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/atomic.inline.hpp"
45 #include "runtime/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/interfaceSupport.hpp"
48 #include "runtime/init.hpp"
49 #include "runtime/java.hpp"
50 #include "runtime/javaCalls.hpp"
51 #include "runtime/mutexLocker.hpp"
52 #include "runtime/objectMonitor.hpp"
53 #include "runtime/orderAccess.inline.hpp"
54 #include "runtime/osThread.hpp"
55 #include "runtime/perfMemory.hpp"
56 #include "runtime/sharedRuntime.hpp"
57 #include "runtime/statSampler.hpp"
58 #include "runtime/stubRoutines.hpp"
59 #include "runtime/thread.inline.hpp"
60 #include "runtime/threadCritical.hpp"
61 #include "runtime/timer.hpp"
62 #include "services/attachListener.hpp"
63 #include "services/memTracker.hpp"
64 #include "services/runtimeService.hpp"
65 #include "utilities/decoder.hpp"
66 #include "utilities/defaultStream.hpp"
67 #include "utilities/events.hpp"
68 #include "utilities/elfFile.hpp"
69 #include "utilities/growableArray.hpp"
70 #include "utilities/vmError.hpp"
71
72 // put OS-includes here
73 # include <sys/types.h>
74 # include <sys/mman.h>
75 # include <sys/stat.h>
76 # include <sys/select.h>
77 # include <pthread.h>
78 # include <signal.h>
79 # include <errno.h>
80 # include <dlfcn.h>
81 # include <stdio.h>
82 # include <unistd.h>
83 # include <sys/resource.h>
84 # include <pthread.h>
85 # include <sys/stat.h>
86 # include <sys/time.h>
87 # include <sys/times.h>
88 # include <sys/utsname.h>
89 # include <sys/socket.h>
90 # include <sys/wait.h>
91 # include <pwd.h>
92 # include <poll.h>
93 # include <semaphore.h>
94 # include <fcntl.h>
95 # include <string.h>
96 # include <syscall.h>
97 # include <sys/sysinfo.h>
98 # include <gnu/libc-version.h>
99 # include <sys/ipc.h>
100 # include <sys/shm.h>
101 # include <link.h>
102 # include <stdint.h>
103 # include <inttypes.h>
104 # include <sys/ioctl.h>
105
106 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
107
108 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
109 // getrusage() is prepared to handle the associated failure.
110 #ifndef RUSAGE_THREAD
111 #define RUSAGE_THREAD (1) /* only the calling thread */
112 #endif
113
114 #define MAX_PATH (2 * K)
115
116 #define MAX_SECS 100000000
117
118 // for timer info max values which include all bits
119 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
120
121 #define LARGEPAGES_BIT (1 << 6)
122 ////////////////////////////////////////////////////////////////////////////////
123 // global variables
124 julong os::Linux::_physical_memory = 0;
125
126 address os::Linux::_initial_thread_stack_bottom = NULL;
127 uintptr_t os::Linux::_initial_thread_stack_size = 0;
128
129 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
130 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
131 Mutex* os::Linux::_createThread_lock = NULL;
132 pthread_t os::Linux::_main_thread;
133 int os::Linux::_page_size = -1;
134 const int os::Linux::_vm_default_page_size = (8 * K);
135 bool os::Linux::_is_floating_stack = false;
136 bool os::Linux::_is_NPTL = false;
137 bool os::Linux::_supports_fast_thread_cpu_time = false;
138 const char * os::Linux::_glibc_version = NULL;
139 const char * os::Linux::_libpthread_version = NULL;
140 pthread_condattr_t os::Linux::_condattr[1];
141
142 static jlong initial_time_count=0;
143
144 static int clock_tics_per_sec = 100;
145
146 // For diagnostics to print a message once. see run_periodic_checks
147 static sigset_t check_signal_done;
148 static bool check_signals = true;
149
150 static pid_t _initial_pid = 0;
151
152 /* Signal number used to suspend/resume a thread */
153
154 /* do not use any signal number less than SIGSEGV, see 4355769 */
155 static int SR_signum = SIGUSR2;
156 sigset_t SR_sigset;
157
158 /* Used to protect dlsym() calls */
159 static pthread_mutex_t dl_mutex;
160
161 // Declarations
162 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
163
164 #ifdef JAVASE_EMBEDDED
165 class MemNotifyThread: public Thread {
166 friend class VMStructs;
167 public:
168 virtual void run();
169
170 private:
171 static MemNotifyThread* _memnotify_thread;
172 int _fd;
173
174 public:
175
176 // Constructor
177 MemNotifyThread(int fd);
178
179 // Tester
180 bool is_memnotify_thread() const { return true; }
181
182 // Printing
183 char* name() const { return (char*)"Linux MemNotify Thread"; }
184
185 // Returns the single instance of the MemNotifyThread
186 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
187
188 // Create and start the single instance of MemNotifyThread
189 static void start();
190 };
191 #endif // JAVASE_EMBEDDED
192
193 // utility functions
194
195 static int SR_initialize();
196
197 julong os::available_memory() {
198 return Linux::available_memory();
199 }
200
201 julong os::Linux::available_memory() {
202 // values in struct sysinfo are "unsigned long"
203 struct sysinfo si;
204 sysinfo(&si);
205
206 return (julong)si.freeram * si.mem_unit;
207 }
208
209 julong os::physical_memory() {
210 return Linux::physical_memory();
211 }
212
213 ////////////////////////////////////////////////////////////////////////////////
214 // environment support
215
216 bool os::getenv(const char* name, char* buf, int len) {
217 const char* val = ::getenv(name);
218 if (val != NULL && strlen(val) < (size_t)len) {
219 strcpy(buf, val);
220 return true;
221 }
222 if (len > 0) buf[0] = 0; // return a null string
223 return false;
224 }
225
226
227 // Return true if user is running as root.
228
229 bool os::have_special_privileges() {
230 static bool init = false;
231 static bool privileges = false;
232 if (!init) {
233 privileges = (getuid() != geteuid()) || (getgid() != getegid());
234 init = true;
235 }
236 return privileges;
237 }
238
239
240 #ifndef SYS_gettid
241 // i386: 224, ia64: 1105, amd64: 186, sparc 143
242 #ifdef __ia64__
243 #define SYS_gettid 1105
244 #elif __i386__
245 #define SYS_gettid 224
246 #elif __amd64__
247 #define SYS_gettid 186
248 #elif __sparc__
249 #define SYS_gettid 143
250 #else
251 #error define gettid for the arch
252 #endif
253 #endif
254
255 // Cpu architecture string
256 #if defined(ZERO)
257 static char cpu_arch[] = ZERO_LIBARCH;
258 #elif defined(IA64)
259 static char cpu_arch[] = "ia64";
260 #elif defined(IA32)
261 static char cpu_arch[] = "i386";
262 #elif defined(AMD64)
263 static char cpu_arch[] = "amd64";
264 #elif defined(ARM)
265 static char cpu_arch[] = "arm";
266 #elif defined(PPC32)
267 static char cpu_arch[] = "ppc";
268 #elif defined(PPC64)
269 static char cpu_arch[] = "ppc64";
270 #elif defined(SPARC)
271 # ifdef _LP64
272 static char cpu_arch[] = "sparcv9";
273 # else
274 static char cpu_arch[] = "sparc";
275 # endif
276 #else
277 #error Add appropriate cpu_arch setting
278 #endif
279
280
281 // pid_t gettid()
282 //
283 // Returns the kernel thread id of the currently running thread. Kernel
284 // thread id is used to access /proc.
285 //
286 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
287 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
288 //
289 pid_t os::Linux::gettid() {
290 int rslt = syscall(SYS_gettid);
291 if (rslt == -1) {
292 // old kernel, no NPTL support
293 return getpid();
294 } else {
295 return (pid_t)rslt;
296 }
297 }
298
299 // Most versions of linux have a bug where the number of processors are
300 // determined by looking at the /proc file system. In a chroot environment,
301 // the system call returns 1. This causes the VM to act as if it is
302 // a single processor and elide locking (see is_MP() call).
303 static bool unsafe_chroot_detected = false;
304 static const char *unstable_chroot_error = "/proc file system not found.\n"
305 "Java may be unstable running multithreaded in a chroot "
306 "environment on Linux when /proc filesystem is not mounted.";
307
308 void os::Linux::initialize_system_info() {
309 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
310 if (processor_count() == 1) {
311 pid_t pid = os::Linux::gettid();
312 char fname[32];
313 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
314 FILE *fp = fopen(fname, "r");
315 if (fp == NULL) {
316 unsafe_chroot_detected = true;
317 } else {
318 fclose(fp);
319 }
320 }
321 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
322 assert(processor_count() > 0, "linux error");
323 }
324
325 void os::init_system_properties_values() {
326 // The next steps are taken in the product version:
327 //
328 // Obtain the JAVA_HOME value from the location of libjvm.so.
329 // This library should be located at:
330 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
331 //
332 // If "/jre/lib/" appears at the right place in the path, then we
333 // assume libjvm.so is installed in a JDK and we use this path.
334 //
335 // Otherwise exit with message: "Could not create the Java virtual machine."
336 //
337 // The following extra steps are taken in the debugging version:
338 //
339 // If "/jre/lib/" does NOT appear at the right place in the path
340 // instead of exit check for $JAVA_HOME environment variable.
341 //
342 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
343 // then we append a fake suffix "hotspot/libjvm.so" to this path so
344 // it looks like libjvm.so is installed there
345 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
346 //
347 // Otherwise exit.
348 //
349 // Important note: if the location of libjvm.so changes this
350 // code needs to be changed accordingly.
351
352 // See ld(1):
353 // The linker uses the following search paths to locate required
354 // shared libraries:
355 // 1: ...
356 // ...
357 // 7: The default directories, normally /lib and /usr/lib.
358 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
359 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
360 #else
361 #define DEFAULT_LIBPATH "/lib:/usr/lib"
362 #endif
363
364 // Base path of extensions installed on the system.
365 #define SYS_EXT_DIR "/usr/java/packages"
366 #define EXTENSIONS_DIR "/lib/ext"
367 #define ENDORSED_DIR "/lib/endorsed"
368
369 // Buffer that fits several sprintfs.
370 // Note that the space for the colon and the trailing null are provided
371 // by the nulls included by the sizeof operator.
372 const size_t bufsize =
373 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
374 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
375 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
376 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
377
378 // sysclasspath, java_home, dll_dir
379 {
380 char *pslash;
381 os::jvm_path(buf, bufsize);
382
383 // Found the full path to libjvm.so.
384 // Now cut the path to <java_home>/jre if we can.
385 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
386 pslash = strrchr(buf, '/');
387 if (pslash != NULL) {
388 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
389 }
390 Arguments::set_dll_dir(buf);
391
392 if (pslash != NULL) {
393 pslash = strrchr(buf, '/');
394 if (pslash != NULL) {
395 *pslash = '\0'; // Get rid of /<arch>.
396 pslash = strrchr(buf, '/');
397 if (pslash != NULL) {
398 *pslash = '\0'; // Get rid of /lib.
399 }
400 }
401 }
402 Arguments::set_java_home(buf);
403 set_boot_path('/', ':');
404 }
405
406 // Where to look for native libraries.
407 //
408 // Note: Due to a legacy implementation, most of the library path
409 // is set in the launcher. This was to accomodate linking restrictions
410 // on legacy Linux implementations (which are no longer supported).
411 // Eventually, all the library path setting will be done here.
412 //
413 // However, to prevent the proliferation of improperly built native
414 // libraries, the new path component /usr/java/packages is added here.
415 // Eventually, all the library path setting will be done here.
416 {
417 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
418 // should always exist (until the legacy problem cited above is
419 // addressed).
420 const char *v = ::getenv("LD_LIBRARY_PATH");
421 const char *v_colon = ":";
422 if (v == NULL) { v = ""; v_colon = ""; }
423 // That's +1 for the colon and +1 for the trailing '\0'.
424 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
425 strlen(v) + 1 +
426 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
427 mtInternal);
428 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
429 Arguments::set_library_path(ld_library_path);
430 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
431 }
432
433 // Extensions directories.
434 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
435 Arguments::set_ext_dirs(buf);
436
437 // Endorsed standards default directory.
438 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
439 Arguments::set_endorsed_dirs(buf);
440
441 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
442
443 #undef DEFAULT_LIBPATH
444 #undef SYS_EXT_DIR
445 #undef EXTENSIONS_DIR
446 #undef ENDORSED_DIR
447 }
448
449 ////////////////////////////////////////////////////////////////////////////////
450 // breakpoint support
451
452 void os::breakpoint() {
453 BREAKPOINT;
454 }
455
456 extern "C" void breakpoint() {
457 // use debugger to set breakpoint here
458 }
459
460 ////////////////////////////////////////////////////////////////////////////////
461 // signal support
462
463 debug_only(static bool signal_sets_initialized = false);
464 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
465
466 bool os::Linux::is_sig_ignored(int sig) {
467 struct sigaction oact;
468 sigaction(sig, (struct sigaction*)NULL, &oact);
469 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
470 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
471 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
472 return true;
473 else
474 return false;
475 }
476
477 void os::Linux::signal_sets_init() {
478 // Should also have an assertion stating we are still single-threaded.
479 assert(!signal_sets_initialized, "Already initialized");
480 // Fill in signals that are necessarily unblocked for all threads in
481 // the VM. Currently, we unblock the following signals:
482 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
483 // by -Xrs (=ReduceSignalUsage));
484 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
485 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
486 // the dispositions or masks wrt these signals.
487 // Programs embedding the VM that want to use the above signals for their
488 // own purposes must, at this time, use the "-Xrs" option to prevent
489 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
490 // (See bug 4345157, and other related bugs).
491 // In reality, though, unblocking these signals is really a nop, since
492 // these signals are not blocked by default.
493 sigemptyset(&unblocked_sigs);
494 sigemptyset(&allowdebug_blocked_sigs);
495 sigaddset(&unblocked_sigs, SIGILL);
496 sigaddset(&unblocked_sigs, SIGSEGV);
497 sigaddset(&unblocked_sigs, SIGBUS);
498 sigaddset(&unblocked_sigs, SIGFPE);
499 #if defined(PPC64)
500 sigaddset(&unblocked_sigs, SIGTRAP);
501 #endif
502 sigaddset(&unblocked_sigs, SR_signum);
503
504 if (!ReduceSignalUsage) {
505 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
506 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
507 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
508 }
509 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
510 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
511 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
512 }
513 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
514 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
515 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
516 }
517 }
518 // Fill in signals that are blocked by all but the VM thread.
519 sigemptyset(&vm_sigs);
520 if (!ReduceSignalUsage)
521 sigaddset(&vm_sigs, BREAK_SIGNAL);
522 debug_only(signal_sets_initialized = true);
523
524 }
525
526 // These are signals that are unblocked while a thread is running Java.
527 // (For some reason, they get blocked by default.)
528 sigset_t* os::Linux::unblocked_signals() {
529 assert(signal_sets_initialized, "Not initialized");
530 return &unblocked_sigs;
531 }
532
533 // These are the signals that are blocked while a (non-VM) thread is
534 // running Java. Only the VM thread handles these signals.
535 sigset_t* os::Linux::vm_signals() {
536 assert(signal_sets_initialized, "Not initialized");
537 return &vm_sigs;
538 }
539
540 // These are signals that are blocked during cond_wait to allow debugger in
541 sigset_t* os::Linux::allowdebug_blocked_signals() {
542 assert(signal_sets_initialized, "Not initialized");
543 return &allowdebug_blocked_sigs;
544 }
545
546 void os::Linux::hotspot_sigmask(Thread* thread) {
547
548 //Save caller's signal mask before setting VM signal mask
549 sigset_t caller_sigmask;
550 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
551
552 OSThread* osthread = thread->osthread();
553 osthread->set_caller_sigmask(caller_sigmask);
554
555 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
556
557 if (!ReduceSignalUsage) {
558 if (thread->is_VM_thread()) {
559 // Only the VM thread handles BREAK_SIGNAL ...
560 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
561 } else {
562 // ... all other threads block BREAK_SIGNAL
563 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
564 }
565 }
566 }
567
568 //////////////////////////////////////////////////////////////////////////////
569 // detecting pthread library
570
571 void os::Linux::libpthread_init() {
572 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
573 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
574 // generic name for earlier versions.
575 // Define macros here so we can build HotSpot on old systems.
576 # ifndef _CS_GNU_LIBC_VERSION
577 # define _CS_GNU_LIBC_VERSION 2
578 # endif
579 # ifndef _CS_GNU_LIBPTHREAD_VERSION
580 # define _CS_GNU_LIBPTHREAD_VERSION 3
581 # endif
582
583 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
584 if (n > 0) {
585 char *str = (char *)malloc(n, mtInternal);
586 confstr(_CS_GNU_LIBC_VERSION, str, n);
587 os::Linux::set_glibc_version(str);
588 } else {
589 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
590 static char _gnu_libc_version[32];
591 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
592 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
593 os::Linux::set_glibc_version(_gnu_libc_version);
594 }
595
596 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
597 if (n > 0) {
598 char *str = (char *)malloc(n, mtInternal);
599 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
600 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
601 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
602 // is the case. LinuxThreads has a hard limit on max number of threads.
603 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
604 // On the other hand, NPTL does not have such a limit, sysconf()
605 // will return -1 and errno is not changed. Check if it is really NPTL.
606 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
607 strstr(str, "NPTL") &&
608 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
609 free(str);
610 os::Linux::set_libpthread_version("linuxthreads");
611 } else {
612 os::Linux::set_libpthread_version(str);
613 }
614 } else {
615 // glibc before 2.3.2 only has LinuxThreads.
616 os::Linux::set_libpthread_version("linuxthreads");
617 }
618
619 if (strstr(libpthread_version(), "NPTL")) {
620 os::Linux::set_is_NPTL();
621 } else {
622 os::Linux::set_is_LinuxThreads();
623 }
624
625 // LinuxThreads have two flavors: floating-stack mode, which allows variable
626 // stack size; and fixed-stack mode. NPTL is always floating-stack.
627 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
628 os::Linux::set_is_floating_stack();
629 }
630 }
631
632 /////////////////////////////////////////////////////////////////////////////
633 // thread stack
634
635 // Force Linux kernel to expand current thread stack. If "bottom" is close
636 // to the stack guard, caller should block all signals.
637 //
638 // MAP_GROWSDOWN:
639 // A special mmap() flag that is used to implement thread stacks. It tells
640 // kernel that the memory region should extend downwards when needed. This
641 // allows early versions of LinuxThreads to only mmap the first few pages
642 // when creating a new thread. Linux kernel will automatically expand thread
643 // stack as needed (on page faults).
644 //
645 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
646 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
647 // region, it's hard to tell if the fault is due to a legitimate stack
648 // access or because of reading/writing non-exist memory (e.g. buffer
649 // overrun). As a rule, if the fault happens below current stack pointer,
650 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
651 // application (see Linux kernel fault.c).
652 //
653 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
654 // stack overflow detection.
655 //
656 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
657 // not use this flag. However, the stack of initial thread is not created
658 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
659 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
660 // and then attach the thread to JVM.
661 //
662 // To get around the problem and allow stack banging on Linux, we need to
663 // manually expand thread stack after receiving the SIGSEGV.
664 //
665 // There are two ways to expand thread stack to address "bottom", we used
666 // both of them in JVM before 1.5:
667 // 1. adjust stack pointer first so that it is below "bottom", and then
668 // touch "bottom"
669 // 2. mmap() the page in question
670 //
671 // Now alternate signal stack is gone, it's harder to use 2. For instance,
672 // if current sp is already near the lower end of page 101, and we need to
673 // call mmap() to map page 100, it is possible that part of the mmap() frame
674 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
675 // That will destroy the mmap() frame and cause VM to crash.
676 //
677 // The following code works by adjusting sp first, then accessing the "bottom"
678 // page to force a page fault. Linux kernel will then automatically expand the
679 // stack mapping.
680 //
681 // _expand_stack_to() assumes its frame size is less than page size, which
682 // should always be true if the function is not inlined.
683
684 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
685 #define NOINLINE
686 #else
687 #define NOINLINE __attribute__ ((noinline))
688 #endif
689
690 static void _expand_stack_to(address bottom) NOINLINE;
691
692 static void _expand_stack_to(address bottom) {
693 address sp;
694 size_t size;
695 volatile char *p;
696
697 // Adjust bottom to point to the largest address within the same page, it
698 // gives us a one-page buffer if alloca() allocates slightly more memory.
699 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
700 bottom += os::Linux::page_size() - 1;
701
702 // sp might be slightly above current stack pointer; if that's the case, we
703 // will alloca() a little more space than necessary, which is OK. Don't use
704 // os::current_stack_pointer(), as its result can be slightly below current
705 // stack pointer, causing us to not alloca enough to reach "bottom".
706 sp = (address)&sp;
707
708 if (sp > bottom) {
709 size = sp - bottom;
710 p = (volatile char *)alloca(size);
711 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
712 p[0] = '\0';
713 }
714 }
715
716 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
717 assert(t!=NULL, "just checking");
718 assert(t->osthread()->expanding_stack(), "expand should be set");
719 assert(t->stack_base() != NULL, "stack_base was not initialized");
720
721 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
722 sigset_t mask_all, old_sigset;
723 sigfillset(&mask_all);
724 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
725 _expand_stack_to(addr);
726 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
727 return true;
728 }
729 return false;
730 }
731
732 //////////////////////////////////////////////////////////////////////////////
733 // create new thread
734
735 static address highest_vm_reserved_address();
736
737 // check if it's safe to start a new thread
738 static bool _thread_safety_check(Thread* thread) {
739 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
740 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
741 // Heap is mmap'ed at lower end of memory space. Thread stacks are
742 // allocated (MAP_FIXED) from high address space. Every thread stack
743 // occupies a fixed size slot (usually 2Mbytes, but user can change
744 // it to other values if they rebuild LinuxThreads).
745 //
746 // Problem with MAP_FIXED is that mmap() can still succeed even part of
747 // the memory region has already been mmap'ed. That means if we have too
748 // many threads and/or very large heap, eventually thread stack will
749 // collide with heap.
750 //
751 // Here we try to prevent heap/stack collision by comparing current
752 // stack bottom with the highest address that has been mmap'ed by JVM
753 // plus a safety margin for memory maps created by native code.
754 //
755 // This feature can be disabled by setting ThreadSafetyMargin to 0
756 //
757 if (ThreadSafetyMargin > 0) {
758 address stack_bottom = os::current_stack_base() - os::current_stack_size();
759
760 // not safe if our stack extends below the safety margin
761 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
762 } else {
763 return true;
764 }
765 } else {
766 // Floating stack LinuxThreads or NPTL:
767 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
768 // there's not enough space left, pthread_create() will fail. If we come
769 // here, that means enough space has been reserved for stack.
770 return true;
771 }
772 }
773
774 // Thread start routine for all newly created threads
775 static void *java_start(Thread *thread) {
776 // Try to randomize the cache line index of hot stack frames.
777 // This helps when threads of the same stack traces evict each other's
778 // cache lines. The threads can be either from the same JVM instance, or
779 // from different JVM instances. The benefit is especially true for
780 // processors with hyperthreading technology.
