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_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.inline.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "services/attachListener.hpp"
64 #include "services/memTracker.hpp"
65 #include "services/runtimeService.hpp"
66 #include "utilities/decoder.hpp"
67 #include "utilities/defaultStream.hpp"
68 #include "utilities/events.hpp"
69 #include "utilities/elfFile.hpp"
70 #include "utilities/growableArray.hpp"
71 #include "utilities/vmError.hpp"
72
73 // put OS-includes here
74 # include <sys/types.h>
75 # include <sys/mman.h>
76 # include <sys/stat.h>
77 # include <sys/select.h>
78 # include <pthread.h>
79 # include <signal.h>
80 # include <errno.h>
81 # include <dlfcn.h>
82 # include <stdio.h>
83 # include <unistd.h>
84 # include <sys/resource.h>
85 # include <pthread.h>
86 # include <sys/stat.h>
87 # include <sys/time.h>
88 # include <sys/times.h>
89 # include <sys/utsname.h>
90 # include <sys/socket.h>
91 # include <sys/wait.h>
92 # include <pwd.h>
93 # include <poll.h>
94 # include <semaphore.h>
95 # include <fcntl.h>
96 # include <string.h>
97 # include <syscall.h>
98 # include <sys/sysinfo.h>
99 # include <gnu/libc-version.h>
100 # include <sys/ipc.h>
101 # include <sys/shm.h>
102 # include <link.h>
103 # include <stdint.h>
104 # include <inttypes.h>
105 # include <sys/ioctl.h>
106
107 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
108
109 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
110 // getrusage() is prepared to handle the associated failure.
111 #ifndef RUSAGE_THREAD
112 #define RUSAGE_THREAD (1) /* only the calling thread */
113 #endif
114
115 #define MAX_PATH (2 * K)
116
117 #define MAX_SECS 100000000
118
119 // for timer info max values which include all bits
120 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
121
122 #define LARGEPAGES_BIT (1 << 6)
123 ////////////////////////////////////////////////////////////////////////////////
124 // global variables
125 julong os::Linux::_physical_memory = 0;
126
127 address os::Linux::_initial_thread_stack_bottom = NULL;
128 uintptr_t os::Linux::_initial_thread_stack_size = 0;
129
130 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
131 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
132 Mutex* os::Linux::_createThread_lock = NULL;
133 pthread_t os::Linux::_main_thread;
134 int os::Linux::_page_size = -1;
135 const int os::Linux::_vm_default_page_size = (8 * K);
136 bool os::Linux::_is_floating_stack = false;
137 bool os::Linux::_is_NPTL = false;
138 bool os::Linux::_supports_fast_thread_cpu_time = false;
139 const char * os::Linux::_glibc_version = NULL;
140 const char * os::Linux::_libpthread_version = NULL;
141 pthread_condattr_t os::Linux::_condattr[1];
142
143 static jlong initial_time_count=0;
144
145 static int clock_tics_per_sec = 100;
146
147 // For diagnostics to print a message once. see run_periodic_checks
148 static sigset_t check_signal_done;
149 static bool check_signals = true;
150
151 static pid_t _initial_pid = 0;
152
153 // Signal number used to suspend/resume a thread
154
155 // do not use any signal number less than SIGSEGV, see 4355769
156 static int SR_signum = SIGUSR2;
157 sigset_t SR_sigset;
158
159 // Used to protect dlsym() calls
160 static pthread_mutex_t dl_mutex;
161
162 // Declarations
163 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
164
165 #ifdef JAVASE_EMBEDDED
166 class MemNotifyThread: public Thread {
167 friend class VMStructs;
168 public:
169 virtual void run();
170
171 private:
172 static MemNotifyThread* _memnotify_thread;
173 int _fd;
174
175 public:
176
177 // Constructor
178 MemNotifyThread(int fd);
179
180 // Tester
181 bool is_memnotify_thread() const { return true; }
182
183 // Printing
184 char* name() const { return (char*)"Linux MemNotify Thread"; }
185
186 // Returns the single instance of the MemNotifyThread
187 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
188
189 // Create and start the single instance of MemNotifyThread
190 static void start();
191 };
192 #endif // JAVASE_EMBEDDED
193
194 // utility functions
195
196 static int SR_initialize();
197
198 julong os::available_memory() {
199 return Linux::available_memory();
200 }
201
202 julong os::Linux::available_memory() {
203 // values in struct sysinfo are "unsigned long"
204 struct sysinfo si;
205 sysinfo(&si);
206
207 return (julong)si.freeram * si.mem_unit;
208 }
209
210 julong os::physical_memory() {
211 return Linux::physical_memory();
212 }
213
214 ////////////////////////////////////////////////////////////////////////////////
215 // environment support
216
217 bool os::getenv(const char* name, char* buf, int len) {
218 const char* val = ::getenv(name);
219 if (val != NULL && strlen(val) < (size_t)len) {
220 strcpy(buf, val);
221 return true;
222 }
223 if (len > 0) buf[0] = 0; // return a null string
224 return false;
225 }
226
227
228 // Return true if user is running as root.
229
230 bool os::have_special_privileges() {
231 static bool init = false;
232 static bool privileges = false;
233 if (!init) {
234 privileges = (getuid() != geteuid()) || (getgid() != getegid());
235 init = true;
236 }
237 return privileges;
238 }
239
240
241 #ifndef SYS_gettid
242 // i386: 224, ia64: 1105, amd64: 186, sparc 143
243 #ifdef __ia64__
244 #define SYS_gettid 1105
245 #elif __i386__
246 #define SYS_gettid 224
247 #elif __amd64__
248 #define SYS_gettid 186
249 #elif __sparc__
250 #define SYS_gettid 143
251 #else
252 #error define gettid for the arch
253 #endif
254 #endif
255
256 // Cpu architecture string
257 #if defined(ZERO)
258 static char cpu_arch[] = ZERO_LIBARCH;
259 #elif defined(IA64)
260 static char cpu_arch[] = "ia64";
261 #elif defined(IA32)
262 static char cpu_arch[] = "i386";
263 #elif defined(AMD64)
264 static char cpu_arch[] = "amd64";
265 #elif defined(ARM)
266 static char cpu_arch[] = "arm";
267 #elif defined(PPC32)
268 static char cpu_arch[] = "ppc";
269 #elif defined(PPC64)
270 static char cpu_arch[] = "ppc64";
271 #elif defined(SPARC)
272 #ifdef _LP64
273 static char cpu_arch[] = "sparcv9";
274 #else
275 static char cpu_arch[] = "sparc";
276 #endif
277 #else
278 #error Add appropriate cpu_arch setting
279 #endif
280
281
282 // pid_t gettid()
283 //
284 // Returns the kernel thread id of the currently running thread. Kernel
285 // thread id is used to access /proc.
286 //
287 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
288 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
289 //
290 pid_t os::Linux::gettid() {
291 int rslt = syscall(SYS_gettid);
292 if (rslt == -1) {
293 // old kernel, no NPTL support
294 return getpid();
295 } else {
296 return (pid_t)rslt;
297 }
298 }
299
300 // Most versions of linux have a bug where the number of processors are
301 // determined by looking at the /proc file system. In a chroot environment,
302 // the system call returns 1. This causes the VM to act as if it is
303 // a single processor and elide locking (see is_MP() call).
304 static bool unsafe_chroot_detected = false;
305 static const char *unstable_chroot_error = "/proc file system not found.\n"
306 "Java may be unstable running multithreaded in a chroot "
307 "environment on Linux when /proc filesystem is not mounted.";
308
309 void os::Linux::initialize_system_info() {
310 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
311 if (processor_count() == 1) {
312 pid_t pid = os::Linux::gettid();
313 char fname[32];
314 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
315 FILE *fp = fopen(fname, "r");
316 if (fp == NULL) {
317 unsafe_chroot_detected = true;
318 } else {
319 fclose(fp);
320 }
321 }
322 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
323 assert(processor_count() > 0, "linux error");
324 }
325
326 void os::init_system_properties_values() {
327 // The next steps are taken in the product version:
328 //
329 // Obtain the JAVA_HOME value from the location of libjvm.so.
330 // This library should be located at:
331 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
332 //
333 // If "/jre/lib/" appears at the right place in the path, then we
334 // assume libjvm.so is installed in a JDK and we use this path.
335 //
336 // Otherwise exit with message: "Could not create the Java virtual machine."
