // above.
6032 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6033 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6034
6035 // execve failed
6036 _exit(-1);
6037
6038 } else {
6039 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6040 // care about the actual exit code, for now.
6041
6042 int status;
6043
6044 // Wait for the child process to exit. This returns immediately if
6045 // the child has already exited. */
6046 while (waitpid(pid, &status, 0) < 0) {
6047 switch (errno) {
6048 case ECHILD: return 0;
6049 case EINTR: break;
6050 default: return -1;
6051 }
6052 }
6053
6054 if (WIFEXITED(status)) {
6055 // The child exited normally; get its exit code.
6056 return WEXITSTATUS(status);
6057 } else if (WIFSIGNALED(status)) {
6058 // The child exited because of a signal
6059 // The best value to return is 0x80 + signal number,
6060 // because that is what all Unix shells do, and because
6061 // it allows callers to distinguish between process exit and
6062 // process death by signal.
6063 return 0x80 + WTERMSIG(status);
6064 } else {
6065 // Unknown exit code; pass it through
6066 return status;
6067 }
6068 }
6069 }
6070
6071 // is_headless_jre()
6072 //
6073 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6074 // in order to report if we are running in a headless jre
6075 //
6076 // Since JDK8 xawt/libmawt.so was moved into the same directory
6077 // as libawt.so, and renamed libawt_xawt.so
6078 //
6079 bool os::is_headless_jre() {
6080 struct stat statbuf;
6081 char buf[MAXPATHLEN];
6082 char libmawtpath[MAXPATHLEN];
6083 const char *xawtstr = "/xawt/libmawt.so";
6084 const char *new_xawtstr = "/libawt_xawt.so";
6085 char *p;
6086
6087 // Get path to libjvm.so
6088 os::jvm_path(buf, sizeof(buf));
6089
6090 // Get rid of libjvm.so
6091 p = strrchr(buf, '/');
6092 if (p == NULL) return false;
6093 else *p = '\0';
6094
6095 // Get rid of client or server
6096 p = strrchr(buf, '/');
6097 if (p == NULL) return false;
6098 else *p = '\0';
6099
6100 // check xawt/libmawt.so
6101 strcpy(libmawtpath, buf);
6102 strcat(libmawtpath, xawtstr);
6103 if (::stat(libmawtpath, &statbuf) == 0) return false;
6104
6105 // check libawt_xawt.so
6106 strcpy(libmawtpath, buf);
6107 strcat(libmawtpath, new_xawtstr);
6108 if (::stat(libmawtpath, &statbuf) == 0) return false;
6109
6110 return true;
6111 }
6112
6113 // Get the default path to the core file
6114 // Returns the length of the string
6115 int os::get_core_path(char* buffer, size_t bufferSize) {
6116 const char* p = get_current_directory(buffer, bufferSize);
6117
6118 if (p == NULL) {
6119 assert(p != NULL, "failed to get current directory");
6120 return 0;
6121 }
6122
6123 return strlen(buffer);
6124 }
6125
6126 #ifdef JAVASE_EMBEDDED
6127 //
6128 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6129 //
6130 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6131
6132 // ctor
6133 //
6134 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6135 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6136 _fd = fd;
6137
6138 if (os::create_thread(this, os::os_thread)) {
6139 _memnotify_thread = this;
6140 os::set_priority(this, NearMaxPriority);
6141 os::start_thread(this);
6142 }
6143 }
6144
6145 // Where all the work gets done
6146 //
6147 void MemNotifyThread::run() {
6148 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6149
6150 // Set up the select arguments
6151 fd_set rfds;
6152 if (_fd != -1) {
6153 FD_ZERO(&rfds);
6154 FD_SET(_fd, &rfds);
6155 }
6156
6157 // Now wait for the mem_notify device to wake up
6158 while (1) {
6159 // Wait for the mem_notify device to signal us..
6160 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6161 if (rc == -1) {
6162 perror("select!\n");
6163 break;
6164 } else if (rc) {
6165 //ssize_t free_before = os::available_memory();
6166 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6167
6168 // The kernel is telling us there is not much memory left...
6169 // try to do something about that
6170
6171 // If we are not already in a GC, try one.