781 static int counter = 0;
782 int pid = os::current_process_id();
783 alloca(((pid ^ counter++) & 7) * 128);
784
785 ThreadLocalStorage::set_thread(thread);
786
787 OSThread* osthread = thread->osthread();
788 Monitor* sync = osthread->startThread_lock();
789
790 // non floating stack LinuxThreads needs extra check, see above
791 if (!_thread_safety_check(thread)) {
792 // notify parent thread
793 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
794 osthread->set_state(ZOMBIE);
795 sync->notify_all();
796 return NULL;
797 }
798
799 // thread_id is kernel thread id (similar to Solaris LWP id)
800 osthread->set_thread_id(os::Linux::gettid());
801
802 if (UseNUMA) {
803 int lgrp_id = os::numa_get_group_id();
804 if (lgrp_id != -1) {
805 thread->set_lgrp_id(lgrp_id);
806 }
807 }
808 // initialize signal mask for this thread
809 os::Linux::hotspot_sigmask(thread);
810
811 // initialize floating point control register
812 os::Linux::init_thread_fpu_state();
813
814 // handshaking with parent thread
815 {
816 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
817
818 // notify parent thread
819 osthread->set_state(INITIALIZED);
820 sync->notify_all();
821
822 // wait until os::start_thread()
823 while (osthread->get_state() == INITIALIZED) {
824 sync->wait(Mutex::_no_safepoint_check_flag);
825 }
826 }
827
828 // call one more level start routine
829 thread->run();
830
831 return 0;
832 }
833
834 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
835 assert(thread->osthread() == NULL, "caller responsible");
836
837 // Allocate the OSThread object
838 OSThread* osthread = new OSThread(NULL, NULL);
839 if (osthread == NULL) {
840 return false;
841 }
842
843 // set the correct thread state
844 osthread->set_thread_type(thr_type);
845
846 // Initial state is ALLOCATED but not INITIALIZED
847 osthread->set_state(ALLOCATED);
848
849 thread->set_osthread(osthread);
850
851 // init thread attributes
852 pthread_attr_t attr;
853 pthread_attr_init(&attr);
854 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
855
856 // stack size
857 if (os::Linux::supports_variable_stack_size()) {
858 // calculate stack size if it's not specified by caller
859 if (stack_size == 0) {
860 stack_size = os::Linux::default_stack_size(thr_type);
861
862 switch (thr_type) {
863 case os::java_thread:
864 // Java threads use ThreadStackSize which default value can be
865 // changed with the flag -Xss
866 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
867 stack_size = JavaThread::stack_size_at_create();
868 break;
869 case os::compiler_thread:
870 if (CompilerThreadStackSize > 0) {
871 stack_size = (size_t)(CompilerThreadStackSize * K);
872 break;
873 } // else fall through:
874 // use VMThreadStackSize if CompilerThreadStackSize is not defined
875 case os::vm_thread:
876 case os::pgc_thread:
877 case os::cgc_thread:
878 case os::watcher_thread:
879 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
880 break;
881 }
882 }
883
884 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
885 pthread_attr_setstacksize(&attr, stack_size);
886 } else {
887 // let pthread_create() pick the default value.
888 }
889
890 // glibc guard page
891 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
892
893 ThreadState state;
894
895 {
896 // Serialize thread creation if we are running with fixed stack LinuxThreads
897 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
898 if (lock) {
899 os::Linux::createThread_lock()->lock_without_safepoint_check();
900 }
901
902 pthread_t tid;
903 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
904
905 pthread_attr_destroy(&attr);
906
907 if (ret != 0) {
908 if (PrintMiscellaneous && (Verbose || WizardMode)) {
909 perror("pthread_create()");
910 }
911 // Need to clean up stuff we've allocated so far
912 thread->set_osthread(NULL);
913 delete osthread;
914 if (lock) os::Linux::createThread_lock()->unlock();
915 return false;
916 }
917
918 // Store pthread info into the OSThread
919 osthread->set_pthread_id(tid);
920
921 // Wait until child thread is either initialized or aborted
922 {
923 Monitor* sync_with_child = osthread->startThread_lock();
924 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
925 while ((state = osthread->get_state()) == ALLOCATED) {
926 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
927 }
928 }
929
930 if (lock) {
931 os::Linux::createThread_lock()->unlock();
932 }
933 }
934
935 // Aborted due to thread limit being reached
936 if (state == ZOMBIE) {
937 thread->set_osthread(NULL);
938 delete osthread;
939 return false;
940 }
941
942 // The thread is returned suspended (in state INITIALIZED),
943 // and is started higher up in the call chain
944 assert(state == INITIALIZED, "race condition");
945 return true;
946 }
947
948 /////////////////////////////////////////////////////////////////////////////
949 // attach existing thread
950
951 // bootstrap the main thread
952 bool os::create_main_thread(JavaThread* thread) {
953 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
954 return create_attached_thread(thread);
955 }
956
957 bool os::create_attached_thread(JavaThread* thread) {
958 #ifdef ASSERT
959 thread->verify_not_published();
960 #endif
961
962 // Allocate the OSThread object
963 OSThread* osthread = new OSThread(NULL, NULL);
964
965 if (osthread == NULL) {
966 return false;
967 }
968
969 // Store pthread info into the OSThread
970 osthread->set_thread_id(os::Linux::gettid());
971 osthread->set_pthread_id(::pthread_self());
972
973 // initialize floating point control register
974 os::Linux::init_thread_fpu_state();
975
976 // Initial thread state is RUNNABLE
977 osthread->set_state(RUNNABLE);
978
979 thread->set_osthread(osthread);
980
981 if (UseNUMA) {
982 int lgrp_id = os::numa_get_group_id();
983 if (lgrp_id != -1) {
984 thread->set_lgrp_id(lgrp_id);
985 }
986 }
987
988 if (os::Linux::is_initial_thread()) {
989 // If current thread is initial thread, its stack is mapped on demand,
990 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
991 // the entire stack region to avoid SEGV in stack banging.
992 // It is also useful to get around the heap-stack-gap problem on SuSE
993 // kernel (see 4821821 for details). We first expand stack to the top
994 // of yellow zone, then enable stack yellow zone (order is significant,
995 // enabling yellow zone first will crash JVM on SuSE Linux), so there
996 // is no gap between the last two virtual memory regions.
997
998 JavaThread *jt = (JavaThread *)thread;
999 address addr = jt->stack_yellow_zone_base();
1000 assert(addr != NULL, "initialization problem?");
1001 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1002
1003 osthread->set_expanding_stack();
1004 os::Linux::manually_expand_stack(jt, addr);
1005 osthread->clear_expanding_stack();
1006 }
1007
1008 // initialize signal mask for this thread
1009 // and save the caller's signal mask
1010 os::Linux::hotspot_sigmask(thread);
1011
1012 return true;
1013 }
1014
1015 void os::pd_start_thread(Thread* thread) {
1016 OSThread * osthread = thread->osthread();
1017 assert(osthread->get_state() != INITIALIZED, "just checking");
1018 Monitor* sync_with_child = osthread->startThread_lock();
1019 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1020 sync_with_child->notify();
1021 }
1022
1023 // Free Linux resources related to the OSThread
1024 void os::free_thread(OSThread* osthread) {
1025 assert(osthread != NULL, "osthread not set");
1026
1027 if (Thread::current()->osthread() == osthread) {
1028 // Restore caller's signal mask
1029 sigset_t sigmask = osthread->caller_sigmask();
1030 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1031 }
1032
1033 delete osthread;
1034 }
1035
1036 //////////////////////////////////////////////////////////////////////////////
1037 // thread local storage
1038
1039 // Restore the thread pointer if the destructor is called. This is in case
1040 // someone from JNI code sets up a destructor with pthread_key_create to run
1041 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1042 // will hang or crash. When detachCurrentThread is called the key will be set
1043 // to null and we will not be called again. If detachCurrentThread is never
1044 // called we could loop forever depending on the pthread implementation.
1045 static void restore_thread_pointer(void* p) {
1046 Thread* thread = (Thread*) p;
1047 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1048 }
1049
1050 int os::allocate_thread_local_storage() {
1051 pthread_key_t key;
1052 int rslt = pthread_key_create(&key, restore_thread_pointer);
1053 assert(rslt == 0, "cannot allocate thread local storage");
1054 return (int)key;
1055 }
1056
1057 // Note: This is currently not used by VM, as we don't destroy TLS key
1058 // on VM exit.
1059 void os::free_thread_local_storage(int index) {
1060 int rslt = pthread_key_delete((pthread_key_t)index);
1061 assert(rslt == 0, "invalid index");
1062 }
1063
1064 void os::thread_local_storage_at_put(int index, void* value) {
1065 int rslt = pthread_setspecific((pthread_key_t)index, value);
1066 assert(rslt == 0, "pthread_setspecific failed");
1067 }
1068
1069 extern "C" Thread* get_thread() {
1070 return ThreadLocalStorage::thread();
1071 }
1072
1073 //////////////////////////////////////////////////////////////////////////////
1074 // initial thread
1075
1076 // Check if current thread is the initial thread, similar to Solaris thr_main.
1077 bool os::Linux::is_initial_thread(void) {
1078 char dummy;
1079 // If called before init complete, thread stack bottom will be null.
1080 // Can be called if fatal error occurs before initialization.
1081 if (initial_thread_stack_bottom() == NULL) return false;
1082 assert(initial_thread_stack_bottom() != NULL &&
1083 initial_thread_stack_size() != 0,
1084 "os::init did not locate initial thread's stack region");
1085 if ((address)&dummy >= initial_thread_stack_bottom() &&
1086 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1087 return true;
1088 else return false;
1089 }
1090
1091 // Find the virtual memory area that contains addr
1092 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1093 FILE *fp = fopen("/proc/self/maps", "r");
1094 if (fp) {
1095 address low, high;
1096 while (!feof(fp)) {
1097 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1098 if (low <= addr && addr < high) {
1099 if (vma_low) *vma_low = low;
1100 if (vma_high) *vma_high = high;
1101 fclose (fp);
1102 return true;
1103 }
1104 }
1105 for (;;) {
1106 int ch = fgetc(fp);
1107 if (ch == EOF || ch == (int)'\n') break;
1108 }
1109 }
1110 fclose(fp);
1111 }
1112 return false;
1113 }
1114
1115 // Locate initial thread stack. This special handling of initial thread stack
1116 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1117 // bogus value for initial thread.
1118 void os::Linux::capture_initial_stack(size_t max_size) {
1119 // stack size is the easy part, get it from RLIMIT_STACK
1120 size_t stack_size;
1121 struct rlimit rlim;
1122 getrlimit(RLIMIT_STACK, &rlim);
1123 stack_size = rlim.rlim_cur;
1124
1125 // 6308388: a bug in ld.so will relocate its own .data section to the
1126 // lower end of primordial stack; reduce ulimit -s value a little bit
1127 // so we won't install guard page on ld.so's data section.
1128 stack_size -= 2 * page_size();
1129
1130 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1131 // 7.1, in both cases we will get 2G in return value.
1132 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1133 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1134 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1135 // in case other parts in glibc still assumes 2M max stack size.
1136 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1137 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1138 if (stack_size > 2 * K * K IA64_ONLY(*2))
1139 stack_size = 2 * K * K IA64_ONLY(*2);
1140 // Try to figure out where the stack base (top) is. This is harder.
1141 //
1142 // When an application is started, glibc saves the initial stack pointer in
1143 // a global variable "__libc_stack_end", which is then used by system
1144 // libraries. __libc_stack_end should be pretty close to stack top. The
1145 // variable is available since the very early days. However, because it is
1146 // a private interface, it could disappear in the future.
1147 //
1148 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1149 // to __libc_stack_end, it is very close to stack top, but isn't the real
1150 // stack top. Note that /proc may not exist if VM is running as a chroot
1151 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1152 // /proc/<pid>/stat could change in the future (though unlikely).
1153 //
1154 // We try __libc_stack_end first. If that doesn't work, look for
1155 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1156 // as a hint, which should work well in most cases.
1157
1158 uintptr_t stack_start;
1159
1160 // try __libc_stack_end first
1161 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1162 if (p && *p) {
1163 stack_start = *p;
1164 } else {
1165 // see if we can get the start_stack field from /proc/self/stat
1166 FILE *fp;
1167 int pid;
1168 char state;
1169 int ppid;
1170 int pgrp;
1171 int session;
1172 int nr;
1173 int tpgrp;
1174 unsigned long flags;
1175 unsigned long minflt;
1176 unsigned long cminflt;
1177 unsigned long majflt;
1178 unsigned long cmajflt;
1179 unsigned long utime;
1180 unsigned long stime;
1181 long cutime;
1182 long cstime;
1183 long prio;
1184 long nice;
1185 long junk;
1186 long it_real;
1187 uintptr_t start;
1188 uintptr_t vsize;
1189 intptr_t rss;
1190 uintptr_t rsslim;
1191 uintptr_t scodes;
1192 uintptr_t ecode;
1193 int i;
1194
1195 // Figure what the primordial thread stack base is. Code is inspired
1196 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1197 // followed by command name surrounded by parentheses, state, etc.
1198 char stat[2048];
1199 int statlen;
1200
1201 fp = fopen("/proc/self/stat", "r");
1202 if (fp) {
1203 statlen = fread(stat, 1, 2047, fp);
1204 stat[statlen] = '\0';
1205 fclose(fp);
1206
1207 // Skip pid and the command string. Note that we could be dealing with
1208 // weird command names, e.g. user could decide to rename java launcher
1209 // to "java 1.4.2 :)", then the stat file would look like
1210 // 1234 (java 1.4.2 :)) R ... ...
1211 // We don't really need to know the command string, just find the last
1212 // occurrence of ")" and then start parsing from there. See bug 4726580.
1213 char * s = strrchr(stat, ')');
1214
1215 i = 0;
1216 if (s) {
1217 // Skip blank chars
1218 do s++; while (isspace(*s));
1219
1220 #define _UFM UINTX_FORMAT
1221 #define _DFM INTX_FORMAT
1222
1223 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1224 /* 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 */
1225 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,
1226 &state, /* 3 %c */
1227 &ppid, /* 4 %d */
1228 &pgrp, /* 5 %d */
1229 &session, /* 6 %d */
1230 &nr, /* 7 %d */
1231 &tpgrp, /* 8 %d */
1232 &flags, /* 9 %lu */
1233 &minflt, /* 10 %lu */
1234 &cminflt, /* 11 %lu */
1235 &majflt, /* 12 %lu */
1236 &cmajflt, /* 13 %lu */
1237 &utime, /* 14 %lu */
1238 &stime, /* 15 %lu */
1239 &cutime, /* 16 %ld */
1240 &cstime, /* 17 %ld */
1241 &prio, /* 18 %ld */
1242 &nice, /* 19 %ld */
1243 &junk, /* 20 %ld */
1244 &it_real, /* 21 %ld */
1245 &start, /* 22 UINTX_FORMAT */
1246 &vsize, /* 23 UINTX_FORMAT */
1247 &rss, /* 24 INTX_FORMAT */
1248 &rsslim, /* 25 UINTX_FORMAT */
1249 &scodes, /* 26 UINTX_FORMAT */
1250 &ecode, /* 27 UINTX_FORMAT */
1251 &stack_start); /* 28 UINTX_FORMAT */
1252 }
1253
1254 #undef _UFM
1255 #undef _DFM
1256
1257 if (i != 28 - 2) {
1258 assert(false, "Bad conversion from /proc/self/stat");
1259 // product mode - assume we are the initial thread, good luck in the
1260 // embedded case.
1261 warning("Can't detect initial thread stack location - bad conversion");
1262 stack_start = (uintptr_t) &rlim;
1263 }
1264 } else {
1265 // For some reason we can't open /proc/self/stat (for example, running on
1266 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1267 // most cases, so don't abort:
1268 warning("Can't detect initial thread stack location - no /proc/self/stat");
1269 stack_start = (uintptr_t) &rlim;
1270 }
1271 }
1272
1273 // Now we have a pointer (stack_start) very close to the stack top, the
1274 // next thing to do is to figure out the exact location of stack top. We
1275 // can find out the virtual memory area that contains stack_start by
1276 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1277 // and its upper limit is the real stack top. (again, this would fail if
1278 // running inside chroot, because /proc may not exist.)
1279
1280 uintptr_t stack_top;
1281 address low, high;
1282 if (find_vma((address)stack_start, &low, &high)) {
1283 // success, "high" is the true stack top. (ignore "low", because initial
1284 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1285 stack_top = (uintptr_t)high;
1286 } else {
1287 // failed, likely because /proc/self/maps does not exist
1288 warning("Can't detect initial thread stack location - find_vma failed");
1289 // best effort: stack_start is normally within a few pages below the real
1290 // stack top, use it as stack top, and reduce stack size so we won't put
1291 // guard page outside stack.
1292 stack_top = stack_start;
1293 stack_size -= 16 * page_size();
1294 }
1295
1296 // stack_top could be partially down the page so align it
1297 stack_top = align_size_up(stack_top, page_size());
1298
1299 if (max_size && stack_size > max_size) {
1300 _initial_thread_stack_size = max_size;
1301 } else {
1302 _initial_thread_stack_size = stack_size;
1303 }
1304
1305 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1306 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1307 }
1308
1309 ////////////////////////////////////////////////////////////////////////////////
1310 // time support
1311
1312 // Time since start-up in seconds to a fine granularity.
1313 // Used by VMSelfDestructTimer and the MemProfiler.
1314 double os::elapsedTime() {
1315
1316 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1317 }
1318
1319 jlong os::elapsed_counter() {
1320 return javaTimeNanos() - initial_time_count;
1321 }
1322
1323 jlong os::elapsed_frequency() {
1324 return NANOSECS_PER_SEC; // nanosecond resolution
1325 }
1326
1327 bool os::supports_vtime() { return true; }
1328 bool os::enable_vtime() { return false; }
1329 bool os::vtime_enabled() { return false; }
1330
1331 double os::elapsedVTime() {
1332 struct rusage usage;
1333 int retval = getrusage(RUSAGE_THREAD, &usage);
1334 if (retval == 0) {
1335 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);
1336 } else {
1337 // better than nothing, but not much
1338 return elapsedTime();
1339 }
1340 }
1341
1342 jlong os::javaTimeMillis() {
1343 timeval time;
1344 int status = gettimeofday(&time, NULL);
1345 assert(status != -1, "linux error");
1346 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1347 }
1348
1349 #ifndef CLOCK_MONOTONIC
1350 #define CLOCK_MONOTONIC (1)
1351 #endif
1352
1353 void os::Linux::clock_init() {
1354 // we do dlopen's in this particular order due to bug in linux
1355 // dynamical loader (see 6348968) leading to crash on exit
1356 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1357 if (handle == NULL) {
1358 handle = dlopen("librt.so", RTLD_LAZY);
1359 }
1360
1361 if (handle) {
1362 int (*clock_getres_func)(clockid_t, struct timespec*) =
1363 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1364 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1365 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1366 if (clock_getres_func && clock_gettime_func) {
1367 // See if monotonic clock is supported by the kernel. Note that some
1368 // early implementations simply return kernel jiffies (updated every
1369 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1370 // for nano time (though the monotonic property is still nice to have).
1371 // It's fixed in newer kernels, however clock_getres() still returns
1372 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1373 // resolution for now. Hopefully as people move to new kernels, this
1374 // won't be a problem.
1375 struct timespec res;
1376 struct timespec tp;
1377 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1378 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1379 // yes, monotonic clock is supported
1380 _clock_gettime = clock_gettime_func;
1381 return;
1382 } else {
1383 // close librt if there is no monotonic clock
1384 dlclose(handle);
1385 }
1386 }
1387 }
1388 warning("No monotonic clock was available - timed services may " \
1389 "be adversely affected if the time-of-day clock changes");
1390 }
1391
1392 #ifndef SYS_clock_getres
1393
1394 #if defined(IA32) || defined(AMD64)
1395 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1396 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1397 #else
1398 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1399 #define sys_clock_getres(x,y) -1
1400 #endif
1401
1402 #else
1403 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1404 #endif
1405
1406 void os::Linux::fast_thread_clock_init() {
1407 if (!UseLinuxPosixThreadCPUClocks) {
1408 return;
1409 }
1410 clockid_t clockid;
1411 struct timespec tp;
1412 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1413 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1414
1415 // Switch to using fast clocks for thread cpu time if
1416 // the sys_clock_getres() returns 0 error code.
1417 // Note, that some kernels may support the current thread
1418 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1419 // returned by the pthread_getcpuclockid().
1420 // If the fast Posix clocks are supported then the sys_clock_getres()
1421 // must return at least tp.tv_sec == 0 which means a resolution
1422 // better than 1 sec. This is extra check for reliability.
1423
1424 if(pthread_getcpuclockid_func &&
1425 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1426 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1427
1428 _supports_fast_thread_cpu_time = true;
1429 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1430 }
1431 }
1432
1433 jlong os::javaTimeNanos() {
1434 if (Linux::supports_monotonic_clock()) {
1435 struct timespec tp;
1436 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1437 assert(status == 0, "gettime error");
1438 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1439 return result;
1440 } else {
1441 timeval time;
1442 int status = gettimeofday(&time, NULL);
1443 assert(status != -1, "linux error");
1444 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1445 return 1000 * usecs;
1446 }
1447 }
1448
1449 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1450 if (Linux::supports_monotonic_clock()) {
1451 info_ptr->max_value = ALL_64_BITS;
1452
1453 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1454 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1455 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1456 } else {
1457 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1458 info_ptr->max_value = ALL_64_BITS;
1459
1460 // gettimeofday is a real time clock so it skips
1461 info_ptr->may_skip_backward = true;
1462 info_ptr->may_skip_forward = true;
1463 }
1464
1465 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1466 }
1467
1468 // Return the real, user, and system times in seconds from an
1469 // arbitrary fixed point in the past.
1470 bool os::getTimesSecs(double* process_real_time,
1471 double* process_user_time,
1472 double* process_system_time) {
1473 struct tms ticks;
1474 clock_t real_ticks = times(&ticks);
1475
1476 if (real_ticks == (clock_t) (-1)) {
1477 return false;
1478 } else {
1479 double ticks_per_second = (double) clock_tics_per_sec;
1480 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1481 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1482 *process_real_time = ((double) real_ticks) / ticks_per_second;
1483
1484 return true;
1485 }
1486 }
1487
1488
1489 char * os::local_time_string(char *buf, size_t buflen) {
1490 struct tm t;
1491 time_t long_time;
1492 time(&long_time);
1493 localtime_r(&long_time, &t);
1494 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1495 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1496 t.tm_hour, t.tm_min, t.tm_sec);
1497 return buf;
1498 }
1499
1500 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1501 return localtime_r(clock, res);
1502 }
1503
1504 ////////////////////////////////////////////////////////////////////////////////
1505 // runtime exit support
1506
1507 // Note: os::shutdown() might be called very early during initialization, or
1508 // called from signal handler. Before adding something to os::shutdown(), make
1509 // sure it is async-safe and can handle partially initialized VM.
1510 void os::shutdown() {
1511
1512 // allow PerfMemory to attempt cleanup of any persistent resources
1513 perfMemory_exit();
1514
1515 // needs to remove object in file system
1516 AttachListener::abort();
1517
1518 // flush buffered output, finish log files
1519 ostream_abort();
1520
1521 // Check for abort hook
1522 abort_hook_t abort_hook = Arguments::abort_hook();
1523 if (abort_hook != NULL) {
1524 abort_hook();
1525 }
1526
1527 }
1528
1529 // Note: os::abort() might be called very early during initialization, or
1530 // called from signal handler. Before adding something to os::abort(), make
1531 // sure it is async-safe and can handle partially initialized VM.
1532 void os::abort(bool dump_core) {
1533 os::shutdown();
1534 if (dump_core) {
1535 #ifndef PRODUCT
1536 fdStream out(defaultStream::output_fd());
1537 out.print_raw("Current thread is ");
1538 char buf[16];
1539 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1540 out.print_raw_cr(buf);
1541 out.print_raw_cr("Dumping core ...");
1542 #endif
1543 ::abort(); // dump core
1544 }
1545
1546 ::exit(1);
1547 }
1548
1549 // Die immediately, no exit hook, no abort hook, no cleanup.