337 //
338 // The following extra steps are taken in the debugging version:
339 //
340 // If "/jre/lib/" does NOT appear at the right place in the path
341 // instead of exit check for $JAVA_HOME environment variable.
342 //
343 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
344 // then we append a fake suffix "hotspot/libjvm.so" to this path so
345 // it looks like libjvm.so is installed there
346 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
347 //
348 // Otherwise exit.
349 //
350 // Important note: if the location of libjvm.so changes this
351 // code needs to be changed accordingly.
352
353 // See ld(1):
354 // The linker uses the following search paths to locate required
355 // shared libraries:
356 // 1: ...
357 // ...
358 // 7: The default directories, normally /lib and /usr/lib.
359 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
360 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
361 #else
362 #define DEFAULT_LIBPATH "/lib:/usr/lib"
363 #endif
364
365 // Base path of extensions installed on the system.
366 #define SYS_EXT_DIR "/usr/java/packages"
367 #define EXTENSIONS_DIR "/lib/ext"
368 #define ENDORSED_DIR "/lib/endorsed"
369
370 // Buffer that fits several sprintfs.
371 // Note that the space for the colon and the trailing null are provided
372 // by the nulls included by the sizeof operator.
373 const size_t bufsize =
374 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
375 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
376 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
377 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
378
379 // sysclasspath, java_home, dll_dir
380 {
381 char *pslash;
382 os::jvm_path(buf, bufsize);
383
384 // Found the full path to libjvm.so.
385 // Now cut the path to <java_home>/jre if we can.
386 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
387 pslash = strrchr(buf, '/');
388 if (pslash != NULL) {
389 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
390 }
391 Arguments::set_dll_dir(buf);
392
393 if (pslash != NULL) {
394 pslash = strrchr(buf, '/');
395 if (pslash != NULL) {
396 *pslash = '\0'; // Get rid of /<arch>.
397 pslash = strrchr(buf, '/');
398 if (pslash != NULL) {
399 *pslash = '\0'; // Get rid of /lib.
400 }
401 }
402 }
403 Arguments::set_java_home(buf);
404 set_boot_path('/', ':');
405 }
406
407 // Where to look for native libraries.
408 //
409 // Note: Due to a legacy implementation, most of the library path
410 // is set in the launcher. This was to accomodate linking restrictions
411 // on legacy Linux implementations (which are no longer supported).
412 // Eventually, all the library path setting will be done here.
413 //
414 // However, to prevent the proliferation of improperly built native
415 // libraries, the new path component /usr/java/packages is added here.
416 // Eventually, all the library path setting will be done here.
417 {
418 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
419 // should always exist (until the legacy problem cited above is
420 // addressed).
421 const char *v = ::getenv("LD_LIBRARY_PATH");
422 const char *v_colon = ":";
423 if (v == NULL) { v = ""; v_colon = ""; }
424 // That's +1 for the colon and +1 for the trailing '\0'.
425 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
426 strlen(v) + 1 +
427 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
428 mtInternal);
429 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
430 Arguments::set_library_path(ld_library_path);
431 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
432 }
433
434 // Extensions directories.
435 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
436 Arguments::set_ext_dirs(buf);
437
438 // Endorsed standards default directory.
439 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
440 Arguments::set_endorsed_dirs(buf);
441
442 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
443
444 #undef DEFAULT_LIBPATH
445 #undef SYS_EXT_DIR
446 #undef EXTENSIONS_DIR
447 #undef ENDORSED_DIR
448 }
449
450 ////////////////////////////////////////////////////////////////////////////////
451 // breakpoint support
452
453 void os::breakpoint() {
454 BREAKPOINT;
455 }
456
457 extern "C" void breakpoint() {
458 // use debugger to set breakpoint here
459 }
460
461 ////////////////////////////////////////////////////////////////////////////////
462 // signal support
463
464 debug_only(static bool signal_sets_initialized = false);
465 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
466
467 bool os::Linux::is_sig_ignored(int sig) {
468 struct sigaction oact;
469 sigaction(sig, (struct sigaction*)NULL, &oact);
470 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
471 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
472 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
473 return true;
474 } else {
475 return false;
476 }
477 }
478
479 void os::Linux::signal_sets_init() {
480 // Should also have an assertion stating we are still single-threaded.
481 assert(!signal_sets_initialized, "Already initialized");
482 // Fill in signals that are necessarily unblocked for all threads in
483 // the VM. Currently, we unblock the following signals:
484 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
485 // by -Xrs (=ReduceSignalUsage));
486 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
487 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
488 // the dispositions or masks wrt these signals.
489 // Programs embedding the VM that want to use the above signals for their
490 // own purposes must, at this time, use the "-Xrs" option to prevent
491 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
492 // (See bug 4345157, and other related bugs).
493 // In reality, though, unblocking these signals is really a nop, since
494 // these signals are not blocked by default.
495 sigemptyset(&unblocked_sigs);
496 sigemptyset(&allowdebug_blocked_sigs);
497 sigaddset(&unblocked_sigs, SIGILL);
498 sigaddset(&unblocked_sigs, SIGSEGV);
499 sigaddset(&unblocked_sigs, SIGBUS);
500 sigaddset(&unblocked_sigs, SIGFPE);
501 #if defined(PPC64)
502 sigaddset(&unblocked_sigs, SIGTRAP);
503 #endif
504 sigaddset(&unblocked_sigs, SR_signum);
505
506 if (!ReduceSignalUsage) {
507 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
508 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
509 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
510 }
511 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
512 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
513 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
514 }
515 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
516 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
517 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
518 }
519 }
520 // Fill in signals that are blocked by all but the VM thread.
521 sigemptyset(&vm_sigs);
522 if (!ReduceSignalUsage) {
523 sigaddset(&vm_sigs, BREAK_SIGNAL);
524 }
525 debug_only(signal_sets_initialized = true);
526
527 }
528
529 // These are signals that are unblocked while a thread is running Java.
530 // (For some reason, they get blocked by default.)
531 sigset_t* os::Linux::unblocked_signals() {
532 assert(signal_sets_initialized, "Not initialized");
533 return &unblocked_sigs;
534 }
535
536 // These are the signals that are blocked while a (non-VM) thread is
537 // running Java. Only the VM thread handles these signals.
538 sigset_t* os::Linux::vm_signals() {
539 assert(signal_sets_initialized, "Not initialized");
540 return &vm_sigs;
541 }
542
543 // These are signals that are blocked during cond_wait to allow debugger in
544 sigset_t* os::Linux::allowdebug_blocked_signals() {
545 assert(signal_sets_initialized, "Not initialized");
546 return &allowdebug_blocked_sigs;
547 }
548
549 void os::Linux::hotspot_sigmask(Thread* thread) {
550
551 //Save caller's signal mask before setting VM signal mask
552 sigset_t caller_sigmask;
553 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
554
555 OSThread* osthread = thread->osthread();
556 osthread->set_caller_sigmask(caller_sigmask);
557
558 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
559
560 if (!ReduceSignalUsage) {
561 if (thread->is_VM_thread()) {
562 // Only the VM thread handles BREAK_SIGNAL ...
563 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
564 } else {
565 // ... all other threads block BREAK_SIGNAL
566 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
567 }
568 }
569 }
570
571 //////////////////////////////////////////////////////////////////////////////
572 // detecting pthread library
573
574 void os::Linux::libpthread_init() {
575 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
576 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
577 // generic name for earlier versions.
578 // Define macros here so we can build HotSpot on old systems.
579 #ifndef _CS_GNU_LIBC_VERSION
580 #define _CS_GNU_LIBC_VERSION 2
581 #endif
582 #ifndef _CS_GNU_LIBPTHREAD_VERSION
583 #define _CS_GNU_LIBPTHREAD_VERSION 3
584 #endif
585
586 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
587 if (n > 0) {
588 char *str = (char *)malloc(n, mtInternal);
589 confstr(_CS_GNU_LIBC_VERSION, str, n);
590 os::Linux::set_glibc_version(str);
591 } else {
592 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
593 static char _gnu_libc_version[32];
594 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
595 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
596 os::Linux::set_glibc_version(_gnu_libc_version);
597 }
598
599 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
600 if (n > 0) {
601 char *str = (char *)malloc(n, mtInternal);
602 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
603 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
604 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
605 // is the case. LinuxThreads has a hard limit on max number of threads.