6172 if (!Universe::heap()->is_gc_active()) {
6173 Universe::heap()->collect(GCCause::_allocation_failure);
6174
6175 //ssize_t free_after = os::available_memory();
6176 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6177 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6178 }
6179 // We might want to do something like the following if we find the GC's are not helping...
6180 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6181 }
6182 }
6183 }
6184
6185 //
6186 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6187 //
6188 void MemNotifyThread::start() {
6189 int fd;
6190 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6191 if (fd < 0) {
6192 return;
6193 }
6194
6195 if (memnotify_thread() == NULL) {
6196 new MemNotifyThread(fd);
6197 }
6198 }
6199
6200 #endif // JAVASE_EMBEDDED
6201
6202
6203 /////////////// Unit tests ///////////////
6204
6205 #ifndef PRODUCT
6206
6207 #define test_log(...) \
6208 do {\
6209 if (VerboseInternalVMTests) { \
6210 tty->print_cr(__VA_ARGS__); \
6211 tty->flush(); \
6212 }\
6213 } while (false)
6214
6215 class TestReserveMemorySpecial : AllStatic {
6216 public:
6217 static void small_page_write(void* addr, size_t size) {
6218 size_t page_size = os::vm_page_size();
6219
6220 char* end = (char*)addr + size;
6221 for (char* p = (char*)addr; p < end; p += page_size) {
6222 *p = 1;
6223 }
6224 }
6225
6226 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6227 if (!UseHugeTLBFS) {
6228 return;
6229 }
6230
6231 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6232
6233 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6234
6235 if (addr != NULL) {
6236 small_page_write(addr, size);
6237
6238 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6239 }
6240 }
6241
6242 static void test_reserve_memory_special_huge_tlbfs_only() {
6243 if (!UseHugeTLBFS) {
6244 return;
6245 }
6246
6247 size_t lp = os::large_page_size();
6248
6249 for (size_t size = lp; size <= lp * 10; size += lp) {
6250 test_reserve_memory_special_huge_tlbfs_only(size);
6251 }
6252 }
6253
6254 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6255 if (!UseHugeTLBFS) {
6256 return;
6257 }
6258
6259 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6260 size, alignment);
6261
6262 assert(size >= os::large_page_size(), "Incorrect input to test");
6263
6264 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6265
6266 if (addr != NULL) {
6267 small_page_write(addr, size);
6268
6269 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6270 }
6271 }
6272
6273 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6274 size_t lp = os::large_page_size();
6275 size_t ag = os::vm_allocation_granularity();
6276
6277 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6278 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6279 }
6280 }
6281
6282 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6283 size_t lp = os::large_page_size();
6284 size_t ag = os::vm_allocation_granularity();
6285
6286 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6287 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6288 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6289 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6290 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6291 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6292 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6293 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6294 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6295 }
6296
6297 static void test_reserve_memory_special_huge_tlbfs() {
6298 if (!UseHugeTLBFS) {
6299 return;
6300 }
6301
6302 test_reserve_memory_special_huge_tlbfs_only();
6303 test_reserve_memory_special_huge_tlbfs_mixed();
6304 }
6305
6306 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6307 if (!UseSHM) {
6308 return;
6309 }
6310
6311 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6312
6313 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6314
6315 if (addr != NULL) {
6316 assert(is_ptr_aligned(addr, alignment), "Check");
6317 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6318
6319 small_page_write(addr, size);
6320
6321 os::Linux::release_memory_special_shm(addr, size);
6322 }
6323 }
6324
6325 static void test_reserve_memory_special_shm() {
6326 size_t lp = os::large_page_size();
6327 size_t ag = os::vm_allocation_granularity();
6328
6329 for (size_t size = ag; size < lp * 3; size += ag) {
6330 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6331 test_reserve_memory_special_shm(size, alignment);
6332 }
6333 }
6334 }
6335
6336 static void test() {
6337 test_reserve_memory_special_huge_tlbfs();
6338 test_reserve_memory_special_shm();
6339 }
6340 };
6341
6342 void TestReserveMemorySpecial_test() {
6343 TestReserveMemorySpecial::test();
6344 }
6345
6346 #endif