1550 void os::die() {
1551 // _exit() on LinuxThreads only kills current thread
1552 ::abort();
1553 }
1554
1555
1556 // This method is a copy of JDK's sysGetLastErrorString
1557 // from src/solaris/hpi/src/system_md.c
1558
1559 size_t os::lasterror(char *buf, size_t len) {
1560
1561 if (errno == 0) return 0;
1562
1563 const char *s = ::strerror(errno);
1564 size_t n = ::strlen(s);
1565 if (n >= len) {
1566 n = len - 1;
1567 }
1568 ::strncpy(buf, s, n);
1569 buf[n] = '\0';
1570 return n;
1571 }
1572
1573 intx os::current_thread_id() { return (intx)pthread_self(); }
1574 int os::current_process_id() {
1575
1576 // Under the old linux thread library, linux gives each thread
1577 // its own process id. Because of this each thread will return
1578 // a different pid if this method were to return the result
1579 // of getpid(2). Linux provides no api that returns the pid
1580 // of the launcher thread for the vm. This implementation
1581 // returns a unique pid, the pid of the launcher thread
1582 // that starts the vm 'process'.
1583
1584 // Under the NPTL, getpid() returns the same pid as the
1585 // launcher thread rather than a unique pid per thread.
1586 // Use gettid() if you want the old pre NPTL behaviour.
1587
1588 // if you are looking for the result of a call to getpid() that
1589 // returns a unique pid for the calling thread, then look at the
1590 // OSThread::thread_id() method in osThread_linux.hpp file
1591
1592 return (int)(_initial_pid ? _initial_pid : getpid());
1593 }
1594
1595 // DLL functions
1596
1597 const char* os::dll_file_extension() { return ".so"; }
1598
1599 // This must be hard coded because it's the system's temporary
1600 // directory not the java application's temp directory, ala java.io.tmpdir.
1601 const char* os::get_temp_directory() { return "/tmp"; }
1602
1603 static bool file_exists(const char* filename) {
1604 struct stat statbuf;
1605 if (filename == NULL || strlen(filename) == 0) {
1606 return false;
1607 }
1608 return os::stat(filename, &statbuf) == 0;
1609 }
1610
1611 bool os::dll_build_name(char* buffer, size_t buflen,
1612 const char* pname, const char* fname) {
1613 bool retval = false;
1614 // Copied from libhpi
1615 const size_t pnamelen = pname ? strlen(pname) : 0;
1616
1617 // Return error on buffer overflow.
1618 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1619 return retval;
1620 }
1621
1622 if (pnamelen == 0) {
1623 snprintf(buffer, buflen, "lib%s.so", fname);
1624 retval = true;
1625 } else if (strchr(pname, *os::path_separator()) != NULL) {
1626 int n;
1627 char** pelements = split_path(pname, &n);
1628 if (pelements == NULL) {
1629 return false;
1630 }
1631 for (int i = 0 ; i < n ; i++) {
1632 // Really shouldn't be NULL, but check can't hurt
1633 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1634 continue; // skip the empty path values
1635 }
1636 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1637 if (file_exists(buffer)) {
1638 retval = true;
1639 break;
1640 }
1641 }
1642 // release the storage
1643 for (int i = 0 ; i < n ; i++) {
1644 if (pelements[i] != NULL) {
1645 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1646 }
1647 }
1648 if (pelements != NULL) {
1649 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1650 }
1651 } else {
1652 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1653 retval = true;
1654 }
1655 return retval;
1656 }
1657
1658 // check if addr is inside libjvm.so
1659 bool os::address_is_in_vm(address addr) {
1660 static address libjvm_base_addr;
1661 Dl_info dlinfo;
1662
1663 if (libjvm_base_addr == NULL) {
1664 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1665 libjvm_base_addr = (address)dlinfo.dli_fbase;
1666 }
1667 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1668 }
1669
1670 if (dladdr((void *)addr, &dlinfo) != 0) {
1671 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1672 }
1673
1674 return false;
1675 }
1676
1677 bool os::dll_address_to_function_name(address addr, char *buf,
1678 int buflen, int *offset) {
1679 // buf is not optional, but offset is optional
1680 assert(buf != NULL, "sanity check");
1681
1682 Dl_info dlinfo;
1683
1684 if (dladdr((void*)addr, &dlinfo) != 0) {
1685 // see if we have a matching symbol
1686 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1687 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1688 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1689 }
1690 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1691 return true;
1692 }
1693 // no matching symbol so try for just file info
1694 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1695 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1696 buf, buflen, offset, dlinfo.dli_fname)) {
1697 return true;
1698 }
1699 }
1700 }
1701
1702 buf[0] = '\0';
1703 if (offset != NULL) *offset = -1;
1704 return false;
1705 }
1706
1707 struct _address_to_library_name {
1708 address addr; // input : memory address
1709 size_t buflen; // size of fname
1710 char* fname; // output: library name
1711 address base; // library base addr
1712 };
1713
1714 static int address_to_library_name_callback(struct dl_phdr_info *info,
1715 size_t size, void *data) {
1716 int i;
1717 bool found = false;
1718 address libbase = NULL;
1719 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1720
1721 // iterate through all loadable segments
1722 for (i = 0; i < info->dlpi_phnum; i++) {
1723 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1724 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1725 // base address of a library is the lowest address of its loaded
1726 // segments.
1727 if (libbase == NULL || libbase > segbase) {
1728 libbase = segbase;
1729 }
1730 // see if 'addr' is within current segment
1731 if (segbase <= d->addr &&
1732 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1733 found = true;
1734 }
1735 }
1736 }
1737
1738 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1739 // so dll_address_to_library_name() can fall through to use dladdr() which
1740 // can figure out executable name from argv[0].
1741 if (found && info->dlpi_name && info->dlpi_name[0]) {
1742 d->base = libbase;
1743 if (d->fname) {
1744 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1745 }
1746 return 1;
1747 }
1748 return 0;
1749 }
1750
1751 bool os::dll_address_to_library_name(address addr, char* buf,
1752 int buflen, int* offset) {
1753 // buf is not optional, but offset is optional
1754 assert(buf != NULL, "sanity check");
1755
1756 Dl_info dlinfo;
1757 struct _address_to_library_name data;
1758
1759 // There is a bug in old glibc dladdr() implementation that it could resolve
1760 // to wrong library name if the .so file has a base address != NULL. Here
1761 // we iterate through the program headers of all loaded libraries to find
1762 // out which library 'addr' really belongs to. This workaround can be
1763 // removed once the minimum requirement for glibc is moved to 2.3.x.
1764 data.addr = addr;
1765 data.fname = buf;
1766 data.buflen = buflen;
1767 data.base = NULL;
1768 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1769
1770 if (rslt) {
1771 // buf already contains library name
1772 if (offset) *offset = addr - data.base;
1773 return true;
1774 }
1775 if (dladdr((void*)addr, &dlinfo) != 0) {
1776 if (dlinfo.dli_fname != NULL) {
1777 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1778 }
1779 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1780 *offset = addr - (address)dlinfo.dli_fbase;
1781 }
1782 return true;
1783 }
1784
1785 buf[0] = '\0';
1786 if (offset) *offset = -1;
1787 return false;
1788 }
1789
1790 // Loads .dll/.so and
1791 // in case of error it checks if .dll/.so was built for the
1792 // same architecture as Hotspot is running on
1793
1794
1795 // Remember the stack's state. The Linux dynamic linker will change
1796 // the stack to 'executable' at most once, so we must safepoint only once.
1797 bool os::Linux::_stack_is_executable = false;
1798
1799 // VM operation that loads a library. This is necessary if stack protection
1800 // of the Java stacks can be lost during loading the library. If we
1801 // do not stop the Java threads, they can stack overflow before the stacks
1802 // are protected again.
1803 class VM_LinuxDllLoad: public VM_Operation {
1804 private:
1805 const char *_filename;
1806 char *_ebuf;
1807 int _ebuflen;
1808 void *_lib;
1809 public:
1810 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1811 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1812 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1813 void doit() {
1814 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1815 os::Linux::_stack_is_executable = true;
1816 }
1817 void* loaded_library() { return _lib; }
1818 };
1819
1820 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1821 {
1822 void * result = NULL;
1823 bool load_attempted = false;
1824
1825 // Check whether the library to load might change execution rights
1826 // of the stack. If they are changed, the protection of the stack
1827 // guard pages will be lost. We need a safepoint to fix this.
1828 //
1829 // See Linux man page execstack(8) for more info.
1830 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1831 ElfFile ef(filename);
1832 if (!ef.specifies_noexecstack()) {
1833 if (!is_init_completed()) {
1834 os::Linux::_stack_is_executable = true;
1835 // This is OK - No Java threads have been created yet, and hence no
1836 // stack guard pages to fix.
1837 //
1838 // This should happen only when you are building JDK7 using a very
1839 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1840 //
1841 // Dynamic loader will make all stacks executable after
1842 // this function returns, and will not do that again.
1843 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1844 } else {
1845 warning("You have loaded library %s which might have disabled stack guard. "
1846 "The VM will try to fix the stack guard now.\n"
1847 "It's highly recommended that you fix the library with "
1848 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1849 filename);
1850
1851 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1852 JavaThread *jt = JavaThread::current();
1853 if (jt->thread_state() != _thread_in_native) {
1854 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1855 // that requires ExecStack. Cannot enter safe point. Let's give up.
1856 warning("Unable to fix stack guard. Giving up.");
1857 } else {
1858 if (!LoadExecStackDllInVMThread) {
1859 // This is for the case where the DLL has an static
1860 // constructor function that executes JNI code. We cannot
1861 // load such DLLs in the VMThread.
1862 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1863 }
1864
1865 ThreadInVMfromNative tiv(jt);
1866 debug_only(VMNativeEntryWrapper vew;)
1867
1868 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1869 VMThread::execute(&op);
1870 if (LoadExecStackDllInVMThread) {
1871 result = op.loaded_library();
1872 }
1873 load_attempted = true;
1874 }
1875 }
1876 }
1877 }
1878
1879 if (!load_attempted) {
1880 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1881 }
1882
1883 if (result != NULL) {
1884 // Successful loading
1885 return result;
1886 }
1887
1888 Elf32_Ehdr elf_head;
1889 int diag_msg_max_length=ebuflen-strlen(ebuf);
1890 char* diag_msg_buf=ebuf+strlen(ebuf);
1891
1892 if (diag_msg_max_length==0) {
1893 // No more space in ebuf for additional diagnostics message
1894 return NULL;
1895 }
1896
1897
1898 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1899
1900 if (file_descriptor < 0) {
1901 // Can't open library, report dlerror() message
1902 return NULL;
1903 }
1904
1905 bool failed_to_read_elf_head=
1906 (sizeof(elf_head)!=
1907 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1908
1909 ::close(file_descriptor);
1910 if (failed_to_read_elf_head) {
1911 // file i/o error - report dlerror() msg
1912 return NULL;
1913 }
1914
1915 typedef struct {
1916 Elf32_Half code; // Actual value as defined in elf.h
1917 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1918 char elf_class; // 32 or 64 bit
1919 char endianess; // MSB or LSB
1920 char* name; // String representation
1921 } arch_t;
1922
1923 #ifndef EM_486
1924 #define EM_486 6 /* Intel 80486 */
1925 #endif
1926
1927 static const arch_t arch_array[]={
1928 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1929 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1930 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1931 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1932 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1933 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1934 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1935 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1936 #if defined(VM_LITTLE_ENDIAN)
1937 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1938 #else
1939 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1940 #endif
1941 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1942 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1943 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1944 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1945 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1946 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1947 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1948 };
1949
1950 #if (defined IA32)
1951 static Elf32_Half running_arch_code=EM_386;
1952 #elif (defined AMD64)
1953 static Elf32_Half running_arch_code=EM_X86_64;
1954 #elif (defined IA64)
1955 static Elf32_Half running_arch_code=EM_IA_64;
1956 #elif (defined __sparc) && (defined _LP64)
1957 static Elf32_Half running_arch_code=EM_SPARCV9;
1958 #elif (defined __sparc) && (!defined _LP64)
1959 static Elf32_Half running_arch_code=EM_SPARC;
1960 #elif (defined __powerpc64__)
1961 static Elf32_Half running_arch_code=EM_PPC64;
1962 #elif (defined __powerpc__)
1963 static Elf32_Half running_arch_code=EM_PPC;
1964 #elif (defined ARM)
1965 static Elf32_Half running_arch_code=EM_ARM;
1966 #elif (defined S390)
1967 static Elf32_Half running_arch_code=EM_S390;
1968 #elif (defined ALPHA)
1969 static Elf32_Half running_arch_code=EM_ALPHA;
1970 #elif (defined MIPSEL)
1971 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1972 #elif (defined PARISC)
1973 static Elf32_Half running_arch_code=EM_PARISC;
1974 #elif (defined MIPS)
1975 static Elf32_Half running_arch_code=EM_MIPS;
1976 #elif (defined M68K)
1977 static Elf32_Half running_arch_code=EM_68K;
1978 #else
1979 #error Method os::dll_load requires that one of following is defined:\
1980 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1981 #endif
1982
1983 // Identify compatability class for VM's architecture and library's architecture
1984 // Obtain string descriptions for architectures
1985
1986 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1987 int running_arch_index=-1;
1988
1989 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1990 if (running_arch_code == arch_array[i].code) {
1991 running_arch_index = i;
1992 }
1993 if (lib_arch.code == arch_array[i].code) {
1994 lib_arch.compat_class = arch_array[i].compat_class;
1995 lib_arch.name = arch_array[i].name;
1996 }
1997 }
1998
1999 assert(running_arch_index != -1,
2000 "Didn't find running architecture code (running_arch_code) in arch_array");
2001 if (running_arch_index == -1) {
2002 // Even though running architecture detection failed
2003 // we may still continue with reporting dlerror() message
2004 return NULL;
2005 }
2006
2007 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2008 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2009 return NULL;
2010 }
2011
2012 #ifndef S390
2013 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2014 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2015 return NULL;
2016 }
2017 #endif // !S390
2018
2019 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2020 if ( lib_arch.name!=NULL ) {
2021 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2022 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2023 lib_arch.name, arch_array[running_arch_index].name);
2024 } else {
2025 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2026 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2027 lib_arch.code,
2028 arch_array[running_arch_index].name);
2029 }
2030 }
2031
2032 return NULL;
2033 }
2034
2035 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2036 void * result = ::dlopen(filename, RTLD_LAZY);
2037 if (result == NULL) {
2038 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2039 ebuf[ebuflen-1] = '\0';
2040 }
2041 return result;
2042 }
2043
2044 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2045 void * result = NULL;
2046 if (LoadExecStackDllInVMThread) {
2047 result = dlopen_helper(filename, ebuf, ebuflen);
2048 }
2049
2050 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2051 // library that requires an executable stack, or which does not have this
2052 // stack attribute set, dlopen changes the stack attribute to executable. The
2053 // read protection of the guard pages gets lost.
2054 //
2055 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2056 // may have been queued at the same time.
2057
2058 if (!_stack_is_executable) {
2059 JavaThread *jt = Threads::first();
2060
2061 while (jt) {
2062 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2063 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2064 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2065 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2066 warning("Attempt to reguard stack yellow zone failed.");
2067 }
2068 }
2069 jt = jt->next();
2070 }
2071 }
2072
2073 return result;
2074 }
2075
2076 /*
2077 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2078 * chances are you might want to run the generated bits against glibc-2.0
2079 * libdl.so, so always use locking for any version of glibc.
2080 */
2081 void* os::dll_lookup(void* handle, const char* name) {
2082 pthread_mutex_lock(&dl_mutex);
2083 void* res = dlsym(handle, name);
2084 pthread_mutex_unlock(&dl_mutex);
2085 return res;
2086 }
2087
2088 void* os::get_default_process_handle() {
2089 return (void*)::dlopen(NULL, RTLD_LAZY);
2090 }
2091
2092 static bool _print_ascii_file(const char* filename, outputStream* st) {
2093 int fd = ::open(filename, O_RDONLY);
2094 if (fd == -1) {
2095 return false;
2096 }
2097
2098 char buf[32];
2099 int bytes;
2100 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2101 st->print_raw(buf, bytes);
2102 }
2103
2104 ::close(fd);
2105
2106 return true;
2107 }
2108
2109 void os::print_dll_info(outputStream *st) {
2110 st->print_cr("Dynamic libraries:");
2111
2112 char fname[32];
2113 pid_t pid = os::Linux::gettid();
2114
2115 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2116
2117 if (!_print_ascii_file(fname, st)) {
2118 st->print("Can not get library information for pid = %d\n", pid);
2119 }
2120 }
2121
2122 void os::print_os_info_brief(outputStream* st) {
2123 os::Linux::print_distro_info(st);
2124
2125 os::Posix::print_uname_info(st);
2126
2127 os::Linux::print_libversion_info(st);
2128
2129 }
2130
2131 void os::print_os_info(outputStream* st) {
2132 st->print("OS:");
2133
2134 os::Linux::print_distro_info(st);
2135
2136 os::Posix::print_uname_info(st);
2137
2138 // Print warning if unsafe chroot environment detected
2139 if (unsafe_chroot_detected) {
2140 st->print("WARNING!! ");
2141 st->print_cr("%s", unstable_chroot_error);
2142 }
2143
2144 os::Linux::print_libversion_info(st);
2145
2146 os::Posix::print_rlimit_info(st);
2147
2148 os::Posix::print_load_average(st);
2149
2150 os::Linux::print_full_memory_info(st);
2151 }
2152
2153 // Try to identify popular distros.
2154 // Most Linux distributions have a /etc/XXX-release file, which contains
2155 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2156 // file that also contains the OS version string. Some have more than one
2157 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2158 // /etc/redhat-release.), so the order is important.
2159 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2160 // their own specific XXX-release file as well as a redhat-release file.
2161 // Because of this the XXX-release file needs to be searched for before the
2162 // redhat-release file.
2163 // Since Red Hat has a lsb-release file that is not very descriptive the
2164 // search for redhat-release needs to be before lsb-release.
2165 // Since the lsb-release file is the new standard it needs to be searched
2166 // before the older style release files.
2167 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2168 // next to last resort. The os-release file is a new standard that contains
2169 // distribution information and the system-release file seems to be an old
2170 // standard that has been replaced by the lsb-release and os-release files.
2171 // Searching for the debian_version file is the last resort. It contains
2172 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2173 // "Debian " is printed before the contents of the debian_version file.
2174 void os::Linux::print_distro_info(outputStream* st) {
2175 if (!_print_ascii_file("/etc/oracle-release", st) &&
2176 !_print_ascii_file("/etc/mandriva-release", st) &&
2177 !_print_ascii_file("/etc/mandrake-release", st) &&
2178 !_print_ascii_file("/etc/sun-release", st) &&
2179 !_print_ascii_file("/etc/redhat-release", st) &&
2180 !_print_ascii_file("/etc/lsb-release", st) &&
2181 !_print_ascii_file("/etc/SuSE-release", st) &&
2182 !_print_ascii_file("/etc/turbolinux-release", st) &&
2183 !_print_ascii_file("/etc/gentoo-release", st) &&
2184 !_print_ascii_file("/etc/ltib-release", st) &&
2185 !_print_ascii_file("/etc/angstrom-version", st) &&
2186 !_print_ascii_file("/etc/system-release", st) &&
2187 !_print_ascii_file("/etc/os-release", st)) {
2188
2189 if (file_exists("/etc/debian_version")) {
2190 st->print("Debian ");
2191 _print_ascii_file("/etc/debian_version", st);
2192 } else {
2193 st->print("Linux");
2194 }
2195 }
2196 st->cr();
2197 }
2198
2199 void os::Linux::print_libversion_info(outputStream* st) {
2200 // libc, pthread
2201 st->print("libc:");
2202 st->print("%s ", os::Linux::glibc_version());
2203 st->print("%s ", os::Linux::libpthread_version());
2204 if (os::Linux::is_LinuxThreads()) {
2205 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2206 }
2207 st->cr();
2208 }
2209
2210 void os::Linux::print_full_memory_info(outputStream* st) {
2211 st->print("\n/proc/meminfo:\n");
2212 _print_ascii_file("/proc/meminfo", st);
2213 st->cr();
2214 }
2215
2216 void os::print_memory_info(outputStream* st) {
2217
2218 st->print("Memory:");
2219 st->print(" %dk page", os::vm_page_size()>>10);
2220
2221 // values in struct sysinfo are "unsigned long"
2222 struct sysinfo si;
2223 sysinfo(&si);
2224
2225 st->print(", physical " UINT64_FORMAT "k",
2226 os::physical_memory() >> 10);
2227 st->print("(" UINT64_FORMAT "k free)",
2228 os::available_memory() >> 10);
2229 st->print(", swap " UINT64_FORMAT "k",
2230 ((jlong)si.totalswap * si.mem_unit) >> 10);
2231 st->print("(" UINT64_FORMAT "k free)",
2232 ((jlong)si.freeswap * si.mem_unit) >> 10);
2233 st->cr();
2234 }
2235
2236 void os::pd_print_cpu_info(outputStream* st) {
2237 st->print("\n/proc/cpuinfo:\n");
2238 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2239 st->print(" <Not Available>");
2240 }
2241 st->cr();
2242 }
2243
2244 void os::print_siginfo(outputStream* st, void* siginfo) {
2245 const siginfo_t* si = (const siginfo_t*)siginfo;
2246
2247 os::Posix::print_siginfo_brief(st, si);
2248 #if INCLUDE_CDS
2249 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2250 UseSharedSpaces) {
2251 FileMapInfo* mapinfo = FileMapInfo::current_info();
2252 if (mapinfo->is_in_shared_space(si->si_addr)) {
2253 st->print("\n\nError accessing class data sharing archive." \
2254 " Mapped file inaccessible during execution, " \
2255 " possible disk/network problem.");
2256 }
2257 }
2258 #endif
2259 st->cr();
2260 }
2261
2262
2263 static void print_signal_handler(outputStream* st, int sig,
2264 char* buf, size_t buflen);
2265
2266 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2267 st->print_cr("Signal Handlers:");
2268 print_signal_handler(st, SIGSEGV, buf, buflen);
2269 print_signal_handler(st, SIGBUS , buf, buflen);
2270 print_signal_handler(st, SIGFPE , buf, buflen);
2271 print_signal_handler(st, SIGPIPE, buf, buflen);
2272 print_signal_handler(st, SIGXFSZ, buf, buflen);
2273 print_signal_handler(st, SIGILL , buf, buflen);
2274 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2275 print_signal_handler(st, SR_signum, buf, buflen);
2276 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2277 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2278 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2279 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2280 #if defined(PPC64)
2281 print_signal_handler(st, SIGTRAP, buf, buflen);
2282 #endif
2283 }
2284
2285 static char saved_jvm_path[MAXPATHLEN] = {0};
2286
2287 // Find the full path to the current module, libjvm.so
2288 void os::jvm_path(char *buf, jint buflen) {
2289 // Error checking.
2290 if (buflen < MAXPATHLEN) {
2291 assert(false, "must use a large-enough buffer");
2292 buf[0] = '\0';
2293 return;
2294 }
2295 // Lazy resolve the path to current module.
2296 if (saved_jvm_path[0] != 0) {
2297 strcpy(buf, saved_jvm_path);
2298 return;
2299 }
2300
2301 char dli_fname[MAXPATHLEN];
2302 bool ret = dll_address_to_library_name(
2303 CAST_FROM_FN_PTR(address, os::jvm_path),
2304 dli_fname, sizeof(dli_fname), NULL);
2305 assert(ret, "cannot locate libjvm");
2306 char *rp = NULL;
2307 if (ret && dli_fname[0] != '\0') {
2308 rp = realpath(dli_fname, buf);
2309 }
2310 if (rp == NULL)
2311 return;
2312
2313 if (Arguments::created_by_gamma_launcher()) {
2314 // Support for the gamma launcher. Typical value for buf is
2315 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2316 // the right place in the string, then assume we are installed in a JDK and
2317 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2318 // up the path so it looks like libjvm.so is installed there (append a
2319 // fake suffix hotspot/libjvm.so).