606 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
607 // On the other hand, NPTL does not have such a limit, sysconf()
608 // will return -1 and errno is not changed. Check if it is really NPTL.
609 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
610 strstr(str, "NPTL") &&
611 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
612 free(str);
613 os::Linux::set_libpthread_version("linuxthreads");
614 } else {
615 os::Linux::set_libpthread_version(str);
616 }
617 } else {
618 // glibc before 2.3.2 only has LinuxThreads.
619 os::Linux::set_libpthread_version("linuxthreads");
620 }
621
622 if (strstr(libpthread_version(), "NPTL")) {
623 os::Linux::set_is_NPTL();
624 } else {
625 os::Linux::set_is_LinuxThreads();
626 }
627
628 // LinuxThreads have two flavors: floating-stack mode, which allows variable
629 // stack size; and fixed-stack mode. NPTL is always floating-stack.
630 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
631 os::Linux::set_is_floating_stack();
632 }
633 }
634
635 /////////////////////////////////////////////////////////////////////////////
636 // thread stack
637
638 // Force Linux kernel to expand current thread stack. If "bottom" is close
639 // to the stack guard, caller should block all signals.
640 //
641 // MAP_GROWSDOWN:
642 // A special mmap() flag that is used to implement thread stacks. It tells
643 // kernel that the memory region should extend downwards when needed. This
644 // allows early versions of LinuxThreads to only mmap the first few pages
645 // when creating a new thread. Linux kernel will automatically expand thread
646 // stack as needed (on page faults).
647 //
648 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
649 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
650 // region, it's hard to tell if the fault is due to a legitimate stack
651 // access or because of reading/writing non-exist memory (e.g. buffer
652 // overrun). As a rule, if the fault happens below current stack pointer,
653 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
654 // application (see Linux kernel fault.c).
655 //
656 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
657 // stack overflow detection.
658 //
659 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
660 // not use this flag. However, the stack of initial thread is not created
661 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
662 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
663 // and then attach the thread to JVM.
664 //
665 // To get around the problem and allow stack banging on Linux, we need to
666 // manually expand thread stack after receiving the SIGSEGV.
667 //
668 // There are two ways to expand thread stack to address "bottom", we used
669 // both of them in JVM before 1.5:
670 // 1. adjust stack pointer first so that it is below "bottom", and then
671 // touch "bottom"
672 // 2. mmap() the page in question
673 //
674 // Now alternate signal stack is gone, it's harder to use 2. For instance,
675 // if current sp is already near the lower end of page 101, and we need to
676 // call mmap() to map page 100, it is possible that part of the mmap() frame
677 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
678 // That will destroy the mmap() frame and cause VM to crash.
679 //
680 // The following code works by adjusting sp first, then accessing the "bottom"
681 // page to force a page fault. Linux kernel will then automatically expand the
682 // stack mapping.
683 //
684 // _expand_stack_to() assumes its frame size is less than page size, which
685 // should always be true if the function is not inlined.
686
687 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
688 #define NOINLINE
689 #else
690 #define NOINLINE __attribute__ ((noinline))
691 #endif
692
693 static void _expand_stack_to(address bottom) NOINLINE;
694
695 static void _expand_stack_to(address bottom) {
696 address sp;
697 size_t size;
698 volatile char *p;
699
700 // Adjust bottom to point to the largest address within the same page, it
701 // gives us a one-page buffer if alloca() allocates slightly more memory.
702 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
703 bottom += os::Linux::page_size() - 1;
704
705 // sp might be slightly above current stack pointer; if that's the case, we
706 // will alloca() a little more space than necessary, which is OK. Don't use
707 // os::current_stack_pointer(), as its result can be slightly below current
708 // stack pointer, causing us to not alloca enough to reach "bottom".
709 sp = (address)&sp;
710
711 if (sp > bottom) {
712 size = sp - bottom;
713 p = (volatile char *)alloca(size);
714 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
715 p[0] = '\0';
716 }
717 }
718
719 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
720 assert(t!=NULL, "just checking");
721 assert(t->osthread()->expanding_stack(), "expand should be set");
722 assert(t->stack_base() != NULL, "stack_base was not initialized");
723
724 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
725 sigset_t mask_all, old_sigset;
726 sigfillset(&mask_all);
727 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
728 _expand_stack_to(addr);
729 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
730 return true;
731 }
732 return false;
733 }
734
735 //////////////////////////////////////////////////////////////////////////////
736 // create new thread
737
738 static address highest_vm_reserved_address();
739
740 // check if it's safe to start a new thread
741 static bool _thread_safety_check(Thread* thread) {
742 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
743 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
744 // Heap is mmap'ed at lower end of memory space. Thread stacks are
745 // allocated (MAP_FIXED) from high address space. Every thread stack
746 // occupies a fixed size slot (usually 2Mbytes, but user can change
747 // it to other values if they rebuild LinuxThreads).
748 //
749 // Problem with MAP_FIXED is that mmap() can still succeed even part of
750 // the memory region has already been mmap'ed. That means if we have too
751 // many threads and/or very large heap, eventually thread stack will
752 // collide with heap.
753 //
754 // Here we try to prevent heap/stack collision by comparing current
755 // stack bottom with the highest address that has been mmap'ed by JVM
756 // plus a safety margin for memory maps created by native code.
757 //
758 // This feature can be disabled by setting ThreadSafetyMargin to 0
759 //
760 if (ThreadSafetyMargin > 0) {
761 address stack_bottom = os::current_stack_base() - os::current_stack_size();
762
763 // not safe if our stack extends below the safety margin
764 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
765 } else {
766 return true;
767 }
768 } else {
769 // Floating stack LinuxThreads or NPTL:
770 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
771 // there's not enough space left, pthread_create() will fail. If we come
772 // here, that means enough space has been reserved for stack.
773 return true;
774 }
775 }
776
777 // Thread start routine for all newly created threads
778 static void *java_start(Thread *thread) {
779 // Try to randomize the cache line index of hot stack frames.
780 // This helps when threads of the same stack traces evict each other's
781 // cache lines. The threads can be either from the same JVM instance, or
782 // from different JVM instances. The benefit is especially true for
783 // processors with hyperthreading technology.
784 static int counter = 0;
785 int pid = os::current_process_id();
786 alloca(((pid ^ counter++) & 7) * 128);
787
788 ThreadLocalStorage::set_thread(thread);
789
790 OSThread* osthread = thread->osthread();
791 Monitor* sync = osthread->startThread_lock();
792
793 // non floating stack LinuxThreads needs extra check, see above
794 if (!_thread_safety_check(thread)) {
795 // notify parent thread
796 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
797 osthread->set_state(ZOMBIE);
798 sync->notify_all();
799 return NULL;
800 }
801
802 // thread_id is kernel thread id (similar to Solaris LWP id)
803 osthread->set_thread_id(os::Linux::gettid());
804
805 if (UseNUMA) {
806 int lgrp_id = os::numa_get_group_id();
807 if (lgrp_id != -1) {
808 thread->set_lgrp_id(lgrp_id);
809 }
810 }
811 // initialize signal mask for this thread
812 os::Linux::hotspot_sigmask(thread);
813
814 // initialize floating point control register
815 os::Linux::init_thread_fpu_state();
816
817 // handshaking with parent thread
818 {
819 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
820
821 // notify parent thread
822 osthread->set_state(INITIALIZED);
823 sync->notify_all();
824
825 // wait until os::start_thread()
826 while (osthread->get_state() == INITIALIZED) {
827 sync->wait(Mutex::_no_safepoint_check_flag);
828 }
829 }
830
831 // call one more level start routine
832 thread->run();
833
834 return 0;
835 }
836
837 bool os::create_thread(Thread* thread, ThreadType thr_type,
838 size_t stack_size) {
839 assert(thread->osthread() == NULL, "caller responsible");
840
841 // Allocate the OSThread object
842 OSThread* osthread = new OSThread(NULL, NULL);
843 if (osthread == NULL) {
844 return false;
845 }
846
847 // set the correct thread state
848 osthread->set_thread_type(thr_type);
849
850 // Initial state is ALLOCATED but not INITIALIZED
851 osthread->set_state(ALLOCATED);
852
853 thread->set_osthread(osthread);
854
855 // init thread attributes
856 pthread_attr_t attr;
857 pthread_attr_init(&attr);
858 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
859
860 // stack size
861 if (os::Linux::supports_variable_stack_size()) {
862 // calculate stack size if it's not specified by caller
863 if (stack_size == 0) {
864 stack_size = os::Linux::default_stack_size(thr_type);
865
866 switch (thr_type) {
867 case os::java_thread:
868 // Java threads use ThreadStackSize which default value can be
869 // changed with the flag -Xss
870 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
871 stack_size = JavaThread::stack_size_at_create();
872 break;
873 case os::compiler_thread:
874 if (CompilerThreadStackSize > 0) {
875 stack_size = (size_t)(CompilerThreadStackSize * K);
876 break;
877 } // else fall through:
878 // use VMThreadStackSize if CompilerThreadStackSize is not defined
879 case os::vm_thread:
880 case os::pgc_thread:
881 case os::cgc_thread:
882 case os::watcher_thread:
883 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
884 break;
885 }
886 }
887
888 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
889 pthread_attr_setstacksize(&attr, stack_size);
890 } else {
891 // let pthread_create() pick the default value.
892 }
893
894 // glibc guard page
895 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
896
897 ThreadState state;
898
899 {
900 // Serialize thread creation if we are running with fixed stack LinuxThreads
901 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
902 if (lock) {
903 os::Linux::createThread_lock()->lock_without_safepoint_check();
904 }
905
906 pthread_t tid;
907 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
908
909 pthread_attr_destroy(&attr);
910
911 if (ret != 0) {
912 if (PrintMiscellaneous && (Verbose || WizardMode)) {
913 perror("pthread_create()");
914 }
915 // Need to clean up stuff we've allocated so far
916 thread->set_osthread(NULL);
917 delete osthread;
918 if (lock) os::Linux::createThread_lock()->unlock();
919 return false;
920 }
921
922 // Store pthread info into the OSThread
923 osthread->set_pthread_id(tid);
924
925 // Wait until child thread is either initialized or aborted
926 {
927 Monitor* sync_with_child = osthread->startThread_lock();
928 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
929 while ((state = osthread->get_state()) == ALLOCATED) {
930 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
931 }
932 }
933
934 if (lock) {
935 os::Linux::createThread_lock()->unlock();
936 }
937 }
938
939 // Aborted due to thread limit being reached
940 if (state == ZOMBIE) {
941 thread->set_osthread(NULL);
942 delete osthread;
943 return false;
944 }
945
946 // The thread is returned suspended (in state INITIALIZED),
947 // and is started higher up in the call chain
948 assert(state == INITIALIZED, "race condition");
949 return true;
950 }
951
952 /////////////////////////////////////////////////////////////////////////////
953 // attach existing thread
954
955 // bootstrap the main thread
956 bool os::create_main_thread(JavaThread* thread) {
957 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
958 return create_attached_thread(thread);
959 }
960
961 bool os::create_attached_thread(JavaThread* thread) {
962 #ifdef ASSERT
963 thread->verify_not_published();
964 #endif
965
966 // Allocate the OSThread object
967 OSThread* osthread = new OSThread(NULL, NULL);
968
969 if (osthread == NULL) {
970 return false;
971 }
972
973 // Store pthread info into the OSThread
974 osthread->set_thread_id(os::Linux::gettid());
975 osthread->set_pthread_id(::pthread_self());
976
977 // initialize floating point control register
978 os::Linux::init_thread_fpu_state();
979
980 // Initial thread state is RUNNABLE
981 osthread->set_state(RUNNABLE);
982
983 thread->set_osthread(osthread);
984
985 if (UseNUMA) {
986 int lgrp_id = os::numa_get_group_id();
987 if (lgrp_id != -1) {
988 thread->set_lgrp_id(lgrp_id);
989 }
990 }
991
992 if (os::Linux::is_initial_thread()) {
993 // If current thread is initial thread, its stack is mapped on demand,
994 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
995 // the entire stack region to avoid SEGV in stack banging.
996 // It is also useful to get around the heap-stack-gap problem on SuSE
997 // kernel (see 4821821 for details). We first expand stack to the top
998 // of yellow zone, then enable stack yellow zone (order is significant,
999 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1000 // is no gap between the last two virtual memory regions.
1001
1002 JavaThread *jt = (JavaThread *)thread;
1003 address addr = jt->stack_yellow_zone_base();
1004 assert(addr != NULL, "initialization problem?");
1005 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1006
1007 osthread->set_expanding_stack();
1008 os::Linux::manually_expand_stack(jt, addr);
1009 osthread->clear_expanding_stack();
1010 }
1011
1012 // initialize signal mask for this thread
1013 // and save the caller's signal mask
1014 os::Linux::hotspot_sigmask(thread);
1015
1016 return true;
1017 }
1018
1019 void os::pd_start_thread(Thread* thread) {
1020 OSThread * osthread = thread->osthread();
1021 assert(osthread->get_state() != INITIALIZED, "just checking");
1022 Monitor* sync_with_child = osthread->startThread_lock();
1023 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1024 sync_with_child->notify();
1025 }
1026
1027 // Free Linux resources related to the OSThread
1028 void os::free_thread(OSThread* osthread) {
1029 assert(osthread != NULL, "osthread not set");
1030
1031 if (Thread::current()->osthread() == osthread) {
1032 // Restore caller's signal mask
1033 sigset_t sigmask = osthread->caller_sigmask();
1034 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1035 }
1036
1037 delete osthread;
1038 }
1039
1040 //////////////////////////////////////////////////////////////////////////////
1041 // thread local storage
1042
1043 // Restore the thread pointer if the destructor is called. This is in case
1044 // someone from JNI code sets up a destructor with pthread_key_create to run
1045 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1046 // will hang or crash. When detachCurrentThread is called the key will be set
1047 // to null and we will not be called again. If detachCurrentThread is never
1048 // called we could loop forever depending on the pthread implementation.
1049 static void restore_thread_pointer(void* p) {
1050 Thread* thread = (Thread*) p;
1051 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1052 }
1053
1054 int os::allocate_thread_local_storage() {
1055 pthread_key_t key;
1056 int rslt = pthread_key_create(&key, restore_thread_pointer);
1057 assert(rslt == 0, "cannot allocate thread local storage");
1058 return (int)key;
1059 }
1060
1061 // Note: This is currently not used by VM, as we don't destroy TLS key
1062 // on VM exit.
1063 void os::free_thread_local_storage(int index) {
1064 int rslt = pthread_key_delete((pthread_key_t)index);
1065 assert(rslt == 0, "invalid index");
1066 }
1067
1068 void os::thread_local_storage_at_put(int index, void* value) {
1069 int rslt = pthread_setspecific((pthread_key_t)index, value);
1070 assert(rslt == 0, "pthread_setspecific failed");
1071 }
1072
1073 extern "C" Thread* get_thread() {
1074 return ThreadLocalStorage::thread();
1075 }
1076
1077 //////////////////////////////////////////////////////////////////////////////
1078 // initial thread
1079
1080 // Check if current thread is the initial thread, similar to Solaris thr_main.
1081 bool os::Linux::is_initial_thread(void) {
1082 char dummy;
1083 // If called before init complete, thread stack bottom will be null.
1084 // Can be called if fatal error occurs before initialization.
1085 if (initial_thread_stack_bottom() == NULL) return false;
1086 assert(initial_thread_stack_bottom() != NULL &&
1087 initial_thread_stack_size() != 0,
1088 "os::init did not locate initial thread's stack region");
1089 if ((address)&dummy >= initial_thread_stack_bottom() &&
1090 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1091 return true;
1092 } else {
1093 return false;
1094 }
1095 }
1096
1097 // Find the virtual memory area that contains addr
1098 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1099 FILE *fp = fopen("/proc/self/maps", "r");
1100 if (fp) {
1101 address low, high;
1102 while (!feof(fp)) {
1103 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1104 if (low <= addr && addr < high) {
1105 if (vma_low) *vma_low = low;
1106 if (vma_high) *vma_high = high;
1107 fclose(fp);
1108 return true;
1109 }
1110 }
1111 for (;;) {
1112 int ch = fgetc(fp);
1113 if (ch == EOF || ch == (int)'\n') break;
1114 }
1115 }
1116 fclose(fp);
1117 }
1118 return false;
1119 }
1120
1121 // Locate initial thread stack. This special handling of initial thread stack
1122 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1123 // bogus value for initial thread.