2320 const char *p = buf + strlen(buf) - 1;
2321 for (int count = 0; p > buf && count < 5; ++count) {
2322 for (--p; p > buf && *p != '/'; --p)
2323 /* empty */ ;
2324 }
2325
2326 if (strncmp(p, "/jre/lib/", 9) != 0) {
2327 // Look for JAVA_HOME in the environment.
2328 char* java_home_var = ::getenv("JAVA_HOME");
2329 if (java_home_var != NULL && java_home_var[0] != 0) {
2330 char* jrelib_p;
2331 int len;
2332
2333 // Check the current module name "libjvm.so".
2334 p = strrchr(buf, '/');
2335 assert(strstr(p, "/libjvm") == p, "invalid library name");
2336
2337 rp = realpath(java_home_var, buf);
2338 if (rp == NULL)
2339 return;
2340
2341 // determine if this is a legacy image or modules image
2342 // modules image doesn't have "jre" subdirectory
2343 len = strlen(buf);
2344 assert(len < buflen, "Ran out of buffer room");
2345 jrelib_p = buf + len;
2346 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2347 if (0 != access(buf, F_OK)) {
2348 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2349 }
2350
2351 if (0 == access(buf, F_OK)) {
2352 // Use current module name "libjvm.so"
2353 len = strlen(buf);
2354 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2355 } else {
2356 // Go back to path of .so
2357 rp = realpath(dli_fname, buf);
2358 if (rp == NULL)
2359 return;
2360 }
2361 }
2362 }
2363 }
2364
2365 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2366 }
2367
2368 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2369 // no prefix required, not even "_"
2370 }
2371
2372 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2373 // no suffix required
2374 }
2375
2376 ////////////////////////////////////////////////////////////////////////////////
2377 // sun.misc.Signal support
2378
2379 static volatile jint sigint_count = 0;
2380
2381 static void
2382 UserHandler(int sig, void *siginfo, void *context) {
2383 // 4511530 - sem_post is serialized and handled by the manager thread. When
2384 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2385 // don't want to flood the manager thread with sem_post requests.
2386 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2387 return;
2388
2389 // Ctrl-C is pressed during error reporting, likely because the error
2390 // handler fails to abort. Let VM die immediately.
2391 if (sig == SIGINT && is_error_reported()) {
2392 os::die();
2393 }
2394
2395 os::signal_notify(sig);
2396 }
2397
2398 void* os::user_handler() {
2399 return CAST_FROM_FN_PTR(void*, UserHandler);
2400 }
2401
2402 class Semaphore : public StackObj {
2403 public:
2404 Semaphore();
2405 ~Semaphore();
2406 void signal();
2407 void wait();
2408 bool trywait();
2409 bool timedwait(unsigned int sec, int nsec);
2410 private:
2411 sem_t _semaphore;
2412 };
2413
2414 Semaphore::Semaphore() {
2415 sem_init(&_semaphore, 0, 0);
2416 }
2417
2418 Semaphore::~Semaphore() {
2419 sem_destroy(&_semaphore);
2420 }
2421
2422 void Semaphore::signal() {
2423 sem_post(&_semaphore);
2424 }
2425
2426 void Semaphore::wait() {
2427 sem_wait(&_semaphore);
2428 }
2429
2430 bool Semaphore::trywait() {
2431 return sem_trywait(&_semaphore) == 0;
2432 }
2433
2434 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2435
2436 struct timespec ts;
2437 // Semaphore's are always associated with CLOCK_REALTIME
2438 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2439 // see unpackTime for discussion on overflow checking
2440 if (sec >= MAX_SECS) {
2441 ts.tv_sec += MAX_SECS;
2442 ts.tv_nsec = 0;
2443 } else {
2444 ts.tv_sec += sec;
2445 ts.tv_nsec += nsec;
2446 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2447 ts.tv_nsec -= NANOSECS_PER_SEC;
2448 ++ts.tv_sec; // note: this must be <= max_secs
2449 }
2450 }
2451
2452 while (1) {
2453 int result = sem_timedwait(&_semaphore, &ts);
2454 if (result == 0) {
2455 return true;
2456 } else if (errno == EINTR) {
2457 continue;
2458 } else if (errno == ETIMEDOUT) {
2459 return false;
2460 } else {
2461 return false;
2462 }
2463 }
2464 }
2465
2466 extern "C" {
2467 typedef void (*sa_handler_t)(int);
2468 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2469 }
2470
2471 void* os::signal(int signal_number, void* handler) {
2472 struct sigaction sigAct, oldSigAct;
2473
2474 sigfillset(&(sigAct.sa_mask));
2475 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2476 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2477
2478 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2479 // -1 means registration failed
2480 return (void *)-1;
2481 }
2482
2483 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2484 }
2485
2486 void os::signal_raise(int signal_number) {
2487 ::raise(signal_number);
2488 }
2489
2490 /*
2491 * The following code is moved from os.cpp for making this
2492 * code platform specific, which it is by its very nature.
2493 */
2494
2495 // Will be modified when max signal is changed to be dynamic
2496 int os::sigexitnum_pd() {
2497 return NSIG;
2498 }
2499
2500 // a counter for each possible signal value
2501 static volatile jint pending_signals[NSIG+1] = { 0 };
2502
2503 // Linux(POSIX) specific hand shaking semaphore.
2504 static sem_t sig_sem;
2505 static Semaphore sr_semaphore;
2506
2507 void os::signal_init_pd() {
2508 // Initialize signal structures
2509 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2510
2511 // Initialize signal semaphore
2512 ::sem_init(&sig_sem, 0, 0);
2513 }
2514
2515 void os::signal_notify(int sig) {
2516 Atomic::inc(&pending_signals[sig]);
2517 ::sem_post(&sig_sem);
2518 }
2519
2520 static int check_pending_signals(bool wait) {
2521 Atomic::store(0, &sigint_count);
2522 for (;;) {
2523 for (int i = 0; i < NSIG + 1; i++) {
2524 jint n = pending_signals[i];
2525 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2526 return i;
2527 }
2528 }
2529 if (!wait) {
2530 return -1;
2531 }
2532 JavaThread *thread = JavaThread::current();
2533 ThreadBlockInVM tbivm(thread);
2534
2535 bool threadIsSuspended;
2536 do {
2537 thread->set_suspend_equivalent();
2538 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2539 ::sem_wait(&sig_sem);
2540
2541 // were we externally suspended while we were waiting?
2542 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2543 if (threadIsSuspended) {
2544 //
2545 // The semaphore has been incremented, but while we were waiting
2546 // another thread suspended us. We don't want to continue running
2547 // while suspended because that would surprise the thread that
2548 // suspended us.
2549 //
2550 ::sem_post(&sig_sem);
2551
2552 thread->java_suspend_self();
2553 }
2554 } while (threadIsSuspended);
2555 }
2556 }
2557
2558 int os::signal_lookup() {
2559 return check_pending_signals(false);
2560 }
2561
2562 int os::signal_wait() {
2563 return check_pending_signals(true);
2564 }
2565
2566 ////////////////////////////////////////////////////////////////////////////////
2567 // Virtual Memory
2568
2569 int os::vm_page_size() {
2570 // Seems redundant as all get out
2571 assert(os::Linux::page_size() != -1, "must call os::init");
2572 return os::Linux::page_size();
2573 }
2574
2575 // Solaris allocates memory by pages.
2576 int os::vm_allocation_granularity() {
2577 assert(os::Linux::page_size() != -1, "must call os::init");
2578 return os::Linux::page_size();
2579 }
2580
2581 // Rationale behind this function:
2582 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2583 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2584 // samples for JITted code. Here we create private executable mapping over the code cache
2585 // and then we can use standard (well, almost, as mapping can change) way to provide
2586 // info for the reporting script by storing timestamp and location of symbol
2587 void linux_wrap_code(char* base, size_t size) {
2588 static volatile jint cnt = 0;
2589
2590 if (!UseOprofile) {
2591 return;
2592 }
2593
2594 char buf[PATH_MAX+1];
2595 int num = Atomic::add(1, &cnt);
2596
2597 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2598 os::get_temp_directory(), os::current_process_id(), num);
2599 unlink(buf);
2600
2601 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2602
2603 if (fd != -1) {
2604 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2605 if (rv != (off_t)-1) {
2606 if (::write(fd, "", 1) == 1) {
2607 mmap(base, size,
2608 PROT_READ|PROT_WRITE|PROT_EXEC,
2609 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2610 }
2611 }
2612 ::close(fd);
2613 unlink(buf);
2614 }
2615 }
2616
2617 static bool recoverable_mmap_error(int err) {
2618 // See if the error is one we can let the caller handle. This
2619 // list of errno values comes from JBS-6843484. I can't find a
2620 // Linux man page that documents this specific set of errno
2621 // values so while this list currently matches Solaris, it may
2622 // change as we gain experience with this failure mode.
2623 switch (err) {
2624 case EBADF:
2625 case EINVAL:
2626 case ENOTSUP:
2627 // let the caller deal with these errors
2628 return true;
2629
2630 default:
2631 // Any remaining errors on this OS can cause our reserved mapping
2632 // to be lost. That can cause confusion where different data
2633 // structures think they have the same memory mapped. The worst
2634 // scenario is if both the VM and a library think they have the
2635 // same memory mapped.
2636 return false;
2637 }
2638 }
2639
2640 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2641 int err) {
2642 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2643 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2644 strerror(err), err);
2645 }
2646
2647 static void warn_fail_commit_memory(char* addr, size_t size,
2648 size_t alignment_hint, bool exec,
2649 int err) {
2650 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2651 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2652 alignment_hint, exec, strerror(err), err);
2653 }
2654
2655 // NOTE: Linux kernel does not really reserve the pages for us.
2656 // All it does is to check if there are enough free pages
2657 // left at the time of mmap(). This could be a potential
2658 // problem.
2659 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2660 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2661 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2662 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2663 if (res != (uintptr_t) MAP_FAILED) {
2664 if (UseNUMAInterleaving) {
2665 numa_make_global(addr, size);
2666 }
2667 return 0;
2668 }
2669
2670 int err = errno; // save errno from mmap() call above
2671
2672 if (!recoverable_mmap_error(err)) {
2673 warn_fail_commit_memory(addr, size, exec, err);
2674 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2675 }
2676
2677 return err;
2678 }
2679
2680 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2681 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2682 }
2683
2684 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2685 const char* mesg) {
2686 assert(mesg != NULL, "mesg must be specified");
2687 int err = os::Linux::commit_memory_impl(addr, size, exec);
2688 if (err != 0) {
2689 // the caller wants all commit errors to exit with the specified mesg:
2690 warn_fail_commit_memory(addr, size, exec, err);
2691 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2692 }
2693 }
2694
2695 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2696 #ifndef MAP_HUGETLB
2697 #define MAP_HUGETLB 0x40000
2698 #endif
2699
2700 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2701 #ifndef MADV_HUGEPAGE
2702 #define MADV_HUGEPAGE 14
2703 #endif
2704
2705 int os::Linux::commit_memory_impl(char* addr, size_t size,
2706 size_t alignment_hint, bool exec) {
2707 int err = os::Linux::commit_memory_impl(addr, size, exec);
2708 if (err == 0) {
2709 realign_memory(addr, size, alignment_hint);
2710 }
2711 return err;
2712 }
2713
2714 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2715 bool exec) {
2716 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2717 }
2718
2719 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2720 size_t alignment_hint, bool exec,
2721 const char* mesg) {
2722 assert(mesg != NULL, "mesg must be specified");
2723 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2724 if (err != 0) {
2725 // the caller wants all commit errors to exit with the specified mesg:
2726 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2727 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2728 }
2729 }
2730
2731 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2732 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2733 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2734 // be supported or the memory may already be backed by huge pages.
2735 ::madvise(addr, bytes, MADV_HUGEPAGE);
2736 }
2737 }
2738
2739 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2740 // This method works by doing an mmap over an existing mmaping and effectively discarding
2741 // the existing pages. However it won't work for SHM-based large pages that cannot be
2742 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2743 // small pages on top of the SHM segment. This method always works for small pages, so we
2744 // allow that in any case.
2745 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2746 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2747 }
2748 }
2749
2750 void os::numa_make_global(char *addr, size_t bytes) {
2751 Linux::numa_interleave_memory(addr, bytes);
2752 }
2753
2754 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2755 // bind policy to MPOL_PREFERRED for the current thread.
2756 #define USE_MPOL_PREFERRED 0
2757
2758 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2759 // To make NUMA and large pages more robust when both enabled, we need to ease
2760 // the requirements on where the memory should be allocated. MPOL_BIND is the
2761 // default policy and it will force memory to be allocated on the specified
2762 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2763 // the specified node, but will not force it. Using this policy will prevent
2764 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2765 // free large pages.
2766 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2767 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2768 }
2769
2770 bool os::numa_topology_changed() { return false; }
2771
2772 size_t os::numa_get_groups_num() {
2773 int max_node = Linux::numa_max_node();
2774 return max_node > 0 ? max_node + 1 : 1;
2775 }
2776
2777 int os::numa_get_group_id() {
2778 int cpu_id = Linux::sched_getcpu();
2779 if (cpu_id != -1) {
2780 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2781 if (lgrp_id != -1) {
2782 return lgrp_id;
2783 }
2784 }
2785 return 0;
2786 }
2787
2788 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2789 for (size_t i = 0; i < size; i++) {
2790 ids[i] = i;
2791 }
2792 return size;
2793 }
2794
2795 bool os::get_page_info(char *start, page_info* info) {
2796 return false;
2797 }
2798
2799 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2800 return end;
2801 }
2802
2803
2804 int os::Linux::sched_getcpu_syscall(void) {
2805 unsigned int cpu;
2806 int retval = -1;
2807
2808 #if defined(IA32)
2809 # ifndef SYS_getcpu
2810 # define SYS_getcpu 318
2811 # endif
2812 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2813 #elif defined(AMD64)
2814 // Unfortunately we have to bring all these macros here from vsyscall.h
2815 // to be able to compile on old linuxes.
2816 # define __NR_vgetcpu 2
2817 # define VSYSCALL_START (-10UL << 20)
2818 # define VSYSCALL_SIZE 1024
2819 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2820 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2821 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2822 retval = vgetcpu(&cpu, NULL, NULL);
2823 #endif
2824
2825 return (retval == -1) ? retval : cpu;
2826 }
2827
2828 // Something to do with the numa-aware allocator needs these symbols
2829 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2830 extern "C" JNIEXPORT void numa_error(char *where) { }
2831 extern "C" JNIEXPORT int fork1() { return fork(); }
2832
2833
2834 // If we are running with libnuma version > 2, then we should
2835 // be trying to use symbols with versions 1.1
2836 // If we are running with earlier version, which did not have symbol versions,
2837 // we should use the base version.
2838 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2839 void *f = dlvsym(handle, name, "libnuma_1.1");
2840 if (f == NULL) {
2841 f = dlsym(handle, name);
2842 }
2843 return f;
2844 }
2845
2846 bool os::Linux::libnuma_init() {
2847 // sched_getcpu() should be in libc.
2848 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2849 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2850
2851 // If it's not, try a direct syscall.
2852 if (sched_getcpu() == -1)
2853 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2854
2855 if (sched_getcpu() != -1) { // Does it work?
2856 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2857 if (handle != NULL) {
2858 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2859 libnuma_dlsym(handle, "numa_node_to_cpus")));
2860 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2861 libnuma_dlsym(handle, "numa_max_node")));
2862 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2863 libnuma_dlsym(handle, "numa_available")));
2864 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2865 libnuma_dlsym(handle, "numa_tonode_memory")));
2866 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2867 libnuma_dlsym(handle, "numa_interleave_memory")));
2868 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2869 libnuma_dlsym(handle, "numa_set_bind_policy")));
2870
2871
2872 if (numa_available() != -1) {
2873 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2874 // Create a cpu -> node mapping
2875 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2876 rebuild_cpu_to_node_map();
2877 return true;
2878 }
2879 }
2880 }
2881 return false;
2882 }
2883
2884 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2885 // The table is later used in get_node_by_cpu().
2886 void os::Linux::rebuild_cpu_to_node_map() {
2887 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2888 // in libnuma (possible values are starting from 16,
2889 // and continuing up with every other power of 2, but less
2890 // than the maximum number of CPUs supported by kernel), and
2891 // is a subject to change (in libnuma version 2 the requirements
2892 // are more reasonable) we'll just hardcode the number they use
2893 // in the library.
2894 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2895
2896 size_t cpu_num = os::active_processor_count();
2897 size_t cpu_map_size = NCPUS / BitsPerCLong;
2898 size_t cpu_map_valid_size =
2899 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2900
2901 cpu_to_node()->clear();
2902 cpu_to_node()->at_grow(cpu_num - 1);
2903 size_t node_num = numa_get_groups_num();
2904
2905 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2906 for (size_t i = 0; i < node_num; i++) {
2907 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2908 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2909 if (cpu_map[j] != 0) {
2910 for (size_t k = 0; k < BitsPerCLong; k++) {
2911 if (cpu_map[j] & (1UL << k)) {
2912 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2913 }
2914 }
2915 }
2916 }
2917 }
2918 }
2919 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2920 }
2921
2922 int os::Linux::get_node_by_cpu(int cpu_id) {
2923 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2924 return cpu_to_node()->at(cpu_id);
2925 }
2926 return -1;
2927 }
2928
2929 GrowableArray<int>* os::Linux::_cpu_to_node;
2930 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2931 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2932 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2933 os::Linux::numa_available_func_t os::Linux::_numa_available;
2934 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2935 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2936 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2937 unsigned long* os::Linux::_numa_all_nodes;
2938
2939 bool os::pd_uncommit_memory(char* addr, size_t size) {
2940 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2941 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2942 return res != (uintptr_t) MAP_FAILED;
2943 }
2944
2945 static
2946 address get_stack_commited_bottom(address bottom, size_t size) {
2947 address nbot = bottom;
2948 address ntop = bottom + size;
2949
2950 size_t page_sz = os::vm_page_size();
2951 unsigned pages = size / page_sz;
2952
2953 unsigned char vec[1];
2954 unsigned imin = 1, imax = pages + 1, imid;
2955 int mincore_return_value = 0;
2956
2957 assert(imin <= imax, "Unexpected page size");
2958
2959 while (imin < imax) {
2960 imid = (imax + imin) / 2;
2961 nbot = ntop - (imid * page_sz);
2962
2963 // Use a trick with mincore to check whether the page is mapped or not.
2964 // mincore sets vec to 1 if page resides in memory and to 0 if page
2965 // is swapped output but if page we are asking for is unmapped
2966 // it returns -1,ENOMEM
2967 mincore_return_value = mincore(nbot, page_sz, vec);
2968
2969 if (mincore_return_value == -1) {
2970 // Page is not mapped go up
2971 // to find first mapped page
2972 if (errno != EAGAIN) {
2973 assert(errno == ENOMEM, "Unexpected mincore errno");
2974 imax = imid;
2975 }
2976 } else {
2977 // Page is mapped go down
2978 // to find first not mapped page
2979 imin = imid + 1;
2980 }
2981 }
2982
2983 nbot = nbot + page_sz;
2984
2985 // Adjust stack bottom one page up if last checked page is not mapped
2986 if (mincore_return_value == -1) {
2987 nbot = nbot + page_sz;
2988 }
2989
2990 return nbot;
2991 }
2992
2993
2994 // Linux uses a growable mapping for the stack, and if the mapping for
2995 // the stack guard pages is not removed when we detach a thread the
2996 // stack cannot grow beyond the pages where the stack guard was
2997 // mapped. If at some point later in the process the stack expands to
2998 // that point, the Linux kernel cannot expand the stack any further
2999 // because the guard pages are in the way, and a segfault occurs.
3000 //
3001 // However, it's essential not to split the stack region by unmapping
3002 // a region (leaving a hole) that's already part of the stack mapping,
3003 // so if the stack mapping has already grown beyond the guard pages at
3004 // the time we create them, we have to truncate the stack mapping.
3005 // So, we need to know the extent of the stack mapping when
3006 // create_stack_guard_pages() is called.
3007
3008 // We only need this for stacks that are growable: at the time of
3009 // writing thread stacks don't use growable mappings (i.e. those
3010 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3011 // only applies to the main thread.
3012
3013 // If the (growable) stack mapping already extends beyond the point
3014 // where we're going to put our guard pages, truncate the mapping at
3015 // that point by munmap()ping it. This ensures that when we later
3016 // munmap() the guard pages we don't leave a hole in the stack
3017 // mapping. This only affects the main/initial thread
3018
3019 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3020
3021 if (os::Linux::is_initial_thread()) {
3022 // As we manually grow stack up to bottom inside create_attached_thread(),
3023 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3024 // we don't need to do anything special.
3025 // Check it first, before calling heavy function.
3026 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3027 unsigned char vec[1];
3028
3029 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3030 // Fallback to slow path on all errors, including EAGAIN
3031 stack_extent = (uintptr_t) get_stack_commited_bottom(
3032 os::Linux::initial_thread_stack_bottom(),
3033 (size_t)addr - stack_extent);
3034 }
3035
3036 if (stack_extent < (uintptr_t)addr) {
3037 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3038 }
3039 }
3040
3041 return os::commit_memory(addr, size, !ExecMem);
3042 }
3043
3044 // If this is a growable mapping, remove the guard pages entirely by
3045 // munmap()ping them. If not, just call uncommit_memory(). This only
3046 // affects the main/initial thread, but guard against future OS changes
3047 // It's safe to always unmap guard pages for initial thread because we
3048 // always place it right after end of the mapped region
3049
3050 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3051 uintptr_t stack_extent, stack_base;
3052
3053 if (os::Linux::is_initial_thread()) {
3054 return ::munmap(addr, size) == 0;
3055 }
3056
3057 return os::uncommit_memory(addr, size);
3058 }
3059
3060 static address _highest_vm_reserved_address = NULL;
3061
3062 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3063 // at 'requested_addr'. If there are existing memory mappings at the same
3064 // location, however, they will be overwritten. If 'fixed' is false,
3065 // 'requested_addr' is only treated as a hint, the return value may or
3066 // may not start from the requested address. Unlike Linux mmap(), this
3067 // function returns NULL to indicate failure.
3068 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3069 char * addr;
3070 int flags;
3071
3072 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3073 if (fixed) {
3074 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3075 flags |= MAP_FIXED;
3076 }
3077
3078 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3079 // touch an uncommitted page. Otherwise, the read/write might
3080 // succeed if we have enough swap space to back the physical page.
3081 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3082 flags, -1, 0);
3083
3084 if (addr != MAP_FAILED) {
3085 // anon_mmap() should only get called during VM initialization,
3086 // don't need lock (actually we can skip locking even it can be called
3087 // from multiple threads, because _highest_vm_reserved_address is just a
3088 // hint about the upper limit of non-stack memory regions.)
3089 if ((address)addr + bytes > _highest_vm_reserved_address) {
3090 _highest_vm_reserved_address = (address)addr + bytes;
3091 }
3092 }
3093
3094 return addr == MAP_FAILED ? NULL : addr;
3095 }
3096
3097 // Don't update _highest_vm_reserved_address, because there might be memory
3098 // regions above addr + size. If so, releasing a memory region only creates
3099 // a hole in the address space, it doesn't help prevent heap-stack collision.