1124 void os::Linux::capture_initial_stack(size_t max_size) {
1125 // stack size is the easy part, get it from RLIMIT_STACK
1126 size_t stack_size;
1127 struct rlimit rlim;
1128 getrlimit(RLIMIT_STACK, &rlim);
1129 stack_size = rlim.rlim_cur;
1130
1131 // 6308388: a bug in ld.so will relocate its own .data section to the
1132 // lower end of primordial stack; reduce ulimit -s value a little bit
1133 // so we won't install guard page on ld.so's data section.
1134 stack_size -= 2 * page_size();
1135
1136 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1137 // 7.1, in both cases we will get 2G in return value.
1138 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1139 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1140 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1141 // in case other parts in glibc still assumes 2M max stack size.
1142 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1143 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1144 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1145 stack_size = 2 * K * K IA64_ONLY(*2);
1146 }
1147 // Try to figure out where the stack base (top) is. This is harder.
1148 //
1149 // When an application is started, glibc saves the initial stack pointer in
1150 // a global variable "__libc_stack_end", which is then used by system
1151 // libraries. __libc_stack_end should be pretty close to stack top. The
1152 // variable is available since the very early days. However, because it is
1153 // a private interface, it could disappear in the future.
1154 //
1155 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1156 // to __libc_stack_end, it is very close to stack top, but isn't the real
1157 // stack top. Note that /proc may not exist if VM is running as a chroot
1158 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1159 // /proc/<pid>/stat could change in the future (though unlikely).
1160 //
1161 // We try __libc_stack_end first. If that doesn't work, look for
1162 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1163 // as a hint, which should work well in most cases.
1164
1165 uintptr_t stack_start;
1166
1167 // try __libc_stack_end first
1168 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1169 if (p && *p) {
1170 stack_start = *p;
1171 } else {
1172 // see if we can get the start_stack field from /proc/self/stat
1173 FILE *fp;
1174 int pid;
1175 char state;
1176 int ppid;
1177 int pgrp;
1178 int session;
1179 int nr;
1180 int tpgrp;
1181 unsigned long flags;
1182 unsigned long minflt;
1183 unsigned long cminflt;
1184 unsigned long majflt;
1185 unsigned long cmajflt;
1186 unsigned long utime;
1187 unsigned long stime;
1188 long cutime;
1189 long cstime;
1190 long prio;
1191 long nice;
1192 long junk;
1193 long it_real;
1194 uintptr_t start;
1195 uintptr_t vsize;
1196 intptr_t rss;
1197 uintptr_t rsslim;
1198 uintptr_t scodes;
1199 uintptr_t ecode;
1200 int i;
1201
1202 // Figure what the primordial thread stack base is. Code is inspired
1203 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1204 // followed by command name surrounded by parentheses, state, etc.
1205 char stat[2048];
1206 int statlen;
1207
1208 fp = fopen("/proc/self/stat", "r");
1209 if (fp) {
1210 statlen = fread(stat, 1, 2047, fp);
1211 stat[statlen] = '\0';
1212 fclose(fp);
1213
1214 // Skip pid and the command string. Note that we could be dealing with
1215 // weird command names, e.g. user could decide to rename java launcher
1216 // to "java 1.4.2 :)", then the stat file would look like
1217 // 1234 (java 1.4.2 :)) R ... ...
1218 // We don't really need to know the command string, just find the last
1219 // occurrence of ")" and then start parsing from there. See bug 4726580.
1220 char * s = strrchr(stat, ')');
1221
1222 i = 0;
1223 if (s) {
1224 // Skip blank chars
1225 do s++; while (isspace(*s));
1226
1227 #define _UFM UINTX_FORMAT
1228 #define _DFM INTX_FORMAT
1229
1230 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1231 // 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
1232 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,
1233 &state, // 3 %c
1234 &ppid, // 4 %d
1235 &pgrp, // 5 %d
1236 &session, // 6 %d
1237 &nr, // 7 %d
1238 &tpgrp, // 8 %d
1239 &flags, // 9 %lu
1240 &minflt, // 10 %lu
1241 &cminflt, // 11 %lu
1242 &majflt, // 12 %lu
1243 &cmajflt, // 13 %lu
1244 &utime, // 14 %lu
1245 &stime, // 15 %lu
1246 &cutime, // 16 %ld
1247 &cstime, // 17 %ld
1248 &prio, // 18 %ld
1249 &nice, // 19 %ld
1250 &junk, // 20 %ld
1251 &it_real, // 21 %ld
1252 &start, // 22 UINTX_FORMAT
1253 &vsize, // 23 UINTX_FORMAT
1254 &rss, // 24 INTX_FORMAT
1255 &rsslim, // 25 UINTX_FORMAT
1256 &scodes, // 26 UINTX_FORMAT
1257 &ecode, // 27 UINTX_FORMAT
1258 &stack_start); // 28 UINTX_FORMAT
1259 }
1260
1261 #undef _UFM
1262 #undef _DFM
1263
1264 if (i != 28 - 2) {
1265 assert(false, "Bad conversion from /proc/self/stat");
1266 // product mode - assume we are the initial thread, good luck in the
1267 // embedded case.
1268 warning("Can't detect initial thread stack location - bad conversion");
1269 stack_start = (uintptr_t) &rlim;
1270 }
1271 } else {
1272 // For some reason we can't open /proc/self/stat (for example, running on
1273 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1274 // most cases, so don't abort:
1275 warning("Can't detect initial thread stack location - no /proc/self/stat");
1276 stack_start = (uintptr_t) &rlim;
1277 }
1278 }
1279
1280 // Now we have a pointer (stack_start) very close to the stack top, the
1281 // next thing to do is to figure out the exact location of stack top. We
1282 // can find out the virtual memory area that contains stack_start by
1283 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1284 // and its upper limit is the real stack top. (again, this would fail if
1285 // running inside chroot, because /proc may not exist.)
1286
1287 uintptr_t stack_top;
1288 address low, high;
1289 if (find_vma((address)stack_start, &low, &high)) {
1290 // success, "high" is the true stack top. (ignore "low", because initial
1291 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1292 stack_top = (uintptr_t)high;
1293 } else {
1294 // failed, likely because /proc/self/maps does not exist
1295 warning("Can't detect initial thread stack location - find_vma failed");
1296 // best effort: stack_start is normally within a few pages below the real
1297 // stack top, use it as stack top, and reduce stack size so we won't put
1298 // guard page outside stack.
1299 stack_top = stack_start;
1300 stack_size -= 16 * page_size();
1301 }
1302
1303 // stack_top could be partially down the page so align it
1304 stack_top = align_size_up(stack_top, page_size());
1305
1306 if (max_size && stack_size > max_size) {
1307 _initial_thread_stack_size = max_size;
1308 } else {
1309 _initial_thread_stack_size = stack_size;
1310 }
1311
1312 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1313 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1314 }
1315
1316 ////////////////////////////////////////////////////////////////////////////////
1317 // time support
1318
1319 // Time since start-up in seconds to a fine granularity.
1320 // Used by VMSelfDestructTimer and the MemProfiler.
1321 double os::elapsedTime() {
1322
1323 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1324 }
1325
1326 jlong os::elapsed_counter() {
1327 return javaTimeNanos() - initial_time_count;
1328 }
1329
1330 jlong os::elapsed_frequency() {
1331 return NANOSECS_PER_SEC; // nanosecond resolution
1332 }
1333
1334 bool os::supports_vtime() { return true; }
1335 bool os::enable_vtime() { return false; }
1336 bool os::vtime_enabled() { return false; }
1337
1338 double os::elapsedVTime() {
1339 struct rusage usage;
1340 int retval = getrusage(RUSAGE_THREAD, &usage);
1341 if (retval == 0) {
1342 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);
1343 } else {
1344 // better than nothing, but not much
1345 return elapsedTime();
1346 }
1347 }
1348
1349 jlong os::javaTimeMillis() {
1350 timeval time;
1351 int status = gettimeofday(&time, NULL);
1352 assert(status != -1, "linux error");
1353 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1354 }
1355
1356 #ifndef CLOCK_MONOTONIC
1357 #define CLOCK_MONOTONIC (1)
1358 #endif
1359
1360 void os::Linux::clock_init() {
1361 // we do dlopen's in this particular order due to bug in linux
1362 // dynamical loader (see 6348968) leading to crash on exit
1363 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1364 if (handle == NULL) {
1365 handle = dlopen("librt.so", RTLD_LAZY);
1366 }
1367
1368 if (handle) {
1369 int (*clock_getres_func)(clockid_t, struct timespec*) =
1370 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1371 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1372 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1373 if (clock_getres_func && clock_gettime_func) {
1374 // See if monotonic clock is supported by the kernel. Note that some
1375 // early implementations simply return kernel jiffies (updated every
1376 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1377 // for nano time (though the monotonic property is still nice to have).
1378 // It's fixed in newer kernels, however clock_getres() still returns
1379 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1380 // resolution for now. Hopefully as people move to new kernels, this
1381 // won't be a problem.
1382 struct timespec res;
1383 struct timespec tp;
1384 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1385 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1386 // yes, monotonic clock is supported
1387 _clock_gettime = clock_gettime_func;
1388 return;
1389 } else {
1390 // close librt if there is no monotonic clock
1391 dlclose(handle);
1392 }
1393 }
1394 }
1395 warning("No monotonic clock was available - timed services may " \
1396 "be adversely affected if the time-of-day clock changes");
1397 }
1398
1399 #ifndef SYS_clock_getres
1400 #if defined(IA32) || defined(AMD64)
1401 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1402 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1403 #else
1404 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1405 #define sys_clock_getres(x,y) -1
1406 #endif
1407 #else
1408 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1409 #endif
1410
1411 void os::Linux::fast_thread_clock_init() {
1412 if (!UseLinuxPosixThreadCPUClocks) {
1413 return;
1414 }
1415 clockid_t clockid;
1416 struct timespec tp;
1417 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1418 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1419
1420 // Switch to using fast clocks for thread cpu time if
1421 // the sys_clock_getres() returns 0 error code.
1422 // Note, that some kernels may support the current thread
1423 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1424 // returned by the pthread_getcpuclockid().
1425 // If the fast Posix clocks are supported then the sys_clock_getres()
1426 // must return at least tp.tv_sec == 0 which means a resolution
1427 // better than 1 sec. This is extra check for reliability.
1428
1429 if (pthread_getcpuclockid_func &&
1430 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1431 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1432 _supports_fast_thread_cpu_time = true;
1433 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1434 }
1435 }
1436
1437 jlong os::javaTimeNanos() {
1438 if (os::supports_monotonic_clock()) {
1439 struct timespec tp;
1440 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1441 assert(status == 0, "gettime error");
1442 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1443 return result;
1444 } else {
1445 timeval time;
1446 int status = gettimeofday(&time, NULL);
1447 assert(status != -1, "linux error");
1448 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1449 return 1000 * usecs;
1450 }
1451 }
1452
1453 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1454 if (os::supports_monotonic_clock()) {
1455 info_ptr->max_value = ALL_64_BITS;
1456
1457 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1458 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1459 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1460 } else {
1461 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1462 info_ptr->max_value = ALL_64_BITS;
1463
1464 // gettimeofday is a real time clock so it skips
1465 info_ptr->may_skip_backward = true;
1466 info_ptr->may_skip_forward = true;
1467 }
1468
1469 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1470 }
1471
1472 // Return the real, user, and system times in seconds from an
1473 // arbitrary fixed point in the past.
1474 bool os::getTimesSecs(double* process_real_time,
1475 double* process_user_time,
1476 double* process_system_time) {
1477 struct tms ticks;
1478 clock_t real_ticks = times(&ticks);
1479
1480 if (real_ticks == (clock_t) (-1)) {
1481 return false;
1482 } else {
1483 double ticks_per_second = (double) clock_tics_per_sec;
1484 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1485 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1486 *process_real_time = ((double) real_ticks) / ticks_per_second;
1487
1488 return true;
1489 }
1490 }
1491
1492
1493 char * os::local_time_string(char *buf, size_t buflen) {
1494 struct tm t;
1495 time_t long_time;
1496 time(&long_time);
1497 localtime_r(&long_time, &t);
1498 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1499 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1500 t.tm_hour, t.tm_min, t.tm_sec);
1501 return buf;
1502 }
1503
1504 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1505 return localtime_r(clock, res);
1506 }
1507
1508 ////////////////////////////////////////////////////////////////////////////////
1509 // runtime exit support
1510
1511 // Note: os::shutdown() might be called very early during initialization, or
1512 // called from signal handler. Before adding something to os::shutdown(), make
1513 // sure it is async-safe and can handle partially initialized VM.
1514 void os::shutdown() {
1515
1516 // allow PerfMemory to attempt cleanup of any persistent resources
1517 perfMemory_exit();
1518
1519 // needs to remove object in file system
1520 AttachListener::abort();
1521
1522 // flush buffered output, finish log files
1523 ostream_abort();
1524
1525 // Check for abort hook
1526 abort_hook_t abort_hook = Arguments::abort_hook();
1527 if (abort_hook != NULL) {
1528 abort_hook();
1529 }
1530
1531 }
1532
1533 // Note: os::abort() might be called very early during initialization, or
1534 // called from signal handler. Before adding something to os::abort(), make
1535 // sure it is async-safe and can handle partially initialized VM.
1536 void os::abort(bool dump_core) {
1537 os::shutdown();
1538 if (dump_core) {
1539 #ifndef PRODUCT
1540 fdStream out(defaultStream::output_fd());
1541 out.print_raw("Current thread is ");
1542 char buf[16];
1543 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1544 out.print_raw_cr(buf);
1545 out.print_raw_cr("Dumping core ...");
1546 #endif
1547 ::abort(); // dump core
1548 }
1549
1550 ::exit(1);
1551 }
1552
1553 // Die immediately, no exit hook, no abort hook, no cleanup.
1554 void os::die() {
1555 // _exit() on LinuxThreads only kills current thread
1556 ::abort();
1557 }
1558
1559
1560 // This method is a copy of JDK's sysGetLastErrorString
1561 // from src/solaris/hpi/src/system_md.c
1562
1563 size_t os::lasterror(char *buf, size_t len) {
1564 if (errno == 0) return 0;
1565
1566 const char *s = ::strerror(errno);
1567 size_t n = ::strlen(s);
1568 if (n >= len) {
1569 n = len - 1;
1570 }
1571 ::strncpy(buf, s, n);
1572 buf[n] = '\0';
1573 return n;
1574 }
1575
1576 intx os::current_thread_id() { return (intx)pthread_self(); }
1577 int os::current_process_id() {
1578
1579 // Under the old linux thread library, linux gives each thread
1580 // its own process id. Because of this each thread will return
1581 // a different pid if this method were to return the result
1582 // of getpid(2). Linux provides no api that returns the pid
1583 // of the launcher thread for the vm. This implementation
1584 // returns a unique pid, the pid of the launcher thread
1585 // that starts the vm 'process'.
1586
1587 // Under the NPTL, getpid() returns the same pid as the
1588 // launcher thread rather than a unique pid per thread.
1589 // Use gettid() if you want the old pre NPTL behaviour.
1590
1591 // if you are looking for the result of a call to getpid() that
1592 // returns a unique pid for the calling thread, then look at the
1593 // OSThread::thread_id() method in osThread_linux.hpp file
1594
1595 return (int)(_initial_pid ? _initial_pid : getpid());
1596 }
1597
1598 // DLL functions
1599
1600 const char* os::dll_file_extension() { return ".so"; }
1601
1602 // This must be hard coded because it's the system's temporary
1603 // directory not the java application's temp directory, ala java.io.tmpdir.
1604 const char* os::get_temp_directory() { return "/tmp"; }
1605
1606 static bool file_exists(const char* filename) {
1607 struct stat statbuf;
1608 if (filename == NULL || strlen(filename) == 0) {
1609 return false;
1610 }
1611 return os::stat(filename, &statbuf) == 0;
1612 }
1613
1614 bool os::dll_build_name(char* buffer, size_t buflen,
1615 const char* pname, const char* fname) {
1616 bool retval = false;
1617 // Copied from libhpi
1618 const size_t pnamelen = pname ? strlen(pname) : 0;
1619
1620 // Return error on buffer overflow.