3100 //
3101 static int anon_munmap(char * addr, size_t size) {
3102 return ::munmap(addr, size) == 0;
3103 }
3104
3105 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3106 size_t alignment_hint) {
3107 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3108 }
3109
3110 bool os::pd_release_memory(char* addr, size_t size) {
3111 return anon_munmap(addr, size);
3112 }
3113
3114 static address highest_vm_reserved_address() {
3115 return _highest_vm_reserved_address;
3116 }
3117
3118 static bool linux_mprotect(char* addr, size_t size, int prot) {
3119 // Linux wants the mprotect address argument to be page aligned.
3120 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3121
3122 // According to SUSv3, mprotect() should only be used with mappings
3123 // established by mmap(), and mmap() always maps whole pages. Unaligned
3124 // 'addr' likely indicates problem in the VM (e.g. trying to change
3125 // protection of malloc'ed or statically allocated memory). Check the
3126 // caller if you hit this assert.
3127 assert(addr == bottom, "sanity check");
3128
3129 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3130 return ::mprotect(bottom, size, prot) == 0;
3131 }
3132
3133 // Set protections specified
3134 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3135 bool is_committed) {
3136 unsigned int p = 0;
3137 switch (prot) {
3138 case MEM_PROT_NONE: p = PROT_NONE; break;
3139 case MEM_PROT_READ: p = PROT_READ; break;
3140 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3141 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3142 default:
3143 ShouldNotReachHere();
3144 }
3145 // is_committed is unused.
3146 return linux_mprotect(addr, bytes, p);
3147 }
3148
3149 bool os::guard_memory(char* addr, size_t size) {
3150 return linux_mprotect(addr, size, PROT_NONE);
3151 }
3152
3153 bool os::unguard_memory(char* addr, size_t size) {
3154 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3155 }
3156
3157 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3158 bool result = false;
3159 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3160 MAP_ANONYMOUS|MAP_PRIVATE,
3161 -1, 0);
3162 if (p != MAP_FAILED) {
3163 void *aligned_p = align_ptr_up(p, page_size);
3164
3165 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3166
3167 munmap(p, page_size * 2);
3168 }
3169
3170 if (warn && !result) {
3171 warning("TransparentHugePages is not supported by the operating system.");
3172 }
3173
3174 return result;
3175 }
3176
3177 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3178 bool result = false;
3179 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3180 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3181 -1, 0);
3182
3183 if (p != MAP_FAILED) {
3184 // We don't know if this really is a huge page or not.
3185 FILE *fp = fopen("/proc/self/maps", "r");
3186 if (fp) {
3187 while (!feof(fp)) {
3188 char chars[257];
3189 long x = 0;
3190 if (fgets(chars, sizeof(chars), fp)) {
3191 if (sscanf(chars, "%lx-%*x", &x) == 1
3192 && x == (long)p) {
3193 if (strstr (chars, "hugepage")) {
3194 result = true;
3195 break;
3196 }
3197 }
3198 }
3199 }
3200 fclose(fp);
3201 }
3202 munmap(p, page_size);
3203 }
3204
3205 if (warn && !result) {
3206 warning("HugeTLBFS is not supported by the operating system.");
3207 }
3208
3209 return result;
3210 }
3211
3212 /*
3213 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3214 *
3215 * From the coredump_filter documentation:
3216 *
3217 * - (bit 0) anonymous private memory
3218 * - (bit 1) anonymous shared memory
3219 * - (bit 2) file-backed private memory
3220 * - (bit 3) file-backed shared memory
3221 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3222 * effective only if the bit 2 is cleared)
3223 * - (bit 5) hugetlb private memory
3224 * - (bit 6) hugetlb shared memory
3225 */
3226 static void set_coredump_filter(void) {
3227 FILE *f;
3228 long cdm;
3229
3230 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3231 return;
3232 }
3233
3234 if (fscanf(f, "%lx", &cdm) != 1) {
3235 fclose(f);
3236 return;
3237 }
3238
3239 rewind(f);
3240
3241 if ((cdm & LARGEPAGES_BIT) == 0) {
3242 cdm |= LARGEPAGES_BIT;
3243 fprintf(f, "%#lx", cdm);
3244 }
3245
3246 fclose(f);
3247 }
3248
3249 // Large page support
3250
3251 static size_t _large_page_size = 0;
3252
3253 size_t os::Linux::find_large_page_size() {
3254 size_t large_page_size = 0;
3255
3256 // large_page_size on Linux is used to round up heap size. x86 uses either
3257 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3258 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3259 // page as large as 256M.
3260 //
3261 // Here we try to figure out page size by parsing /proc/meminfo and looking
3262 // for a line with the following format:
3263 // Hugepagesize: 2048 kB
3264 //
3265 // If we can't determine the value (e.g. /proc is not mounted, or the text
3266 // format has been changed), we'll use the largest page size supported by
3267 // the processor.
3268
3269 #ifndef ZERO
3270 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3271 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3272 #endif // ZERO
3273
3274 FILE *fp = fopen("/proc/meminfo", "r");
3275 if (fp) {
3276 while (!feof(fp)) {
3277 int x = 0;
3278 char buf[16];
3279 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3280 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3281 large_page_size = x * K;
3282 break;
3283 }
3284 } else {
3285 // skip to next line
3286 for (;;) {
3287 int ch = fgetc(fp);
3288 if (ch == EOF || ch == (int)'\n') break;
3289 }
3290 }
3291 }
3292 fclose(fp);
3293 }
3294
3295 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3296 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3297 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3298 proper_unit_for_byte_size(large_page_size));
3299 }
3300
3301 return large_page_size;
3302 }
3303
3304 size_t os::Linux::setup_large_page_size() {
3305 _large_page_size = Linux::find_large_page_size();
3306 const size_t default_page_size = (size_t)Linux::page_size();
3307 if (_large_page_size > default_page_size) {
3308 _page_sizes[0] = _large_page_size;
3309 _page_sizes[1] = default_page_size;
3310 _page_sizes[2] = 0;
3311 }
3312
3313 return _large_page_size;
3314 }
3315
3316 bool os::Linux::setup_large_page_type(size_t page_size) {
3317 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3318 FLAG_IS_DEFAULT(UseSHM) &&
3319 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3320
3321 // The type of large pages has not been specified by the user.
3322
3323 // Try UseHugeTLBFS and then UseSHM.
3324 UseHugeTLBFS = UseSHM = true;
3325
3326 // Don't try UseTransparentHugePages since there are known
3327 // performance issues with it turned on. This might change in the future.
3328 UseTransparentHugePages = false;
3329 }
3330
3331 if (UseTransparentHugePages) {
3332 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3333 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3334 UseHugeTLBFS = false;
3335 UseSHM = false;
3336 return true;
3337 }
3338 UseTransparentHugePages = false;
3339 }
3340
3341 if (UseHugeTLBFS) {
3342 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3343 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3344 UseSHM = false;
3345 return true;
3346 }
3347 UseHugeTLBFS = false;
3348 }
3349
3350 return UseSHM;
3351 }
3352
3353 void os::large_page_init() {
3354 if (!UseLargePages &&
3355 !UseTransparentHugePages &&
3356 !UseHugeTLBFS &&
3357 !UseSHM) {
3358 // Not using large pages.
3359 return;
3360 }
3361
3362 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3363 // The user explicitly turned off large pages.
3364 // Ignore the rest of the large pages flags.
3365 UseTransparentHugePages = false;
3366 UseHugeTLBFS = false;
3367 UseSHM = false;
3368 return;
3369 }
3370
3371 size_t large_page_size = Linux::setup_large_page_size();
3372 UseLargePages = Linux::setup_large_page_type(large_page_size);
3373
3374 set_coredump_filter();
3375 }
3376
3377 #ifndef SHM_HUGETLB
3378 #define SHM_HUGETLB 04000
3379 #endif
3380
3381 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3382 // "exec" is passed in but not used. Creating the shared image for
3383 // the code cache doesn't have an SHM_X executable permission to check.
3384 assert(UseLargePages && UseSHM, "only for SHM large pages");
3385 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3386
3387 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3388 return NULL; // Fallback to small pages.
3389 }
3390
3391 key_t key = IPC_PRIVATE;
3392 char *addr;
3393
3394 bool warn_on_failure = UseLargePages &&
3395 (!FLAG_IS_DEFAULT(UseLargePages) ||
3396 !FLAG_IS_DEFAULT(UseSHM) ||
3397 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3398 );
3399 char msg[128];
3400
3401 // Create a large shared memory region to attach to based on size.
3402 // Currently, size is the total size of the heap
3403 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3404 if (shmid == -1) {
3405 // Possible reasons for shmget failure:
3406 // 1. shmmax is too small for Java heap.
3407 // > check shmmax value: cat /proc/sys/kernel/shmmax
3408 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3409 // 2. not enough large page memory.
3410 // > check available large pages: cat /proc/meminfo
3411 // > increase amount of large pages:
3412 // echo new_value > /proc/sys/vm/nr_hugepages
3413 // Note 1: different Linux may use different name for this property,
3414 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3415 // Note 2: it's possible there's enough physical memory available but
3416 // they are so fragmented after a long run that they can't
3417 // coalesce into large pages. Try to reserve large pages when
3418 // the system is still "fresh".
3419 if (warn_on_failure) {
3420 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3421 warning("%s", msg);
3422 }
3423 return NULL;
3424 }
3425
3426 // attach to the region
3427 addr = (char*)shmat(shmid, req_addr, 0);
3428 int err = errno;
3429
3430 // Remove shmid. If shmat() is successful, the actual shared memory segment
3431 // will be deleted when it's detached by shmdt() or when the process
3432 // terminates. If shmat() is not successful this will remove the shared
3433 // segment immediately.
3434 shmctl(shmid, IPC_RMID, NULL);
3435
3436 if ((intptr_t)addr == -1) {
3437 if (warn_on_failure) {
3438 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3439 warning("%s", msg);
3440 }
3441 return NULL;
3442 }
3443
3444 return addr;
3445 }
3446
3447 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3448 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3449
3450 bool warn_on_failure = UseLargePages &&
3451 (!FLAG_IS_DEFAULT(UseLargePages) ||
3452 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3453 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3454
3455 if (warn_on_failure) {
3456 char msg[128];
3457 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3458 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3459 warning("%s", msg);
3460 }
3461 }
3462
3463 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3464 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3465 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3466 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3467
3468 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3469 char* addr = (char*)::mmap(req_addr, bytes, prot,
3470 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3471 -1, 0);
3472
3473 if (addr == MAP_FAILED) {
3474 warn_on_large_pages_failure(req_addr, bytes, errno);
3475 return NULL;
3476 }
3477
3478 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3479
3480 return addr;
3481 }
3482
3483 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3484 size_t large_page_size = os::large_page_size();
3485
3486 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3487
3488 // Allocate small pages.
3489
3490 char* start;
3491 if (req_addr != NULL) {
3492 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3493 assert(is_size_aligned(bytes, alignment), "Must be");
3494 start = os::reserve_memory(bytes, req_addr);
3495 assert(start == NULL || start == req_addr, "Must be");
3496 } else {
3497 start = os::reserve_memory_aligned(bytes, alignment);
3498 }
3499
3500 if (start == NULL) {
3501 return NULL;
3502 }
3503
3504 assert(is_ptr_aligned(start, alignment), "Must be");
3505
3506 if (MemTracker::tracking_level() > NMT_minimal) {
3507 // os::reserve_memory_special will record this memory area.
3508 // Need to release it here to prevent overlapping reservations.
3509 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3510 tkr.record((address)start, bytes);
3511 }
3512
3513 char* end = start + bytes;
3514
3515 // Find the regions of the allocated chunk that can be promoted to large pages.
3516 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3517 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3518
3519 size_t lp_bytes = lp_end - lp_start;
3520
3521 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3522
3523 if (lp_bytes == 0) {
3524 // The mapped region doesn't even span the start and the end of a large page.
3525 // Fall back to allocate a non-special area.
3526 ::munmap(start, end - start);
3527 return NULL;
3528 }
3529
3530 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3531
3532
3533 void* result;
3534
3535 if (start != lp_start) {
3536 result = ::mmap(start, lp_start - start, prot,
3537 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3538 -1, 0);
3539 if (result == MAP_FAILED) {
3540 ::munmap(lp_start, end - lp_start);
3541 return NULL;
3542 }
3543 }
3544
3545 result = ::mmap(lp_start, lp_bytes, prot,
3546 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3547 -1, 0);
3548 if (result == MAP_FAILED) {
3549 warn_on_large_pages_failure(req_addr, bytes, errno);
3550 // If the mmap above fails, the large pages region will be unmapped and we
3551 // have regions before and after with small pages. Release these regions.
3552 //
3553 // | mapped | unmapped | mapped |
3554 // ^ ^ ^ ^
3555 // start lp_start lp_end end
3556 //
3557 ::munmap(start, lp_start - start);
3558 ::munmap(lp_end, end - lp_end);
3559 return NULL;
3560 }
3561
3562 if (lp_end != end) {
3563 result = ::mmap(lp_end, end - lp_end, prot,
3564 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3565 -1, 0);
3566 if (result == MAP_FAILED) {
3567 ::munmap(start, lp_end - start);
3568 return NULL;
3569 }
3570 }
3571
3572 return start;
3573 }
3574
3575 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3576 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3577 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3578 assert(is_power_of_2(alignment), "Must be");
3579 assert(is_power_of_2(os::large_page_size()), "Must be");
3580 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3581
3582 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3583 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3584 } else {
3585 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3586 }
3587 }
3588
3589 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3590 assert(UseLargePages, "only for large pages");
3591
3592 char* addr;
3593 if (UseSHM) {
3594 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3595 } else {
3596 assert(UseHugeTLBFS, "must be");
3597 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3598 }
3599
3600 if (addr != NULL) {
3601 if (UseNUMAInterleaving) {
3602 numa_make_global(addr, bytes);
3603 }
3604
3605 // The memory is committed
3606 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3607 }
3608
3609 return addr;
3610 }
3611
3612 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3613 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3614 return shmdt(base) == 0;
3615 }
3616
3617 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3618 return pd_release_memory(base, bytes);
3619 }
3620
3621 bool os::release_memory_special(char* base, size_t bytes) {
3622 bool res;
3623 if (MemTracker::tracking_level() > NMT_minimal) {
3624 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3625 res = os::Linux::release_memory_special_impl(base, bytes);
3626 if (res) {
3627 tkr.record((address)base, bytes);
3628 }
3629
3630 } else {
3631 res = os::Linux::release_memory_special_impl(base, bytes);
3632 }
3633 return res;
3634 }
3635
3636 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3637 assert(UseLargePages, "only for large pages");
3638 bool res;
3639
3640 if (UseSHM) {
3641 res = os::Linux::release_memory_special_shm(base, bytes);
3642 } else {
3643 assert(UseHugeTLBFS, "must be");
3644 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3645 }
3646 return res;
3647 }
3648
3649 size_t os::large_page_size() {
3650 return _large_page_size;
3651 }
3652
3653 // With SysV SHM the entire memory region must be allocated as shared
3654 // memory.
3655 // HugeTLBFS allows application to commit large page memory on demand.
3656 // However, when committing memory with HugeTLBFS fails, the region
3657 // that was supposed to be committed will lose the old reservation
3658 // and allow other threads to steal that memory region. Because of this
3659 // behavior we can't commit HugeTLBFS memory.
3660 bool os::can_commit_large_page_memory() {
3661 return UseTransparentHugePages;
3662 }
3663
3664 bool os::can_execute_large_page_memory() {
3665 return UseTransparentHugePages || UseHugeTLBFS;
3666 }
3667
3668 // Reserve memory at an arbitrary address, only if that area is
3669 // available (and not reserved for something else).
3670
3671 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3672 const int max_tries = 10;
3673 char* base[max_tries];
3674 size_t size[max_tries];
3675 const size_t gap = 0x000000;
3676
3677 // Assert only that the size is a multiple of the page size, since
3678 // that's all that mmap requires, and since that's all we really know
3679 // about at this low abstraction level. If we need higher alignment,
3680 // we can either pass an alignment to this method or verify alignment
3681 // in one of the methods further up the call chain. See bug 5044738.
3682 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3683
3684 // Repeatedly allocate blocks until the block is allocated at the
3685 // right spot. Give up after max_tries. Note that reserve_memory() will
3686 // automatically update _highest_vm_reserved_address if the call is
3687 // successful. The variable tracks the highest memory address every reserved
3688 // by JVM. It is used to detect heap-stack collision if running with
3689 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3690 // space than needed, it could confuse the collision detecting code. To
3691 // solve the problem, save current _highest_vm_reserved_address and
3692 // calculate the correct value before return.
3693 address old_highest = _highest_vm_reserved_address;
3694
3695 // Linux mmap allows caller to pass an address as hint; give it a try first,
3696 // if kernel honors the hint then we can return immediately.
3697 char * addr = anon_mmap(requested_addr, bytes, false);
3698 if (addr == requested_addr) {
3699 return requested_addr;
3700 }
3701
3702 if (addr != NULL) {
3703 // mmap() is successful but it fails to reserve at the requested address
3704 anon_munmap(addr, bytes);
3705 }
3706
3707 int i;
3708 for (i = 0; i < max_tries; ++i) {
3709 base[i] = reserve_memory(bytes);
3710
3711 if (base[i] != NULL) {
3712 // Is this the block we wanted?
3713 if (base[i] == requested_addr) {
3714 size[i] = bytes;
3715 break;
3716 }
3717
3718 // Does this overlap the block we wanted? Give back the overlapped
3719 // parts and try again.
3720
3721 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3722 if (top_overlap >= 0 && top_overlap < bytes) {
3723 unmap_memory(base[i], top_overlap);
3724 base[i] += top_overlap;
3725 size[i] = bytes - top_overlap;
3726 } else {
3727 size_t bottom_overlap = base[i] + bytes - requested_addr;
3728 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3729 unmap_memory(requested_addr, bottom_overlap);
3730 size[i] = bytes - bottom_overlap;
3731 } else {
3732 size[i] = bytes;
3733 }
3734 }
3735 }
3736 }
3737
3738 // Give back the unused reserved pieces.
3739
3740 for (int j = 0; j < i; ++j) {
3741 if (base[j] != NULL) {
3742 unmap_memory(base[j], size[j]);
3743 }
3744 }
3745
3746 if (i < max_tries) {
3747 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3748 return requested_addr;
3749 } else {
3750 _highest_vm_reserved_address = old_highest;
3751 return NULL;
3752 }
3753 }
3754
3755 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3756 return ::read(fd, buf, nBytes);
3757 }
3758
3759 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3760 // Solaris uses poll(), linux uses park().
3761 // Poll() is likely a better choice, assuming that Thread.interrupt()
3762 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3763 // SIGSEGV, see 4355769.
3764
3765 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3766 assert(thread == Thread::current(), "thread consistency check");
3767
3768 ParkEvent * const slp = thread->_SleepEvent ;
3769 slp->reset() ;
3770 OrderAccess::fence() ;
3771
3772 if (interruptible) {
3773 jlong prevtime = javaTimeNanos();
3774
3775 for (;;) {
3776 if (os::is_interrupted(thread, true)) {
3777 return OS_INTRPT;
3778 }
3779
3780 jlong newtime = javaTimeNanos();
3781
3782 if (newtime - prevtime < 0) {
3783 // time moving backwards, should only happen if no monotonic clock
3784 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3785 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3786 } else {
3787 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3788 }
3789
3790 if(millis <= 0) {
3791 return OS_OK;
3792 }
3793
3794 prevtime = newtime;
3795
3796 {
3797 assert(thread->is_Java_thread(), "sanity check");
3798 JavaThread *jt = (JavaThread *) thread;
3799 ThreadBlockInVM tbivm(jt);
3800 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3801
3802 jt->set_suspend_equivalent();
3803 // cleared by handle_special_suspend_equivalent_condition() or
3804 // java_suspend_self() via check_and_wait_while_suspended()
3805
3806 slp->park(millis);
3807
3808 // were we externally suspended while we were waiting?
3809 jt->check_and_wait_while_suspended();
3810 }
3811 }
3812 } else {
3813 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3814 jlong prevtime = javaTimeNanos();
3815
3816 for (;;) {
3817 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3818 // the 1st iteration ...
3819 jlong newtime = javaTimeNanos();
3820
3821 if (newtime - prevtime < 0) {
3822 // time moving backwards, should only happen if no monotonic clock
3823 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3824 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3825 } else {
3826 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3827 }
3828
3829 if(millis <= 0) break ;
3830
3831 prevtime = newtime;
3832 slp->park(millis);
3833 }
3834 return OS_OK ;
3835 }
3836 }
3837
3838 //
3839 // Short sleep, direct OS call.
3840 //
3841 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3842 // sched_yield(2) will actually give up the CPU:
3843 //
3844 // * Alone on this pariticular CPU, keeps running.
3845 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3846 // (pre 2.6.39).
3847 //
3848 // So calling this with 0 is an alternative.
3849 //
3850 void os::naked_short_sleep(jlong ms) {
3851 struct timespec req;
3852
3853 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3854 req.tv_sec = 0;
3855 if (ms > 0) {
3856 req.tv_nsec = (ms % 1000) * 1000000;
3857 }
3858 else {
3859 req.tv_nsec = 1;
3860 }
3861
3862 nanosleep(&req, NULL);
3863
3864 return;
3865 }
3866
3867 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3868 void os::infinite_sleep() {
3869 while (true) { // sleep forever ...
3870 ::sleep(100); // ... 100 seconds at a time
3871 }
3872 }
3873
3874 // Used to convert frequent JVM_Yield() to nops
3875 bool os::dont_yield() {
3876 return DontYieldALot;
3877 }
3878
3879 void os::yield() {
3880 sched_yield();
3881 }
3882
3883 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3884
3885 void os::yield_all(int attempts) {
3886 // Yields to all threads, including threads with lower priorities
3887 // Threads on Linux are all with same priority. The Solaris style
3888 // os::yield_all() with nanosleep(1ms) is not necessary.
3889 sched_yield();
3890 }
3891
3892 // Called from the tight loops to possibly influence time-sharing heuristics
3893 void os::loop_breaker(int attempts) {
3894 os::yield_all(attempts);
3895 }
3896
3897 ////////////////////////////////////////////////////////////////////////////////
3898 // thread priority support
3899
3900 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3901 // only supports dynamic priority, static priority must be zero. For real-time
3902 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3903 // However, for large multi-threaded applications, SCHED_RR is not only slower
3904 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3905 // of 5 runs - Sep 2005).
3906 //
3907 // The following code actually changes the niceness of kernel-thread/LWP. It
3908 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3909 // not the entire user process, and user level threads are 1:1 mapped to kernel
3910 // threads. It has always been the case, but could change in the future. For
3911 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3912 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3913
3914 int os::java_to_os_priority[CriticalPriority + 1] = {
3915 19, // 0 Entry should never be used
3916
3917 4, // 1 MinPriority
3918 3, // 2
3919 2, // 3
3920
3921 1, // 4
3922 0, // 5 NormPriority
3923 -1, // 6
3924
3925 -2, // 7
3926 -3, // 8
3927 -4, // 9 NearMaxPriority
3928
3929 -5, // 10 MaxPriority
3930
3931 -5 // 11 CriticalPriority
3932 };
3933
3934 static int prio_init() {
3935 if (ThreadPriorityPolicy == 1) {
3936 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3937 // if effective uid is not root. Perhaps, a more elegant way of doing
3938 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3939 if (geteuid() != 0) {
3940 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3941 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3942 }
3943 ThreadPriorityPolicy = 0;
3944 }
3945 }
3946 if (UseCriticalJavaThreadPriority) {
3947 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3948 }
3949 return 0;
3950 }
3951
3952 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3953 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3954
3955 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3956 return (ret == 0) ? OS_OK : OS_ERR;
3957 }
3958
3959 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3960 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3961 *priority_ptr = java_to_os_priority[NormPriority];
3962 return OS_OK;
3963 }
3964
3965 errno = 0;
3966 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3967 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3968 }
3969
3970 // Hint to the underlying OS that a task switch would not be good.
3971 // Void return because it's a hint and can fail.
3972 void os::hint_no_preempt() {}
3973
3974 ////////////////////////////////////////////////////////////////////////////////
3975 // suspend/resume support
3976
3977 // the low-level signal-based suspend/resume support is a remnant from the
3978 // old VM-suspension that used to be for java-suspension, safepoints etc,
3979 // within hotspot. Now there is a single use-case for this:
3980 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3981 // that runs in the watcher thread.
3982 // The remaining code is greatly simplified from the more general suspension
3983 // code that used to be used.
3984 //
3985 // The protocol is quite simple:
3986 // - suspend:
3987 // - sends a signal to the target thread
3988 // - polls the suspend state of the osthread using a yield loop
3989 // - target thread signal handler (SR_handler) sets suspend state
3990 // and blocks in sigsuspend until continued
3991 // - resume:
3992 // - sets target osthread state to continue
3993 // - sends signal to end the sigsuspend loop in the SR_handler
3994 //
3995 // Note that the SR_lock plays no role in this suspend/resume protocol.
3996 //
3997
3998 static void resume_clear_context(OSThread *osthread) {
3999 osthread->set_ucontext(NULL);
4000 osthread->set_siginfo(NULL);
4001 }
4002
4003 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4004 osthread->set_ucontext(context);
4005 osthread->set_siginfo(siginfo);
4006 }
4007
4008 //
4009 // Handler function invoked when a thread's execution is suspended or
4010 // resumed. We have to be careful that only async-safe functions are
4011 // called here (Note: most pthread functions are not async safe and
4012 // should be avoided.)
4013 //
4014 // Note: sigwait() is a more natural fit than sigsuspend() from an
4015 // interface point of view, but sigwait() prevents the signal hander
4016 // from being run. libpthread would get very confused by not having
4017 // its signal handlers run and prevents sigwait()'s use with the
4018 // mutex granting granting signal.
4019 //
4020 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4021 //
4022 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4023 // Save and restore errno to avoid confusing native code with EINTR
4024 // after sigsuspend.
4025 int old_errno = errno;
4026
4027 Thread* thread = Thread::current();
4028 OSThread* osthread = thread->osthread();
4029 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4030
4031 os::SuspendResume::State current = osthread->sr.state();
4032 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4033 suspend_save_context(osthread, siginfo, context);
4034
4035 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4036 os::SuspendResume::State state = osthread->sr.suspended();
4037 if (state == os::SuspendResume::SR_SUSPENDED) {
4038 sigset_t suspend_set; // signals for sigsuspend()
4039
4040 // get current set of blocked signals and unblock resume signal
4041 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4042 sigdelset(&suspend_set, SR_signum);
4043
4044 sr_semaphore.signal();
4045 // wait here until we are resumed
4046 while (1) {
4047 sigsuspend(&suspend_set);
4048
4049 os::SuspendResume::State result = osthread->sr.running();
4050 if (result == os::SuspendResume::SR_RUNNING) {
4051 sr_semaphore.signal();
4052 break;
4053 }
4054 }
4055
4056 } else if (state == os::SuspendResume::SR_RUNNING) {
4057 // request was cancelled, continue
4058 } else {
4059 ShouldNotReachHere();
4060 }
4061
4062 resume_clear_context(osthread);
4063 } else if (current == os::SuspendResume::SR_RUNNING) {
4064 // request was cancelled, continue
4065 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4066 // ignore
4067 } else {
4068 // ignore
4069 }
4070
4071 errno = old_errno;
4072 }
4073
4074
4075 static int SR_initialize() {
4076 struct sigaction act;
4077 char *s;
4078 /* Get signal number to use for suspend/resume */
4079 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4080 int sig = ::strtol(s, 0, 10);
4081 if (sig > 0 || sig < _NSIG) {
4082 SR_signum = sig;
4083 }
4084 }
4085
4086 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4087 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4088
4089 sigemptyset(&SR_sigset);
4090 sigaddset(&SR_sigset, SR_signum);
4091
4092 /* Set up signal handler for suspend/resume */
4093 act.sa_flags = SA_RESTART|SA_SIGINFO;
4094 act.sa_handler = (void (*)(int)) SR_handler;
4095
4096 // SR_signum is blocked by default.
4097 // 4528190 - We also need to block pthread restart signal (32 on all
4098 // supported Linux platforms). Note that LinuxThreads need to block
4099 // this signal for all threads to work properly. So we don't have
4100 // to use hard-coded signal number when setting up the mask.
4101 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4102
4103 if (sigaction(SR_signum, &act, 0) == -1) {
4104 return -1;
4105 }
4106
4107 // Save signal flag
4108 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4109 return 0;
4110 }
4111
4112 static int sr_notify(OSThread* osthread) {
4113 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4114 assert_status(status == 0, status, "pthread_kill");
4115 return status;
4116 }
4117
4118 // "Randomly" selected value for how long we want to spin
4119 // before bailing out on suspending a thread, also how often
4120 // we send a signal to a thread we want to resume
4121 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4122 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4123
4124 // returns true on success and false on error - really an error is fatal
4125 // but this seems the normal response to library errors
4126 static bool do_suspend(OSThread* osthread) {
4127 assert(osthread->sr.is_running(), "thread should be running");
4128 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4129
4130 // mark as suspended and send signal
4131 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4132 // failed to switch, state wasn't running?
4133 ShouldNotReachHere();
4134 return false;
4135 }
4136
4137 if (sr_notify(osthread) != 0) {
4138 ShouldNotReachHere();
4139 }
4140
4141 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4142 while (true) {
4143 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4144 break;
4145 } else {
4146 // timeout
4147 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4148 if (cancelled == os::SuspendResume::SR_RUNNING) {
4149 return false;
4150 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4151 // make sure that we consume the signal on the semaphore as well
4152 sr_semaphore.wait();
4153 break;
4154 } else {
4155 ShouldNotReachHere();
4156 return false;
4157 }
4158 }
4159 }
4160
4161 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4162 return true;
4163 }
4164
4165 static void do_resume(OSThread* osthread) {
4166 assert(osthread->sr.is_suspended(), "thread should be suspended");
4167 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4168
4169 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4170 // failed to switch to WAKEUP_REQUEST
4171 ShouldNotReachHere();
4172 return;
4173 }
4174
4175 while (true) {
4176 if (sr_notify(osthread) == 0) {
4177 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4178 if (osthread->sr.is_running()) {
4179 return;
4180 }
4181 }
4182 } else {
4183 ShouldNotReachHere();
4184 }
4185 }
4186
4187 guarantee(osthread->sr.is_running(), "Must be running!");
4188 }
4189
4190 ////////////////////////////////////////////////////////////////////////////////
4191 // interrupt support
4192
4193 void os::interrupt(Thread* thread) {
4194 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4195 "possibility of dangling Thread pointer");
4196
4197 OSThread* osthread = thread->osthread();
4198
4199 if (!osthread->interrupted()) {
4200 osthread->set_interrupted(true);
4201 // More than one thread can get here with the same value of osthread,
4202 // resulting in multiple notifications. We do, however, want the store
4203 // to interrupted() to be visible to other threads before we execute unpark().
4204 OrderAccess::fence();
4205 ParkEvent * const slp = thread->_SleepEvent ;
4206 if (slp != NULL) slp->unpark() ;
4207 }
4208
4209 // For JSR166. Unpark even if interrupt status already was set
4210 if (thread->is_Java_thread())
4211 ((JavaThread*)thread)->parker()->unpark();
4212
4213 ParkEvent * ev = thread->_ParkEvent ;
4214 if (ev != NULL) ev->unpark() ;
4215
4216 }
4217
4218 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4219 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4220 "possibility of dangling Thread pointer");
4221
4222 OSThread* osthread = thread->osthread();
4223
4224 bool interrupted = osthread->interrupted();
4225
4226 if (interrupted && clear_interrupted) {
4227 osthread->set_interrupted(false);
4228 // consider thread->_SleepEvent->reset() ... optional optimization
4229 }
4230
4231 return interrupted;
4232 }
4233
4234 ///////////////////////////////////////////////////////////////////////////////////
4235 // signal handling (except suspend/resume)
4236
4237 // This routine may be used by user applications as a "hook" to catch signals.
4238 // The user-defined signal handler must pass unrecognized signals to this
4239 // routine, and if it returns true (non-zero), then the signal handler must
4240 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4241 // routine will never retun false (zero), but instead will execute a VM panic
4242 // routine kill the process.
4243 //
4244 // If this routine returns false, it is OK to call it again. This allows
4245 // the user-defined signal handler to perform checks either before or after
4246 // the VM performs its own checks. Naturally, the user code would be making
4247 // a serious error if it tried to handle an exception (such as a null check
4248 // or breakpoint) that the VM was generating for its own correct operation.
4249 //
4250 // This routine may recognize any of the following kinds of signals:
4251 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4252 // It should be consulted by handlers for any of those signals.
4253 //
4254 // The caller of this routine must pass in the three arguments supplied
4255 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4256 // field of the structure passed to sigaction(). This routine assumes that
4257 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4258 //
4259 // Note that the VM will print warnings if it detects conflicting signal
4260 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4261 //
4262 extern "C" JNIEXPORT int
4263 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4264 void* ucontext, int abort_if_unrecognized);
4265
4266 void signalHandler(int sig, siginfo_t* info, void* uc) {
4267 assert(info != NULL && uc != NULL, "it must be old kernel");
4268 int orig_errno = errno; // Preserve errno value over signal handler.
4269 JVM_handle_linux_signal(sig, info, uc, true);
4270 errno = orig_errno;
4271 }
4272
4273
4274 // This boolean allows users to forward their own non-matching signals
4275 // to JVM_handle_linux_signal, harmlessly.
4276 bool os::Linux::signal_handlers_are_installed = false;
4277
4278 // For signal-chaining
4279 struct sigaction os::Linux::sigact[MAXSIGNUM];
4280 unsigned int os::Linux::sigs = 0;
4281 bool os::Linux::libjsig_is_loaded = false;
4282 typedef struct sigaction *(*get_signal_t)(int);
4283 get_signal_t os::Linux::get_signal_action = NULL;
4284
4285 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4286 struct sigaction *actp = NULL;
4287
4288 if (libjsig_is_loaded) {
4289 // Retrieve the old signal handler from libjsig
4290 actp = (*get_signal_action)(sig);
4291 }
4292 if (actp == NULL) {
4293 // Retrieve the preinstalled signal handler from jvm
4294 actp = get_preinstalled_handler(sig);
4295 }
4296
4297 return actp;
4298 }
4299
4300 static bool call_chained_handler(struct sigaction *actp, int sig,
4301 siginfo_t *siginfo, void *context) {
4302 // Call the old signal handler
4303 if (actp->sa_handler == SIG_DFL) {
4304 // It's more reasonable to let jvm treat it as an unexpected exception
4305 // instead of taking the default action.
4306 return false;
4307 } else if (actp->sa_handler != SIG_IGN) {
4308 if ((actp->sa_flags & SA_NODEFER) == 0) {
4309 // automaticlly block the signal
4310 sigaddset(&(actp->sa_mask), sig);
4311 }
4312
4313 sa_handler_t hand;
4314 sa_sigaction_t sa;
4315 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4316 // retrieve the chained handler
4317 if (siginfo_flag_set) {
4318 sa = actp->sa_sigaction;
4319 } else {
4320 hand = actp->sa_handler;
4321 }
4322
4323 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4324 actp->sa_handler = SIG_DFL;
4325 }
4326
4327 // try to honor the signal mask
4328 sigset_t oset;
4329 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4330
4331 // call into the chained handler
4332 if (siginfo_flag_set) {
4333 (*sa)(sig, siginfo, context);
4334 } else {
4335 (*hand)(sig);
4336 }
4337
4338 // restore the signal mask
4339 pthread_sigmask(SIG_SETMASK, &oset, 0);
4340 }
4341 // Tell jvm's signal handler the signal is taken care of.
4342 return true;
4343 }
4344
4345 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4346 bool chained = false;
4347 // signal-chaining
4348 if (UseSignalChaining) {
4349 struct sigaction *actp = get_chained_signal_action(sig);
4350 if (actp != NULL) {
4351 chained = call_chained_handler(actp, sig, siginfo, context);
4352 }
4353 }
4354 return chained;
4355 }
4356
4357 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4358 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4359 return &sigact[sig];
4360 }
4361 return NULL;
4362 }
4363
4364 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4365 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4366 sigact[sig] = oldAct;
4367 sigs |= (unsigned int)1 << sig;
4368 }
4369
4370 // for diagnostic
4371 int os::Linux::sigflags[MAXSIGNUM];
4372
4373 int os::Linux::get_our_sigflags(int sig) {
4374 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4375 return sigflags[sig];
4376 }
4377
4378 void os::Linux::set_our_sigflags(int sig, int flags) {
4379 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4380 sigflags[sig] = flags;
4381 }
4382
4383 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4384 // Check for overwrite.
4385 struct sigaction oldAct;
4386 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4387
4388 void* oldhand = oldAct.sa_sigaction
4389 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4390 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4391 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4392 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4393 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4394 if (AllowUserSignalHandlers || !set_installed) {
4395 // Do not overwrite; user takes responsibility to forward to us.
4396 return;
4397 } else if (UseSignalChaining) {
4398 // save the old handler in jvm
4399 save_preinstalled_handler(sig, oldAct);
4400 // libjsig also interposes the sigaction() call below and saves the
4401 // old sigaction on it own.
4402 } else {
4403 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4404 "%#lx for signal %d.", (long)oldhand, sig));
4405 }
4406 }
4407
4408 struct sigaction sigAct;
4409 sigfillset(&(sigAct.sa_mask));
4410 sigAct.sa_handler = SIG_DFL;
4411 if (!set_installed) {
4412 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4413 } else {
4414 sigAct.sa_sigaction = signalHandler;
4415 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4416 }
4417 // Save flags, which are set by ours
4418 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4419 sigflags[sig] = sigAct.sa_flags;
4420
4421 int ret = sigaction(sig, &sigAct, &oldAct);
4422 assert(ret == 0, "check");
4423
4424 void* oldhand2 = oldAct.sa_sigaction
4425 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4426 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4427 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4428 }
4429
4430 // install signal handlers for signals that HotSpot needs to
4431 // handle in order to support Java-level exception handling.
4432
4433 void os::Linux::install_signal_handlers() {
4434 if (!signal_handlers_are_installed) {
4435 signal_handlers_are_installed = true;
4436
4437 // signal-chaining
4438 typedef void (*signal_setting_t)();
4439 signal_setting_t begin_signal_setting = NULL;
4440 signal_setting_t end_signal_setting = NULL;
4441 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4442 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4443 if (begin_signal_setting != NULL) {
4444 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4445 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4446 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4447 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4448 libjsig_is_loaded = true;
4449 assert(UseSignalChaining, "should enable signal-chaining");
4450 }
4451 if (libjsig_is_loaded) {
4452 // Tell libjsig jvm is setting signal handlers
4453 (*begin_signal_setting)();
4454 }
4455
4456 set_signal_handler(SIGSEGV, true);
4457 set_signal_handler(SIGPIPE, true);
4458 set_signal_handler(SIGBUS, true);
4459 set_signal_handler(SIGILL, true);
4460 set_signal_handler(SIGFPE, true);
4461 #if defined(PPC64)
4462 set_signal_handler(SIGTRAP, true);
4463 #endif
4464 set_signal_handler(SIGXFSZ, true);
4465
4466 if (libjsig_is_loaded) {
4467 // Tell libjsig jvm finishes setting signal handlers
4468 (*end_signal_setting)();
4469 }
4470
4471 // We don't activate signal checker if libjsig is in place, we trust ourselves
4472 // and if UserSignalHandler is installed all bets are off.
4473 // Log that signal checking is off only if -verbose:jni is specified.
4474 if (CheckJNICalls) {
4475 if (libjsig_is_loaded) {
4476 if (PrintJNIResolving) {
4477 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4478 }
4479 check_signals = false;
4480 }
4481 if (AllowUserSignalHandlers) {
4482 if (PrintJNIResolving) {
4483 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4484 }
4485 check_signals = false;
4486 }
4487 }
4488 }
4489 }
4490
4491 // This is the fastest way to get thread cpu time on Linux.
4492 // Returns cpu time (user+sys) for any thread, not only for current.
4493 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4494 // It might work on 2.6.10+ with a special kernel/glibc patch.
4495 // For reference, please, see IEEE Std 1003.1-2004:
4496 // http://www.unix.org/single_unix_specification
4497
4498 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4499 struct timespec tp;
4500 int rc = os::Linux::clock_gettime(clockid, &tp);
4501 assert(rc == 0, "clock_gettime is expected to return 0 code");
4502
4503 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4504 }
4505
4506 /////
4507 // glibc on Linux platform uses non-documented flag
4508 // to indicate, that some special sort of signal
4509 // trampoline is used.
4510 // We will never set this flag, and we should
4511 // ignore this flag in our diagnostic
4512 #ifdef SIGNIFICANT_SIGNAL_MASK
4513 #undef SIGNIFICANT_SIGNAL_MASK
4514 #endif
4515 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4516
4517 static const char* get_signal_handler_name(address handler,
4518 char* buf, int buflen) {
4519 int offset;
4520 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4521 if (found) {
4522 // skip directory names
4523 const char *p1, *p2;
4524 p1 = buf;
4525 size_t len = strlen(os::file_separator());
4526 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4527 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4528 } else {
4529 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4530 }
4531 return buf;
4532 }
4533
4534 static void print_signal_handler(outputStream* st, int sig,
4535 char* buf, size_t buflen) {
4536 struct sigaction sa;
4537
4538 sigaction(sig, NULL, &sa);
4539
4540 // See comment for SIGNIFICANT_SIGNAL_MASK define
4541 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4542
4543 st->print("%s: ", os::exception_name(sig, buf, buflen));
4544
4545 address handler = (sa.sa_flags & SA_SIGINFO)
4546 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4547 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4548
4549 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4550 st->print("SIG_DFL");
4551 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4552 st->print("SIG_IGN");
4553 } else {
4554 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4555 }
4556
4557 st->print(", sa_mask[0]=");
4558 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4559
4560 address rh = VMError::get_resetted_sighandler(sig);
4561 // May be, handler was resetted by VMError?
4562 if(rh != NULL) {
4563 handler = rh;
4564 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4565 }
4566
4567 st->print(", sa_flags=");
4568 os::Posix::print_sa_flags(st, sa.sa_flags);
4569
4570 // Check: is it our handler?
4571 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4572 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4573 // It is our signal handler
4574 // check for flags, reset system-used one!
4575 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4576 st->print(
4577 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4578 os::Linux::get_our_sigflags(sig));
4579 }
4580 }
4581 st->cr();
4582 }
4583
4584
4585 #define DO_SIGNAL_CHECK(sig) \
4586 if (!sigismember(&check_signal_done, sig)) \
4587 os::Linux::check_signal_handler(sig)
4588
4589 // This method is a periodic task to check for misbehaving JNI applications
4590 // under CheckJNI, we can add any periodic checks here
4591
4592 void os::run_periodic_checks() {
4593
4594 if (check_signals == false) return;
4595
4596 // SEGV and BUS if overridden could potentially prevent
4597 // generation of hs*.log in the event of a crash, debugging
4598 // such a case can be very challenging, so we absolutely
4599 // check the following for a good measure:
4600 DO_SIGNAL_CHECK(SIGSEGV);
4601 DO_SIGNAL_CHECK(SIGILL);
4602 DO_SIGNAL_CHECK(SIGFPE);
4603 DO_SIGNAL_CHECK(SIGBUS);
4604 DO_SIGNAL_CHECK(SIGPIPE);
4605 DO_SIGNAL_CHECK(SIGXFSZ);
4606 #if defined(PPC64)
4607 DO_SIGNAL_CHECK(SIGTRAP);
4608 #endif
4609
4610 // ReduceSignalUsage allows the user to override these handlers
4611 // see comments at the very top and jvm_solaris.h
4612 if (!ReduceSignalUsage) {
4613 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4614 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4615 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4616 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4617 }
4618
4619 DO_SIGNAL_CHECK(SR_signum);
4620 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4621 }
4622
4623 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4624
4625 static os_sigaction_t os_sigaction = NULL;
4626
4627 void os::Linux::check_signal_handler(int sig) {
4628 char buf[O_BUFLEN];
4629 address jvmHandler = NULL;
4630
4631
4632 struct sigaction act;
4633 if (os_sigaction == NULL) {
4634 // only trust the default sigaction, in case it has been interposed
4635 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4636 if (os_sigaction == NULL) return;
4637 }
4638
4639 os_sigaction(sig, (struct sigaction*)NULL, &act);
4640
4641
4642 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4643
4644 address thisHandler = (act.sa_flags & SA_SIGINFO)
4645 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4646 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4647
4648
4649 switch(sig) {
4650 case SIGSEGV:
4651 case SIGBUS:
4652 case SIGFPE:
4653 case SIGPIPE:
4654 case SIGILL:
4655 case SIGXFSZ:
4656 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4657 break;
4658
4659 case SHUTDOWN1_SIGNAL:
4660 case SHUTDOWN2_SIGNAL:
4661 case SHUTDOWN3_SIGNAL:
4662 case BREAK_SIGNAL:
4663 jvmHandler = (address)user_handler();
4664 break;
4665
4666 case INTERRUPT_SIGNAL:
4667 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4668 break;
4669
4670 default:
4671 if (sig == SR_signum) {
4672 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4673 } else {
4674 return;
4675 }
4676 break;
4677 }
4678
4679 if (thisHandler != jvmHandler) {
4680 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4681 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4682 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4683 // No need to check this sig any longer
4684 sigaddset(&check_signal_done, sig);
4685 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4686 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4687 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4688 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4689 // No need to check this sig any longer
4690 sigaddset(&check_signal_done, sig);
4691 }
4692
4693 // Dump all the signal
4694 if (sigismember(&check_signal_done, sig)) {
4695 print_signal_handlers(tty, buf, O_BUFLEN);
4696 }
4697 }
4698
4699 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4700
4701 extern bool signal_name(int signo, char* buf, size_t len);
4702
4703 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4704 if (0 < exception_code && exception_code <= SIGRTMAX) {
4705 // signal
4706 if (!signal_name(exception_code, buf, size)) {
4707 jio_snprintf(buf, size, "SIG%d", exception_code);
4708 }
4709 return buf;
4710 } else {
4711 return NULL;
4712 }
4713 }
4714
4715 // this is called _before_ the most of global arguments have been parsed
4716 void os::init(void) {
4717 char dummy; /* used to get a guess on initial stack address */
4718 // first_hrtime = gethrtime();
4719
4720 // With LinuxThreads the JavaMain thread pid (primordial thread)
4721 // is different than the pid of the java launcher thread.
4722 // So, on Linux, the launcher thread pid is passed to the VM
4723 // via the sun.java.launcher.pid property.
4724 // Use this property instead of getpid() if it was correctly passed.
4725 // See bug 6351349.
4726 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4727
4728 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4729
4730 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4731
4732 init_random(1234567);
4733
4734 ThreadCritical::initialize();
4735
4736 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4737 if (Linux::page_size() == -1) {
4738 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4739 strerror(errno)));
4740 }
4741 init_page_sizes((size_t) Linux::page_size());
4742
4743 Linux::initialize_system_info();
4744
4745 // main_thread points to the aboriginal thread
4746 Linux::_main_thread = pthread_self();
4747
4748 Linux::clock_init();
4749 initial_time_count = javaTimeNanos();
4750
4751 // pthread_condattr initialization for monotonic clock
4752 int status;
4753 pthread_condattr_t* _condattr = os::Linux::condAttr();
4754 if ((status = pthread_condattr_init(_condattr)) != 0) {
4755 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4756 }
4757 // Only set the clock if CLOCK_MONOTONIC is available
4758 if (Linux::supports_monotonic_clock()) {
4759 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4760 if (status == EINVAL) {
4761 warning("Unable to use monotonic clock with relative timed-waits" \
4762 " - changes to the time-of-day clock may have adverse affects");
4763 } else {
4764 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4765 }
4766 }
4767 }
4768 // else it defaults to CLOCK_REALTIME
4769
4770 pthread_mutex_init(&dl_mutex, NULL);
4771
4772 // If the pagesize of the VM is greater than 8K determine the appropriate
4773 // number of initial guard pages. The user can change this with the
4774 // command line arguments, if needed.