1621 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1622 return retval;
1623 }
1624
1625 if (pnamelen == 0) {
1626 snprintf(buffer, buflen, "lib%s.so", fname);
1627 retval = true;
1628 } else if (strchr(pname, *os::path_separator()) != NULL) {
1629 int n;
1630 char** pelements = split_path(pname, &n);
1631 if (pelements == NULL) {
1632 return false;
1633 }
1634 for (int i = 0; i < n; i++) {
1635 // Really shouldn't be NULL, but check can't hurt
1636 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1637 continue; // skip the empty path values
1638 }
1639 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1640 if (file_exists(buffer)) {
1641 retval = true;
1642 break;
1643 }
1644 }
1645 // release the storage
1646 for (int i = 0; i < n; i++) {
1647 if (pelements[i] != NULL) {
1648 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1649 }
1650 }
1651 if (pelements != NULL) {
1652 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1653 }
1654 } else {
1655 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1656 retval = true;
1657 }
1658 return retval;
1659 }
1660
1661 // check if addr is inside libjvm.so
1662 bool os::address_is_in_vm(address addr) {
1663 static address libjvm_base_addr;
1664 Dl_info dlinfo;
1665
1666 if (libjvm_base_addr == NULL) {
1667 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1668 libjvm_base_addr = (address)dlinfo.dli_fbase;
1669 }
1670 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1671 }
1672
1673 if (dladdr((void *)addr, &dlinfo) != 0) {
1674 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1675 }
1676
1677 return false;
1678 }
1679
1680 bool os::dll_address_to_function_name(address addr, char *buf,
1681 int buflen, int *offset) {
1682 // buf is not optional, but offset is optional
1683 assert(buf != NULL, "sanity check");
1684
1685 Dl_info dlinfo;
1686
1687 if (dladdr((void*)addr, &dlinfo) != 0) {
1688 // see if we have a matching symbol
1689 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1690 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1691 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1692 }
1693 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1694 return true;
1695 }
1696 // no matching symbol so try for just file info
1697 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1698 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1699 buf, buflen, offset, dlinfo.dli_fname)) {
1700 return true;
1701 }
1702 }
1703 }
1704
1705 buf[0] = '\0';
1706 if (offset != NULL) *offset = -1;
1707 return false;
1708 }
1709
1710 struct _address_to_library_name {
1711 address addr; // input : memory address
1712 size_t buflen; // size of fname
1713 char* fname; // output: library name
1714 address base; // library base addr
1715 };
1716
1717 static int address_to_library_name_callback(struct dl_phdr_info *info,
1718 size_t size, void *data) {
1719 int i;
1720 bool found = false;
1721 address libbase = NULL;
1722 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1723
1724 // iterate through all loadable segments
1725 for (i = 0; i < info->dlpi_phnum; i++) {
1726 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1727 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1728 // base address of a library is the lowest address of its loaded
1729 // segments.
1730 if (libbase == NULL || libbase > segbase) {
1731 libbase = segbase;
1732 }
1733 // see if 'addr' is within current segment
1734 if (segbase <= d->addr &&
1735 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1736 found = true;
1737 }
1738 }
1739 }
1740
1741 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1742 // so dll_address_to_library_name() can fall through to use dladdr() which
1743 // can figure out executable name from argv[0].
1744 if (found && info->dlpi_name && info->dlpi_name[0]) {
1745 d->base = libbase;
1746 if (d->fname) {
1747 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1748 }
1749 return 1;
1750 }
1751 return 0;
1752 }
1753
1754 bool os::dll_address_to_library_name(address addr, char* buf,
1755 int buflen, int* offset) {
1756 // buf is not optional, but offset is optional
1757 assert(buf != NULL, "sanity check");
1758
1759 Dl_info dlinfo;
1760 struct _address_to_library_name data;
1761
1762 // There is a bug in old glibc dladdr() implementation that it could resolve
1763 // to wrong library name if the .so file has a base address != NULL. Here
1764 // we iterate through the program headers of all loaded libraries to find
1765 // out which library 'addr' really belongs to. This workaround can be
1766 // removed once the minimum requirement for glibc is moved to 2.3.x.
1767 data.addr = addr;
1768 data.fname = buf;
1769 data.buflen = buflen;
1770 data.base = NULL;
1771 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1772
1773 if (rslt) {
1774 // buf already contains library name
1775 if (offset) *offset = addr - data.base;
1776 return true;
1777 }
1778 if (dladdr((void*)addr, &dlinfo) != 0) {
1779 if (dlinfo.dli_fname != NULL) {
1780 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1781 }
1782 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1783 *offset = addr - (address)dlinfo.dli_fbase;
1784 }
1785 return true;
1786 }
1787
1788 buf[0] = '\0';
1789 if (offset) *offset = -1;
1790 return false;
1791 }
1792
1793 // Loads .dll/.so and
1794 // in case of error it checks if .dll/.so was built for the
1795 // same architecture as Hotspot is running on
1796
1797
1798 // Remember the stack's state. The Linux dynamic linker will change
1799 // the stack to 'executable' at most once, so we must safepoint only once.
1800 bool os::Linux::_stack_is_executable = false;
1801
1802 // VM operation that loads a library. This is necessary if stack protection
1803 // of the Java stacks can be lost during loading the library. If we
1804 // do not stop the Java threads, they can stack overflow before the stacks
1805 // are protected again.
1806 class VM_LinuxDllLoad: public VM_Operation {
1807 private:
1808 const char *_filename;
1809 char *_ebuf;
1810 int _ebuflen;
1811 void *_lib;
1812 public:
1813 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1814 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1815 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1816 void doit() {
1817 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1818 os::Linux::_stack_is_executable = true;
1819 }
1820 void* loaded_library() { return _lib; }
1821 };
1822
1823 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1824 void * result = NULL;
1825 bool load_attempted = false;
1826
1827 // Check whether the library to load might change execution rights
1828 // of the stack. If they are changed, the protection of the stack
1829 // guard pages will be lost. We need a safepoint to fix this.
1830 //
1831 // See Linux man page execstack(8) for more info.
1832 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1833 ElfFile ef(filename);
1834 if (!ef.specifies_noexecstack()) {
1835 if (!is_init_completed()) {
1836 os::Linux::_stack_is_executable = true;
1837 // This is OK - No Java threads have been created yet, and hence no
1838 // stack guard pages to fix.
1839 //
1840 // This should happen only when you are building JDK7 using a very
1841 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1842 //
1843 // Dynamic loader will make all stacks executable after
1844 // this function returns, and will not do that again.
1845 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1846 } else {
1847 warning("You have loaded library %s which might have disabled stack guard. "
1848 "The VM will try to fix the stack guard now.\n"
1849 "It's highly recommended that you fix the library with "
1850 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1851 filename);
1852
1853 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1854 JavaThread *jt = JavaThread::current();
1855 if (jt->thread_state() != _thread_in_native) {
1856 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1857 // that requires ExecStack. Cannot enter safe point. Let's give up.
1858 warning("Unable to fix stack guard. Giving up.");
1859 } else {
1860 if (!LoadExecStackDllInVMThread) {
1861 // This is for the case where the DLL has an static
1862 // constructor function that executes JNI code. We cannot
1863 // load such DLLs in the VMThread.