4775 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4776 StackYellowPages = 1;
4777 StackRedPages = 1;
4778 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4779 }
4780 }
4781
4782 // To install functions for atexit system call
4783 extern "C" {
4784 static void perfMemory_exit_helper() {
4785 perfMemory_exit();
4786 }
4787 }
4788
4789 // this is called _after_ the global arguments have been parsed
4790 jint os::init_2(void)
4791 {
4792 Linux::fast_thread_clock_init();
4793
4794 // Allocate a single page and mark it as readable for safepoint polling
4795 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4796 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4797
4798 os::set_polling_page( polling_page );
4799
4800 #ifndef PRODUCT
4801 if(Verbose && PrintMiscellaneous)
4802 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4803 #endif
4804
4805 if (!UseMembar) {
4806 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4807 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4808 os::set_memory_serialize_page( mem_serialize_page );
4809
4810 #ifndef PRODUCT
4811 if(Verbose && PrintMiscellaneous)
4812 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4813 #endif
4814 }
4815
4816 // initialize suspend/resume support - must do this before signal_sets_init()
4817 if (SR_initialize() != 0) {
4818 perror("SR_initialize failed");
4819 return JNI_ERR;
4820 }
4821
4822 Linux::signal_sets_init();
4823 Linux::install_signal_handlers();
4824
4825 // Check minimum allowable stack size for thread creation and to initialize
4826 // the java system classes, including StackOverflowError - depends on page
4827 // size. Add a page for compiler2 recursion in main thread.
4828 // Add in 2*BytesPerWord times page size to account for VM stack during
4829 // class initialization depending on 32 or 64 bit VM.
4830 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4831 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4832 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4833
4834 size_t threadStackSizeInBytes = ThreadStackSize * K;
4835 if (threadStackSizeInBytes != 0 &&
4836 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4837 tty->print_cr("\nThe stack size specified is too small, "
4838 "Specify at least %dk",
4839 os::Linux::min_stack_allowed/ K);
4840 return JNI_ERR;
4841 }
4842
4843 // Make the stack size a multiple of the page size so that
4844 // the yellow/red zones can be guarded.
4845 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4846 vm_page_size()));
4847
4848 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4849
4850 #if defined(IA32)
4851 workaround_expand_exec_shield_cs_limit();
4852 #endif
4853
4854 Linux::libpthread_init();
4855 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4856 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4857 Linux::glibc_version(), Linux::libpthread_version(),
4858 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4859 }
4860
4861 if (UseNUMA) {
4862 if (!Linux::libnuma_init()) {
4863 UseNUMA = false;
4864 } else {
4865 if ((Linux::numa_max_node() < 1)) {
4866 // There's only one node(they start from 0), disable NUMA.
4867 UseNUMA = false;
4868 }
4869 }
4870 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4871 // we can make the adaptive lgrp chunk resizing work. If the user specified
4872 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4873 // disable adaptive resizing.
4874 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4875 if (FLAG_IS_DEFAULT(UseNUMA)) {
4876 UseNUMA = false;
4877 } else {
4878 if (FLAG_IS_DEFAULT(UseLargePages) &&
4879 FLAG_IS_DEFAULT(UseSHM) &&
4880 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4881 UseLargePages = false;
4882 } else {
4883 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4884 UseAdaptiveSizePolicy = false;
4885 UseAdaptiveNUMAChunkSizing = false;
4886 }
4887 }
4888 }
4889 if (!UseNUMA && ForceNUMA) {
4890 UseNUMA = true;
4891 }
4892 }
4893
4894 if (MaxFDLimit) {
4895 // set the number of file descriptors to max. print out error
4896 // if getrlimit/setrlimit fails but continue regardless.
4897 struct rlimit nbr_files;
4898 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4899 if (status != 0) {
4900 if (PrintMiscellaneous && (Verbose || WizardMode))
4901 perror("os::init_2 getrlimit failed");
4902 } else {
4903 nbr_files.rlim_cur = nbr_files.rlim_max;
4904 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4905 if (status != 0) {
4906 if (PrintMiscellaneous && (Verbose || WizardMode))
4907 perror("os::init_2 setrlimit failed");
4908 }
4909 }
4910 }
4911
4912 // Initialize lock used to serialize thread creation (see os::create_thread)
4913 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4914
4915 // at-exit methods are called in the reverse order of their registration.
4916 // atexit functions are called on return from main or as a result of a
4917 // call to exit(3C). There can be only 32 of these functions registered
4918 // and atexit() does not set errno.
4919
4920 if (PerfAllowAtExitRegistration) {
4921 // only register atexit functions if PerfAllowAtExitRegistration is set.
4922 // atexit functions can be delayed until process exit time, which
4923 // can be problematic for embedded VM situations. Embedded VMs should
4924 // call DestroyJavaVM() to assure that VM resources are released.
4925
4926 // note: perfMemory_exit_helper atexit function may be removed in
4927 // the future if the appropriate cleanup code can be added to the
4928 // VM_Exit VMOperation's doit method.
4929 if (atexit(perfMemory_exit_helper) != 0) {
4930 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4931 }
4932 }
4933
4934 // initialize thread priority policy
4935 prio_init();
4936
4937 return JNI_OK;
4938 }
4939
4940 // this is called at the end of vm_initialization
4941 void os::init_3(void) {
4942 #ifdef JAVASE_EMBEDDED
4943 // Start the MemNotifyThread
4944 if (LowMemoryProtection) {
4945 MemNotifyThread::start();
4946 }
4947 return;
4948 #endif
4949 }
4950
4951 // Mark the polling page as unreadable
4952 void os::make_polling_page_unreadable(void) {
4953 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4954 fatal("Could not disable polling page");
4955 };
4956
4957 // Mark the polling page as readable
4958 void os::make_polling_page_readable(void) {
4959 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4960 fatal("Could not enable polling page");
4961 }
4962 };
4963
4964 int os::active_processor_count() {
4965 // Linux doesn't yet have a (official) notion of processor sets,
4966 // so just return the number of online processors.
4967 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4968 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4969 return online_cpus;
4970 }
4971
4972 void os::set_native_thread_name(const char *name) {
4973 // Not yet implemented.
4974 return;
4975 }
4976
4977 bool os::distribute_processes(uint length, uint* distribution) {
4978 // Not yet implemented.
4979 return false;
4980 }
4981
4982 bool os::bind_to_processor(uint processor_id) {
4983 // Not yet implemented.
4984 return false;
4985 }
4986
4987 ///
4988
4989 void os::SuspendedThreadTask::internal_do_task() {
4990 if (do_suspend(_thread->osthread())) {
4991 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4992 do_task(context);
4993 do_resume(_thread->osthread());
4994 }
4995 }
4996
4997 class PcFetcher : public os::SuspendedThreadTask {
4998 public:
4999 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5000 ExtendedPC result();
5001 protected:
5002 void do_task(const os::SuspendedThreadTaskContext& context);
5003 private:
5004 ExtendedPC _epc;
5005 };
5006
5007 ExtendedPC PcFetcher::result() {
5008 guarantee(is_done(), "task is not done yet.");
5009 return _epc;
5010 }
5011
5012 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5013 Thread* thread = context.thread();
5014 OSThread* osthread = thread->osthread();
5015 if (osthread->ucontext() != NULL) {
5016 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5017 } else {
5018 // NULL context is unexpected, double-check this is the VMThread
5019 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5020 }
5021 }
5022
5023 // Suspends the target using the signal mechanism and then grabs the PC before
5024 // resuming the target. Used by the flat-profiler only
5025 ExtendedPC os::get_thread_pc(Thread* thread) {
5026 // Make sure that it is called by the watcher for the VMThread
5027 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5028 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5029
5030 PcFetcher fetcher(thread);
5031 fetcher.run();
5032 return fetcher.result();
5033 }
5034
5035 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5036 {
5037 if (is_NPTL()) {
5038 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5039 } else {
5040 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5041 // word back to default 64bit precision if condvar is signaled. Java
5042 // wants 53bit precision. Save and restore current value.
5043 int fpu = get_fpu_control_word();
5044 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5045 set_fpu_control_word(fpu);
5046 return status;
5047 }
5048 }
5049
5050 ////////////////////////////////////////////////////////////////////////////////
5051 // debug support
5052
5053 bool os::find(address addr, outputStream* st) {
5054 Dl_info dlinfo;
5055 memset(&dlinfo, 0, sizeof(dlinfo));
5056 if (dladdr(addr, &dlinfo) != 0) {
5057 st->print(PTR_FORMAT ": ", addr);
5058 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5059 st->print("%s+%#x", dlinfo.dli_sname,
5060 addr - (intptr_t)dlinfo.dli_saddr);
5061 } else if (dlinfo.dli_fbase != NULL) {
5062 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5063 } else {
5064 st->print("<absolute address>");
5065 }
5066 if (dlinfo.dli_fname != NULL) {
5067 st->print(" in %s", dlinfo.dli_fname);
5068 }
5069 if (dlinfo.dli_fbase != NULL) {
5070 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5071 }
5072 st->cr();
5073
5074 if (Verbose) {
5075 // decode some bytes around the PC
5076 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5077 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5078 address lowest = (address) dlinfo.dli_sname;
5079 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5080 if (begin < lowest) begin = lowest;
5081 Dl_info dlinfo2;
5082 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5083 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5084 end = (address) dlinfo2.dli_saddr;
5085 Disassembler::decode(begin, end, st);
5086 }
5087 return true;
5088 }
5089 return false;
5090 }
5091
5092 ////////////////////////////////////////////////////////////////////////////////
5093 // misc
5094
5095 // This does not do anything on Linux. This is basically a hook for being
5096 // able to use structured exception handling (thread-local exception filters)
5097 // on, e.g., Win32.
5098 void
5099 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5100 JavaCallArguments* args, Thread* thread) {
5101 f(value, method, args, thread);
5102 }
5103
5104 void os::print_statistics() {
5105 }
5106
5107 int os::message_box(const char* title, const char* message) {
5108 int i;
5109 fdStream err(defaultStream::error_fd());
5110 for (i = 0; i < 78; i++) err.print_raw("=");
5111 err.cr();
5112 err.print_raw_cr(title);
5113 for (i = 0; i < 78; i++) err.print_raw("-");
5114 err.cr();
5115 err.print_raw_cr(message);
5116 for (i = 0; i < 78; i++) err.print_raw("=");
5117 err.cr();
5118
5119 char buf[16];
5120 // Prevent process from exiting upon "read error" without consuming all CPU
5121 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5122
5123 return buf[0] == 'y' || buf[0] == 'Y';
5124 }
5125
5126 int os::stat(const char *path, struct stat *sbuf) {
5127 char pathbuf[MAX_PATH];
5128 if (strlen(path) > MAX_PATH - 1) {
5129 errno = ENAMETOOLONG;
5130 return -1;
5131 }
5132 os::native_path(strcpy(pathbuf, path));
5133 return ::stat(pathbuf, sbuf);
5134 }
5135
5136 bool os::check_heap(bool force) {
5137 return true;
5138 }
5139
5140 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5141 return ::vsnprintf(buf, count, format, args);
5142 }
5143
5144 // Is a (classpath) directory empty?
5145 bool os::dir_is_empty(const char* path) {
5146 DIR *dir = NULL;
5147 struct dirent *ptr;
5148
5149 dir = opendir(path);
5150 if (dir == NULL) return true;
5151
5152 /* Scan the directory */
5153 bool result = true;
5154 char buf[sizeof(struct dirent) + MAX_PATH];
5155 while (result && (ptr = ::readdir(dir)) != NULL) {
5156 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5157 result = false;
5158 }
5159 }
5160 closedir(dir);
5161 return result;
5162 }
5163
5164 // This code originates from JDK's sysOpen and open64_w
5165 // from src/solaris/hpi/src/system_md.c
5166
5167 #ifndef O_DELETE
5168 #define O_DELETE 0x10000
5169 #endif
5170
5171 // Open a file. Unlink the file immediately after open returns
5172 // if the specified oflag has the O_DELETE flag set.
5173 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5174
5175 int os::open(const char *path, int oflag, int mode) {
5176
5177 if (strlen(path) > MAX_PATH - 1) {
5178 errno = ENAMETOOLONG;
5179 return -1;
5180 }
5181 int fd;
5182 int o_delete = (oflag & O_DELETE);
5183 oflag = oflag & ~O_DELETE;
5184
5185 fd = ::open64(path, oflag, mode);
5186 if (fd == -1) return -1;
5187
5188 //If the open succeeded, the file might still be a directory
5189 {
5190 struct stat64 buf64;
5191 int ret = ::fstat64(fd, &buf64);
5192 int st_mode = buf64.st_mode;
5193
5194 if (ret != -1) {
5195 if ((st_mode & S_IFMT) == S_IFDIR) {
5196 errno = EISDIR;
5197 ::close(fd);
5198 return -1;
5199 }
5200 } else {
5201 ::close(fd);
5202 return -1;
5203 }
5204 }
5205
5206 /*
5207 * All file descriptors that are opened in the JVM and not
5208 * specifically destined for a subprocess should have the
5209 * close-on-exec flag set. If we don't set it, then careless 3rd
5210 * party native code might fork and exec without closing all
5211 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5212 * UNIXProcess.c), and this in turn might:
5213 *
5214 * - cause end-of-file to fail to be detected on some file
5215 * descriptors, resulting in mysterious hangs, or
5216 *
5217 * - might cause an fopen in the subprocess to fail on a system
5218 * suffering from bug 1085341.
5219 *
5220 * (Yes, the default setting of the close-on-exec flag is a Unix
5221 * design flaw)
5222 *
5223 * See:
5224 * 1085341: 32-bit stdio routines should support file descriptors >255
5225 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5226 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5227 */
5228 #ifdef FD_CLOEXEC
5229 {
5230 int flags = ::fcntl(fd, F_GETFD);
5231 if (flags != -1)
5232 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5233 }
5234 #endif
5235
5236 if (o_delete != 0) {
5237 ::unlink(path);
5238 }
5239 return fd;
5240 }
5241
5242
5243 // create binary file, rewriting existing file if required
5244 int os::create_binary_file(const char* path, bool rewrite_existing) {
5245 int oflags = O_WRONLY | O_CREAT;
5246 if (!rewrite_existing) {
5247 oflags |= O_EXCL;
5248 }
5249 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5250 }
5251
5252 // return current position of file pointer
5253 jlong os::current_file_offset(int fd) {
5254 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5255 }
5256
5257 // move file pointer to the specified offset
5258 jlong os::seek_to_file_offset(int fd, jlong offset) {
5259 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5260 }
5261
5262 // This code originates from JDK's sysAvailable
5263 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5264
5265 int os::available(int fd, jlong *bytes) {
5266 jlong cur, end;
5267 int mode;
5268 struct stat64 buf64;
5269
5270 if (::fstat64(fd, &buf64) >= 0) {
5271 mode = buf64.st_mode;
5272 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5273 /*
5274 * XXX: is the following call interruptible? If so, this might
5275 * need to go through the INTERRUPT_IO() wrapper as for other
5276 * blocking, interruptible calls in this file.
5277 */
5278 int n;
5279 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5280 *bytes = n;
5281 return 1;
5282 }
5283 }
5284 }
5285 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5286 return 0;
5287 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5288 return 0;
5289 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5290 return 0;
5291 }
5292 *bytes = end - cur;
5293 return 1;
5294 }
5295
5296 int os::socket_available(int fd, jint *pbytes) {
5297 // Linux doc says EINTR not returned, unlike Solaris
5298 int ret = ::ioctl(fd, FIONREAD, pbytes);
5299
5300 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5301 // is expected to return 0 on failure and 1 on success to the jdk.
5302 return (ret < 0) ? 0 : 1;
5303 }
5304
5305 // Map a block of memory.
5306 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5307 char *addr, size_t bytes, bool read_only,
5308 bool allow_exec) {
5309 int prot;
5310 int flags = MAP_PRIVATE;
5311
5312 if (read_only) {
5313 prot = PROT_READ;
5314 } else {
5315 prot = PROT_READ | PROT_WRITE;
5316 }
5317
5318 if (allow_exec) {
5319 prot |= PROT_EXEC;
5320 }
5321
5322 if (addr != NULL) {
5323 flags |= MAP_FIXED;
5324 }
5325
5326 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5327 fd, file_offset);
5328 if (mapped_address == MAP_FAILED) {
5329 return NULL;
5330 }
5331 return mapped_address;
5332 }
5333
5334
5335 // Remap a block of memory.
5336 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5337 char *addr, size_t bytes, bool read_only,
5338 bool allow_exec) {
5339 // same as map_memory() on this OS
5340 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5341 allow_exec);
5342 }
5343
5344
5345 // Unmap a block of memory.
5346 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5347 return munmap(addr, bytes) == 0;
5348 }
5349
5350 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5351
5352 static clockid_t thread_cpu_clockid(Thread* thread) {
5353 pthread_t tid = thread->osthread()->pthread_id();
5354 clockid_t clockid;
5355
5356 // Get thread clockid
5357 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5358 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5359 return clockid;
5360 }
5361
5362 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5363 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5364 // of a thread.
5365 //
5366 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5367 // the fast estimate available on the platform.
5368
5369 jlong os::current_thread_cpu_time() {
5370 if (os::Linux::supports_fast_thread_cpu_time()) {
5371 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5372 } else {
5373 // return user + sys since the cost is the same
5374 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5375 }
5376 }
5377
5378 jlong os::thread_cpu_time(Thread* thread) {
5379 // consistent with what current_thread_cpu_time() returns
5380 if (os::Linux::supports_fast_thread_cpu_time()) {
5381 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5382 } else {
5383 return slow_thread_cpu_time(thread, true /* user + sys */);
5384 }
5385 }
5386
5387 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5388 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5389 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5390 } else {
5391 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5392 }
5393 }
5394
5395 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5396 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5397 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5398 } else {
5399 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5400 }
5401 }
5402
5403 //
5404 // -1 on error.
5405 //
5406
5407 PRAGMA_DIAG_PUSH
5408 PRAGMA_FORMAT_NONLITERAL_IGNORED
5409 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5410 static bool proc_task_unchecked = true;
5411 static const char *proc_stat_path = "/proc/%d/stat";
5412 pid_t tid = thread->osthread()->thread_id();
5413 char *s;
5414 char stat[2048];
5415 int statlen;
5416 char proc_name[64];
5417 int count;
5418 long sys_time, user_time;
5419 char cdummy;
5420 int idummy;
5421 long ldummy;
5422 FILE *fp;
5423
5424 // The /proc/<tid>/stat aggregates per-process usage on
5425 // new Linux kernels 2.6+ where NPTL is supported.
5426 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5427 // See bug 6328462.
5428 // There possibly can be cases where there is no directory
5429 // /proc/self/task, so we check its availability.
5430 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5431 // This is executed only once
5432 proc_task_unchecked = false;
5433 fp = fopen("/proc/self/task", "r");
5434 if (fp != NULL) {
5435 proc_stat_path = "/proc/self/task/%d/stat";
5436 fclose(fp);
5437 }
5438 }
5439
5440 sprintf(proc_name, proc_stat_path, tid);
5441 fp = fopen(proc_name, "r");
5442 if ( fp == NULL ) return -1;
5443 statlen = fread(stat, 1, 2047, fp);
5444 stat[statlen] = '\0';
5445 fclose(fp);
5446
5447 // Skip pid and the command string. Note that we could be dealing with
5448 // weird command names, e.g. user could decide to rename java launcher
5449 // to "java 1.4.2 :)", then the stat file would look like
5450 // 1234 (java 1.4.2 :)) R ... ...
5451 // We don't really need to know the command string, just find the last
5452 // occurrence of ")" and then start parsing from there. See bug 4726580.
5453 s = strrchr(stat, ')');
5454 if (s == NULL ) return -1;
5455
5456 // Skip blank chars
5457 do s++; while (isspace(*s));
5458
5459 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5460 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5461 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5462 &user_time, &sys_time);
5463 if ( count != 13 ) return -1;
5464 if (user_sys_cpu_time) {
5465 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5466 } else {
5467 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5468 }
5469 }
5470 PRAGMA_DIAG_POP
5471
5472 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5473 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5474 info_ptr->may_skip_backward = false; // elapsed time not wall time
5475 info_ptr->may_skip_forward = false; // elapsed time not wall time
5476 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5477 }
5478
5479 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5480 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5481 info_ptr->may_skip_backward = false; // elapsed time not wall time
5482 info_ptr->may_skip_forward = false; // elapsed time not wall time
5483 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5484 }
5485
5486 bool os::is_thread_cpu_time_supported() {
5487 return true;
5488 }
5489
5490 // System loadavg support. Returns -1 if load average cannot be obtained.
5491 // Linux doesn't yet have a (official) notion of processor sets,
5492 // so just return the system wide load average.
5493 int os::loadavg(double loadavg[], int nelem) {
5494 return ::getloadavg(loadavg, nelem);
5495 }
5496
5497 void os::pause() {
5498 char filename[MAX_PATH];
5499 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5500 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5501 } else {
5502 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5503 }
5504
5505 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5506 if (fd != -1) {
5507 struct stat buf;
5508 ::close(fd);
5509 while (::stat(filename, &buf) == 0) {
5510 (void)::poll(NULL, 0, 100);
5511 }
5512 } else {
5513 jio_fprintf(stderr,
5514 "Could not open pause file '%s', continuing immediately.\n", filename);
5515 }
5516 }
5517
5518
5519 // Refer to the comments in os_solaris.cpp park-unpark.
5520 //
5521 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5522 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5523 // For specifics regarding the bug see GLIBC BUGID 261237 :
5524 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5525 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5526 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5527 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5528 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5529 // and monitorenter when we're using 1-0 locking. All those operations may result in
5530 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5531 // of libpthread avoids the problem, but isn't practical.
5532 //
5533 // Possible remedies:
5534 //
5535 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5536 // This is palliative and probabilistic, however. If the thread is preempted
5537 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5538 // than the minimum period may have passed, and the abstime may be stale (in the
5539 // past) resultin in a hang. Using this technique reduces the odds of a hang
5540 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5541 //
5542 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5543 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5544 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5545 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5546 // thread.
5547 //
5548 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5549 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5550 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5551 // This also works well. In fact it avoids kernel-level scalability impediments
5552 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5553 // timers in a graceful fashion.
5554 //
5555 // 4. When the abstime value is in the past it appears that control returns
5556 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5557 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5558 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5559 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5560 // It may be possible to avoid reinitialization by checking the return
5561 // value from pthread_cond_timedwait(). In addition to reinitializing the
5562 // condvar we must establish the invariant that cond_signal() is only called
5563 // within critical sections protected by the adjunct mutex. This prevents
5564 // cond_signal() from "seeing" a condvar that's in the midst of being
5565 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5566 // desirable signal-after-unlock optimization that avoids futile context switching.
5567 //
5568 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5569 // structure when a condvar is used or initialized. cond_destroy() would
5570 // release the helper structure. Our reinitialize-after-timedwait fix
5571 // put excessive stress on malloc/free and locks protecting the c-heap.
5572 //
5573 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5574 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5575 // and only enabling the work-around for vulnerable environments.
5576
5577 // utility to compute the abstime argument to timedwait:
5578 // millis is the relative timeout time
5579 // abstime will be the absolute timeout time
5580 // TODO: replace compute_abstime() with unpackTime()
5581
5582 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5583 if (millis < 0) millis = 0;
5584
5585 jlong seconds = millis / 1000;
5586 millis %= 1000;
5587 if (seconds > 50000000) { // see man cond_timedwait(3T)
5588 seconds = 50000000;
5589 }
5590
5591 if (os::Linux::supports_monotonic_clock()) {
5592 struct timespec now;
5593 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5594 assert_status(status == 0, status, "clock_gettime");
5595 abstime->tv_sec = now.tv_sec + seconds;
5596 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5597 if (nanos >= NANOSECS_PER_SEC) {
5598 abstime->tv_sec += 1;
5599 nanos -= NANOSECS_PER_SEC;
5600 }
5601 abstime->tv_nsec = nanos;
5602 } else {
5603 struct timeval now;
5604 int status = gettimeofday(&now, NULL);
5605 assert(status == 0, "gettimeofday");
5606 abstime->tv_sec = now.tv_sec + seconds;
5607 long usec = now.tv_usec + millis * 1000;
5608 if (usec >= 1000000) {
5609 abstime->tv_sec += 1;
5610 usec -= 1000000;
5611 }
5612 abstime->tv_nsec = usec * 1000;
5613 }
5614 return abstime;
5615 }
5616
5617
5618 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5619 // Conceptually TryPark() should be equivalent to park(0).
5620
5621 int os::PlatformEvent::TryPark() {
5622 for (;;) {
5623 const int v = _Event ;
5624 guarantee ((v == 0) || (v == 1), "invariant") ;
5625 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5626 }
5627 }
5628
5629 void os::PlatformEvent::park() { // AKA "down()"
5630 // Invariant: Only the thread associated with the Event/PlatformEvent
5631 // may call park().
5632 // TODO: assert that _Assoc != NULL or _Assoc == Self
5633 int v ;
5634 for (;;) {
5635 v = _Event ;
5636 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5637 }
5638 guarantee (v >= 0, "invariant") ;
5639 if (v == 0) {
5640 // Do this the hard way by blocking ...
5641 int status = pthread_mutex_lock(_mutex);
5642 assert_status(status == 0, status, "mutex_lock");
5643 guarantee (_nParked == 0, "invariant") ;
5644 ++ _nParked ;
5645 while (_Event < 0) {
5646 status = pthread_cond_wait(_cond, _mutex);
5647 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5648 // Treat this the same as if the wait was interrupted
5649 if (status == ETIME) { status = EINTR; }
5650 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5651 }
5652 -- _nParked ;
5653
5654 _Event = 0 ;
5655 status = pthread_mutex_unlock(_mutex);
5656 assert_status(status == 0, status, "mutex_unlock");
5657 // Paranoia to ensure our locked and lock-free paths interact
5658 // correctly with each other.
5659 OrderAccess::fence();
5660 }
5661 guarantee (_Event >= 0, "invariant") ;
5662 }
5663
5664 int os::PlatformEvent::park(jlong millis) {
5665 guarantee (_nParked == 0, "invariant") ;
5666
5667 int v ;
5668 for (;;) {
5669 v = _Event ;
5670 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5671 }
5672 guarantee (v >= 0, "invariant") ;
5673 if (v != 0) return OS_OK ;
5674
5675 // We do this the hard way, by blocking the thread.
5676 // Consider enforcing a minimum timeout value.
5677 struct timespec abst;
5678 compute_abstime(&abst, millis);
5679
5680 int ret = OS_TIMEOUT;
5681 int status = pthread_mutex_lock(_mutex);
5682 assert_status(status == 0, status, "mutex_lock");
5683 guarantee (_nParked == 0, "invariant") ;
5684 ++_nParked ;
5685
5686 // Object.wait(timo) will return because of
5687 // (a) notification
5688 // (b) timeout
5689 // (c) thread.interrupt
5690 //
5691 // Thread.interrupt and object.notify{All} both call Event::set.
5692 // That is, we treat thread.interrupt as a special case of notification.
5693 // The underlying Solaris implementation, cond_timedwait, admits
5694 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5695 // JVM from making those visible to Java code. As such, we must
5696 // filter out spurious wakeups. We assume all ETIME returns are valid.
5697 //
5698 // TODO: properly differentiate simultaneous notify+interrupt.
5699 // In that case, we should propagate the notify to another waiter.
5700
5701 while (_Event < 0) {
5702 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5703 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5704 pthread_cond_destroy (_cond);
5705 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5706 }
5707 assert_status(status == 0 || status == EINTR ||
5708 status == ETIME || status == ETIMEDOUT,
5709 status, "cond_timedwait");
5710 if (!FilterSpuriousWakeups) break ; // previous semantics
5711 if (status == ETIME || status == ETIMEDOUT) break ;
5712 // We consume and ignore EINTR and spurious wakeups.
5713 }
5714 --_nParked ;
5715 if (_Event >= 0) {
5716 ret = OS_OK;
5717 }
5718 _Event = 0 ;
5719 status = pthread_mutex_unlock(_mutex);
5720 assert_status(status == 0, status, "mutex_unlock");
5721 assert (_nParked == 0, "invariant") ;
5722 // Paranoia to ensure our locked and lock-free paths interact
5723 // correctly with each other.
5724 OrderAccess::fence();
5725 return ret;
5726 }
5727
5728 void os::PlatformEvent::unpark() {
5729 // Transitions for _Event:
5730 // 0 :=> 1
5731 // 1 :=> 1
5732 // -1 :=> either 0 or 1; must signal target thread
5733 // That is, we can safely transition _Event from -1 to either
5734 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5735 // unpark() calls.
5736 // See also: "Semaphores in Plan 9" by Mullender & Cox
5737 //
5738 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5739 // that it will take two back-to-back park() calls for the owning
5740 // thread to block. This has the benefit of forcing a spurious return
5741 // from the first park() call after an unpark() call which will help
5742 // shake out uses of park() and unpark() without condition variables.
5743
5744 if (Atomic::xchg(1, &_Event) >= 0) return;
5745
5746 // Wait for the thread associated with the event to vacate
5747 int status = pthread_mutex_lock(_mutex);
5748 assert_status(status == 0, status, "mutex_lock");
5749 int AnyWaiters = _nParked;
5750 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5751 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5752 AnyWaiters = 0;
5753 pthread_cond_signal(_cond);
5754 }
5755 status = pthread_mutex_unlock(_mutex);
5756 assert_status(status == 0, status, "mutex_unlock");
5757 if (AnyWaiters != 0) {
5758 status = pthread_cond_signal(_cond);
5759 assert_status(status == 0, status, "cond_signal");
5760 }
5761
5762 // Note that we signal() _after dropping the lock for "immortal" Events.
5763 // This is safe and avoids a common class of futile wakeups. In rare
5764 // circumstances this can cause a thread to return prematurely from
5765 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5766 // simply re-test the condition and re-park itself.
5767 }
5768
5769
5770 // JSR166
5771 // -------------------------------------------------------
5772
5773 /*
5774 * The solaris and linux implementations of park/unpark are fairly
5775 * conservative for now, but can be improved. They currently use a
5776 * mutex/condvar pair, plus a a count.
5777 * Park decrements count if > 0, else does a condvar wait. Unpark
5778 * sets count to 1 and signals condvar. Only one thread ever waits
5779 * on the condvar. Contention seen when trying to park implies that someone
5780 * is unparking you, so don't wait. And spurious returns are fine, so there
5781 * is no need to track notifications.
5782 */
5783
5784 /*
5785 * This code is common to linux and solaris and will be moved to a
5786 * common place in dolphin.
5787 *
5788 * The passed in time value is either a relative time in nanoseconds
5789 * or an absolute time in milliseconds. Either way it has to be unpacked
5790 * into suitable seconds and nanoseconds components and stored in the
5791 * given timespec structure.
5792 * Given time is a 64-bit value and the time_t used in the timespec is only
5793 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5794 * overflow if times way in the future are given. Further on Solaris versions
5795 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5796 * number of seconds, in abstime, is less than current_time + 100,000,000.
5797 * As it will be 28 years before "now + 100000000" will overflow we can
5798 * ignore overflow and just impose a hard-limit on seconds using the value
5799 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5800 * years from "now".
5801 */
5802
5803 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5804 assert (time > 0, "convertTime");
5805 time_t max_secs = 0;
5806
5807 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5808 struct timeval now;
5809 int status = gettimeofday(&now, NULL);
5810 assert(status == 0, "gettimeofday");
5811
5812 max_secs = now.tv_sec + MAX_SECS;
5813
5814 if (isAbsolute) {
5815 jlong secs = time / 1000;
5816 if (secs > max_secs) {
5817 absTime->tv_sec = max_secs;
5818 } else {
5819 absTime->tv_sec = secs;
5820 }
5821 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5822 } else {
5823 jlong secs = time / NANOSECS_PER_SEC;
5824 if (secs >= MAX_SECS) {
5825 absTime->tv_sec = max_secs;
5826 absTime->tv_nsec = 0;
5827 } else {
5828 absTime->tv_sec = now.tv_sec + secs;
5829 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5830 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5831 absTime->tv_nsec -= NANOSECS_PER_SEC;
5832 ++absTime->tv_sec; // note: this must be <= max_secs
5833 }
5834 }
5835 }
5836 } else {
5837 // must be relative using monotonic clock
5838 struct timespec now;
5839 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5840 assert_status(status == 0, status, "clock_gettime");
5841 max_secs = now.tv_sec + MAX_SECS;
5842 jlong secs = time / NANOSECS_PER_SEC;
5843 if (secs >= MAX_SECS) {
5844 absTime->tv_sec = max_secs;
5845 absTime->tv_nsec = 0;
5846 } else {
5847 absTime->tv_sec = now.tv_sec + secs;
5848 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5849 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5850 absTime->tv_nsec -= NANOSECS_PER_SEC;
5851 ++absTime->tv_sec; // note: this must be <= max_secs
5852 }
5853 }
5854 }
5855 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5856 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5857 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5858 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5859 }
5860
5861 void Parker::park(bool isAbsolute, jlong time) {
5862 // Ideally we'd do something useful while spinning, such
5863 // as calling unpackTime().
5864
5865 // Optional fast-path check:
5866 // Return immediately if a permit is available.
5867 // We depend on Atomic::xchg() having full barrier semantics
5868 // since we are doing a lock-free update to _counter.
5869 if (Atomic::xchg(0, &_counter) > 0) return;
5870
5871 Thread* thread = Thread::current();
5872 assert(thread->is_Java_thread(), "Must be JavaThread");
5873 JavaThread *jt = (JavaThread *)thread;
5874
5875 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5876 // Check interrupt before trying to wait
5877 if (Thread::is_interrupted(thread, false)) {
5878 return;
5879 }
5880
5881 // Next, demultiplex/decode time arguments
5882 timespec absTime;
5883 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5884 return;
5885 }
5886 if (time > 0) {
5887 unpackTime(&absTime, isAbsolute, time);
5888 }
5889
5890
5891 // Enter safepoint region
5892 // Beware of deadlocks such as 6317397.
5893 // The per-thread Parker:: mutex is a classic leaf-lock.
5894 // In particular a thread must never block on the Threads_lock while
5895 // holding the Parker:: mutex. If safepoints are pending both the
5896 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5897 ThreadBlockInVM tbivm(jt);
5898
5899 // Don't wait if cannot get lock since interference arises from
5900 // unblocking. Also. check interrupt before trying wait
5901 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5902 return;
5903 }
5904
5905 int status ;
5906 if (_counter > 0) { // no wait needed
5907 _counter = 0;
5908 status = pthread_mutex_unlock(_mutex);
5909 assert (status == 0, "invariant") ;
5910 // Paranoia to ensure our locked and lock-free paths interact
5911 // correctly with each other and Java-level accesses.
5912 OrderAccess::fence();
5913 return;
5914 }
5915
5916 #ifdef ASSERT
5917 // Don't catch signals while blocked; let the running threads have the signals.
5918 // (This allows a debugger to break into the running thread.)
5919 sigset_t oldsigs;
5920 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5921 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5922 #endif
5923
5924 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5925 jt->set_suspend_equivalent();
5926 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5927
5928 assert(_cur_index == -1, "invariant");
5929 if (time == 0) {
5930 _cur_index = REL_INDEX; // arbitrary choice when not timed
5931 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5932 } else {
5933 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5934 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5935 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5936 pthread_cond_destroy (&_cond[_cur_index]) ;
5937 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5938 }
5939 }
5940 _cur_index = -1;
5941 assert_status(status == 0 || status == EINTR ||
5942 status == ETIME || status == ETIMEDOUT,
5943 status, "cond_timedwait");
5944
5945 #ifdef ASSERT
5946 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5947 #endif
5948
5949 _counter = 0 ;
5950 status = pthread_mutex_unlock(_mutex) ;
5951 assert_status(status == 0, status, "invariant") ;
5952 // Paranoia to ensure our locked and lock-free paths interact
5953 // correctly with each other and Java-level accesses.
5954 OrderAccess::fence();
5955
5956 // If externally suspended while waiting, re-suspend
5957 if (jt->handle_special_suspend_equivalent_condition()) {
5958 jt->java_suspend_self();
5959 }
5960 }
5961
5962 void Parker::unpark() {
5963 int s, status ;
5964 status = pthread_mutex_lock(_mutex);
5965 assert (status == 0, "invariant") ;
5966 s = _counter;
5967 _counter = 1;
5968 if (s < 1) {
5969 // thread might be parked
5970 if (_cur_index != -1) {
5971 // thread is definitely parked
5972 if (WorkAroundNPTLTimedWaitHang) {
5973 status = pthread_cond_signal (&_cond[_cur_index]);
5974 assert (status == 0, "invariant");
5975 status = pthread_mutex_unlock(_mutex);
5976 assert (status == 0, "invariant");
5977 } else {
5978 status = pthread_mutex_unlock(_mutex);
5979 assert (status == 0, "invariant");
5980 status = pthread_cond_signal (&_cond[_cur_index]);
5981 assert (status == 0, "invariant");
5982 }
5983 } else {
5984 pthread_mutex_unlock(_mutex);
5985 assert (status == 0, "invariant") ;
5986 }
5987 } else {
5988 pthread_mutex_unlock(_mutex);
5989 assert (status == 0, "invariant") ;
5990 }
5991 }
5992
5993
5994 extern char** environ;
5995
5996 #ifndef __NR_fork
5997 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5998 #endif
5999
6000 #ifndef __NR_execve
6001 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
6002 #endif
6003
6004 // Run the specified command in a separate process. Return its exit value,
6005 // or -1 on failure (e.g. can't fork a new process).
6006 // Unlike system(), this function can be called from signal handler. It
6007 // doesn't block SIGINT et al.
6008 int os::fork_and_exec(char* cmd) {
6009 const char * argv[4] = {"sh", "-c", cmd, NULL};
6010
6011 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
6012 // pthread_atfork handlers and reset pthread library. All we need is a
6013 // separate process to execve. Make a direct syscall to fork process.
6014 // On IA64 there's no fork syscall, we have to use fork() and hope for
6015 // the best...
6016 pid_t pid = NOT_IA64(syscall(__NR_fork);)
6017 IA64_ONLY(fork();)
6018
6019 if (pid < 0) {
6020 // fork failed
6021 return -1;
6022
6023 } else if (pid == 0) {
6024 // child process
6025
6026 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6027 // first to kill every thread on the thread list. Because this list is
6028 // not reset by fork() (see notes above), execve() will instead kill
6029 // every thread in the parent process. We know this is the only thread
6030 // in the new process, so make a system call directly.
6031 // IA64 should use normal execve() from glibc to match the glibc fork()
6032 // above.
6033 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6034 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6035
6036 // execve failed
6037 _exit(-1);
6038
6039 } else {
6040 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6041 // care about the actual exit code, for now.
6042
6043 int status;
6044
6045 // Wait for the child process to exit. This returns immediately if
6046 // the child has already exited. */
6047 while (waitpid(pid, &status, 0) < 0) {
6048 switch (errno) {
6049 case ECHILD: return 0;
6050 case EINTR: break;
6051 default: return -1;
6052 }
6053 }
6054
6055 if (WIFEXITED(status)) {
6056 // The child exited normally; get its exit code.
6057 return WEXITSTATUS(status);
6058 } else if (WIFSIGNALED(status)) {
6059 // The child exited because of a signal
6060 // The best value to return is 0x80 + signal number,
6061 // because that is what all Unix shells do, and because
6062 // it allows callers to distinguish between process exit and
6063 // process death by signal.
6064 return 0x80 + WTERMSIG(status);
6065 } else {
6066 // Unknown exit code; pass it through
6067 return status;
6068 }
6069 }
6070 }
6071
6072 // is_headless_jre()
6073 //
6074 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6075 // in order to report if we are running in a headless jre
6076 //
6077 // Since JDK8 xawt/libmawt.so was moved into the same directory
6078 // as libawt.so, and renamed libawt_xawt.so
6079 //
6080 bool os::is_headless_jre() {
6081 struct stat statbuf;
6082 char buf[MAXPATHLEN];
6083 char libmawtpath[MAXPATHLEN];
6084 const char *xawtstr = "/xawt/libmawt.so";
6085 const char *new_xawtstr = "/libawt_xawt.so";
6086 char *p;
6087
6088 // Get path to libjvm.so
6089 os::jvm_path(buf, sizeof(buf));
6090
6091 // Get rid of libjvm.so
6092 p = strrchr(buf, '/');
6093 if (p == NULL) return false;
6094 else *p = '\0';
6095
6096 // Get rid of client or server
6097 p = strrchr(buf, '/');
6098 if (p == NULL) return false;
6099 else *p = '\0';
6100
6101 // check xawt/libmawt.so
6102 strcpy(libmawtpath, buf);
6103 strcat(libmawtpath, xawtstr);
6104 if (::stat(libmawtpath, &statbuf) == 0) return false;
6105
6106 // check libawt_xawt.so
6107 strcpy(libmawtpath, buf);
6108 strcat(libmawtpath, new_xawtstr);
6109 if (::stat(libmawtpath, &statbuf) == 0) return false;
6110
6111 return true;
6112 }
6113
6114 // Get the default path to the core file
6115 // Returns the length of the string
6116 int os::get_core_path(char* buffer, size_t bufferSize) {
6117 const char* p = get_current_directory(buffer, bufferSize);
6118
6119 if (p == NULL) {
6120 assert(p != NULL, "failed to get current directory");
6121 return 0;
6122 }
6123
6124 return strlen(buffer);
6125 }
6126
6127 #ifdef JAVASE_EMBEDDED
6128 //
6129 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6130 //
6131 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6132
6133 // ctor
6134 //
6135 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6136 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6137 _fd = fd;
6138
6139 if (os::create_thread(this, os::os_thread)) {
6140 _memnotify_thread = this;
6141 os::set_priority(this, NearMaxPriority);
6142 os::start_thread(this);
6143 }
6144 }
6145
6146 // Where all the work gets done
6147 //
6148 void MemNotifyThread::run() {
6149 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6150
6151 // Set up the select arguments
6152 fd_set rfds;
6153 if (_fd != -1) {
6154 FD_ZERO(&rfds);
6155 FD_SET(_fd, &rfds);
6156 }
6157
6158 // Now wait for the mem_notify device to wake up
6159 while (1) {
6160 // Wait for the mem_notify device to signal us..
6161 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6162 if (rc == -1) {
6163 perror("select!\n");
6164 break;
6165 } else if (rc) {
6166 //ssize_t free_before = os::available_memory();
6167 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6168
6169 // The kernel is telling us there is not much memory left...
6170 // try to do something about that
6171
6172 // If we are not already in a GC, try one.
6173 if (!Universe::heap()->is_gc_active()) {
6174 Universe::heap()->collect(GCCause::_allocation_failure);
6175
6176 //ssize_t free_after = os::available_memory();
6177 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6178 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6179 }
6180 // We might want to do something like the following if we find the GC's are not helping...
6181 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6182 }
6183 }
6184 }
6185
6186 //
6187 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6188 //
6189 void MemNotifyThread::start() {
6190 int fd;
6191 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6192 if (fd < 0) {
6193 return;
6194 }
6195
6196 if (memnotify_thread() == NULL) {
6197 new MemNotifyThread(fd);
6198 }
6199 }
6200
6201 #endif // JAVASE_EMBEDDED
6202
6203
6204 /////////////// Unit tests ///////////////
6205
6206 #ifndef PRODUCT
6207
6208 #define test_log(...) \
6209 do {\
6210 if (VerboseInternalVMTests) { \
6211 tty->print_cr(__VA_ARGS__); \
6212 tty->flush(); \
6213 }\
6214 } while (false)
6215
6216 class TestReserveMemorySpecial : AllStatic {
6217 public:
6218 static void small_page_write(void* addr, size_t size) {
6219 size_t page_size = os::vm_page_size();
6220
6221 char* end = (char*)addr + size;
6222 for (char* p = (char*)addr; p < end; p += page_size) {
6223 *p = 1;
6224 }
6225 }
6226
6227 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6228 if (!UseHugeTLBFS) {
6229 return;
6230 }
6231
6232 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6233
6234 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6235
6236 if (addr != NULL) {
6237 small_page_write(addr, size);
6238
6239 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6240 }
6241 }
6242
6243 static void test_reserve_memory_special_huge_tlbfs_only() {
6244 if (!UseHugeTLBFS) {
6245 return;
6246 }
6247
6248 size_t lp = os::large_page_size();
6249
6250 for (size_t size = lp; size <= lp * 10; size += lp) {
6251 test_reserve_memory_special_huge_tlbfs_only(size);
6252 }
6253 }
6254
6255 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6256 if (!UseHugeTLBFS) {
6257 return;
6258 }
6259
6260 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6261 size, alignment);
6262
6263 assert(size >= os::large_page_size(), "Incorrect input to test");
6264
6265 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6266
6267 if (addr != NULL) {
6268 small_page_write(addr, size);
6269
6270 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6271 }
6272 }
6273
6274 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6275 size_t lp = os::large_page_size();
6276 size_t ag = os::vm_allocation_granularity();
6277
6278 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6279 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6280 }
6281 }
6282
6283 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6284 size_t lp = os::large_page_size();
6285 size_t ag = os::vm_allocation_granularity();
6286
6287 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6288 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6289 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6290 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
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 - ag);
6293 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6294 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6295 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6296 }
6297
6298 static void test_reserve_memory_special_huge_tlbfs() {
6299 if (!UseHugeTLBFS) {
6300 return;
6301 }
6302
6303 test_reserve_memory_special_huge_tlbfs_only();
6304 test_reserve_memory_special_huge_tlbfs_mixed();
6305 }
6306
6307 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6308 if (!UseSHM) {
6309 return;
6310 }
6311
6312 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6313
6314 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6315
6316 if (addr != NULL) {
6317 assert(is_ptr_aligned(addr, alignment), "Check");
6318 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6319
6320 small_page_write(addr, size);
6321
6322 os::Linux::release_memory_special_shm(addr, size);
6323 }
6324 }
6325
6326 static void test_reserve_memory_special_shm() {
6327 size_t lp = os::large_page_size();
6328 size_t ag = os::vm_allocation_granularity();
6329
6330 for (size_t size = ag; size < lp * 3; size += ag) {
6331 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6332 test_reserve_memory_special_shm(size, alignment);
6333 }
6334 }
6335 }
6336
6337 static void test() {
6338 test_reserve_memory_special_huge_tlbfs();
6339 test_reserve_memory_special_shm();
6340 }
6341 };
6342
6343 void TestReserveMemorySpecial_test() {
6344 TestReserveMemorySpecial::test();
6345 }
6346
6347 #endif