1864 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1865 }
1866
1867 ThreadInVMfromNative tiv(jt);
1868 debug_only(VMNativeEntryWrapper vew;)
1869
1870 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1871 VMThread::execute(&op);
1872 if (LoadExecStackDllInVMThread) {
1873 result = op.loaded_library();
1874 }
1875 load_attempted = true;
1876 }
1877 }
1878 }
1879 }
1880
1881 if (!load_attempted) {
1882 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1883 }
1884
1885 if (result != NULL) {
1886 // Successful loading
1887 return result;
1888 }
1889
1890 Elf32_Ehdr elf_head;
1891 int diag_msg_max_length=ebuflen-strlen(ebuf);
1892 char* diag_msg_buf=ebuf+strlen(ebuf);
1893
1894 if (diag_msg_max_length==0) {
1895 // No more space in ebuf for additional diagnostics message
1896 return NULL;
1897 }
1898
1899
1900 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1901
1902 if (file_descriptor < 0) {
1903 // Can't open library, report dlerror() message
1904 return NULL;
1905 }
1906
1907 bool failed_to_read_elf_head=
1908 (sizeof(elf_head)!=
1909 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1910
1911 ::close(file_descriptor);
1912 if (failed_to_read_elf_head) {
1913 // file i/o error - report dlerror() msg
1914 return NULL;
1915 }
1916
1917 typedef struct {
1918 Elf32_Half code; // Actual value as defined in elf.h
1919 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1920 char elf_class; // 32 or 64 bit
1921 char endianess; // MSB or LSB
1922 char* name; // String representation
1923 } arch_t;
1924
1925 #ifndef EM_486
1926 #define EM_486 6 /* Intel 80486 */
1927 #endif
1928
1929 static const arch_t arch_array[]={
1930 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1931 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1932 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1933 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1934 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1935 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1936 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1937 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1938 #if defined(VM_LITTLE_ENDIAN)
1939 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1940 #else
1941 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1942 #endif
1943 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1944 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1945 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1946 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1947 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1948 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1949 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1950 };
1951
1952 #if (defined IA32)
1953 static Elf32_Half running_arch_code=EM_386;
1954 #elif (defined AMD64)
1955 static Elf32_Half running_arch_code=EM_X86_64;
1956 #elif (defined IA64)
1957 static Elf32_Half running_arch_code=EM_IA_64;
1958 #elif (defined __sparc) && (defined _LP64)
1959 static Elf32_Half running_arch_code=EM_SPARCV9;
1960 #elif (defined __sparc) && (!defined _LP64)
1961 static Elf32_Half running_arch_code=EM_SPARC;
1962 #elif (defined __powerpc64__)
1963 static Elf32_Half running_arch_code=EM_PPC64;
1964 #elif (defined __powerpc__)
1965 static Elf32_Half running_arch_code=EM_PPC;
1966 #elif (defined ARM)
1967 static Elf32_Half running_arch_code=EM_ARM;
1968 #elif (defined S390)
1969 static Elf32_Half running_arch_code=EM_S390;
1970 #elif (defined ALPHA)
1971 static Elf32_Half running_arch_code=EM_ALPHA;
1972 #elif (defined MIPSEL)
1973 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1974 #elif (defined PARISC)
1975 static Elf32_Half running_arch_code=EM_PARISC;
1976 #elif (defined MIPS)
1977 static Elf32_Half running_arch_code=EM_MIPS;
1978 #elif (defined M68K)
1979 static Elf32_Half running_arch_code=EM_68K;
1980 #else
1981 #error Method os::dll_load requires that one of following is defined:\
1982 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1983 #endif
1984
1985 // Identify compatability class for VM's architecture and library's architecture
1986 // Obtain string descriptions for architectures
1987
1988 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1989 int running_arch_index=-1;
1990
1991 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1992 if (running_arch_code == arch_array[i].code) {
1993 running_arch_index = i;
1994 }
1995 if (lib_arch.code == arch_array[i].code) {
1996 lib_arch.compat_class = arch_array[i].compat_class;
1997 lib_arch.name = arch_array[i].name;
1998 }
1999 }
2000
2001 assert(running_arch_index != -1,
2002 "Didn't find running architecture code (running_arch_code) in arch_array");
2003 if (running_arch_index == -1) {
2004 // Even though running architecture detection failed
2005 // we may still continue with reporting dlerror() message
2006 return NULL;
2007 }
2008
2009 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2010 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2011 return NULL;
2012 }
2013
2014 #ifndef S390
2015 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2016 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2017 return NULL;
2018 }
2019 #endif // !S390
2020
2021 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2022 if (lib_arch.name!=NULL) {
2023 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2024 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2025 lib_arch.name, arch_array[running_arch_index].name);
2026 } else {
2027 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2028 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2029 lib_arch.code,
2030 arch_array[running_arch_index].name);
2031 }
2032 }
2033
2034 return NULL;
2035 }
2036
2037 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
2038 int ebuflen) {
2039 void * result = ::dlopen(filename, RTLD_LAZY);
2040 if (result == NULL) {
2041 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2042 ebuf[ebuflen-1] = '\0';
2043 }
2044 return result;
2045 }
2046
2047 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2048 int ebuflen) {
2049 void * result = NULL;
2050 if (LoadExecStackDllInVMThread) {
2051 result = dlopen_helper(filename, ebuf, ebuflen);
2052 }
2053
2054 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2055 // library that requires an executable stack, or which does not have this
2056 // stack attribute set, dlopen changes the stack attribute to executable. The
2057 // read protection of the guard pages gets lost.
2058 //
2059 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2060 // may have been queued at the same time.
2061
2062 if (!_stack_is_executable) {
2063 JavaThread *jt = Threads::first();
2064
2065 while (jt) {
2066 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2067 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2068 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2069 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2070 warning("Attempt to reguard stack yellow zone failed.");
2071 }
2072 }
2073 jt = jt->next();
2074 }
2075 }
2076
2077 return result;
2078 }
2079
2080 // glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2081 // chances are you might want to run the generated bits against glibc-2.0
2082 // libdl.so, so always use locking for any version of glibc.
2083 //
2084 void* os::dll_lookup(void* handle, const char* name) {
2085 pthread_mutex_lock(&dl_mutex);
2086 void* res = dlsym(handle, name);
2087 pthread_mutex_unlock(&dl_mutex);
2088 return res;
2089 }
2090
2091 void* os::get_default_process_handle() {
2092 return (void*)::dlopen(NULL, RTLD_LAZY);
2093 }
2094
2095 static bool _print_ascii_file(const char* filename, outputStream* st) {
2096 int fd = ::open(filename, O_RDONLY);
2097 if (fd == -1) {
2098 return false;
2099 }
2100
2101 char buf[32];
2102 int bytes;
2103 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2104 st->print_raw(buf, bytes);
2105 }
2106
2107 ::close(fd);
2108
2109 return true;
2110 }
2111
2112 void os::print_dll_info(outputStream *st) {
2113 st->print_cr("Dynamic libraries:");
2114
2115 char fname[32];
2116 pid_t pid = os::Linux::gettid();
2117
2118 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2119
2120 if (!_print_ascii_file(fname, st)) {
2121 st->print("Can not get library information for pid = %d\n", pid);
2122 }
2123 }
2124
2125 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2126 FILE *procmapsFile = NULL;
2127
2128 // Open the procfs maps file for the current process
2129 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2130 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2131 char line[PATH_MAX + 100];
2132
2133 // Read line by line from 'file'
2134 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2135 u8 base, top, offset, inode;
2136 char permissions[5];
2137 char device[6];
2138 char name[PATH_MAX + 1];
2139
2140 // Parse fields from line
2141 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2142
2143 // Filter by device id '00:00' so that we only get file system mapped files.
2144 if (strcmp(device, "00:00") != 0) {
2145
2146 // Call callback with the fields of interest
2147 if(callback(name, (address)base, (address)top, param)) {
2148 // Oops abort, callback aborted
2149 fclose(procmapsFile);
2150 return 1;
2151 }
2152 }
2153 }
2154 fclose(procmapsFile);
2155 }
2156 return 0;
2157 }
2158
2159 void os::print_os_info_brief(outputStream* st) {
2160 os::Linux::print_distro_info(st);
2161
2162 os::Posix::print_uname_info(st);
2163
2164 os::Linux::print_libversion_info(st);
2165
2166 }
2167
2168 void os::print_os_info(outputStream* st) {
2169 st->print("OS:");
2170
2171 os::Linux::print_distro_info(st);
2172
2173 os::Posix::print_uname_info(st);
2174
2175 // Print warning if unsafe chroot environment detected
2176 if (unsafe_chroot_detected) {
2177 st->print("WARNING!! ");
2178 st->print_cr("%s", unstable_chroot_error);
2179 }
2180
2181 os::Linux::print_libversion_info(st);
2182
2183 os::Posix::print_rlimit_info(st);
2184
2185 os::Posix::print_load_average(st);
2186
2187 os::Linux::print_full_memory_info(st);
2188 }
2189
2190 // Try to identify popular distros.
2191 // Most Linux distributions have a /etc/XXX-release file, which contains
2192 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2193 // file that also contains the OS version string. Some have more than one
2194 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2195 // /etc/redhat-release.), so the order is important.
2196 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2197 // their own specific XXX-release file as well as a redhat-release file.
2198 // Because of this the XXX-release file needs to be searched for before the
2199 // redhat-release file.
2200 // Since Red Hat has a lsb-release file that is not very descriptive the
2201 // search for redhat-release needs to be before lsb-release.
2202 // Since the lsb-release file is the new standard it needs to be searched
2203 // before the older style release files.
2204 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2205 // next to last resort. The os-release file is a new standard that contains
2206 // distribution information and the system-release file seems to be an old
2207 // standard that has been replaced by the lsb-release and os-release files.
2208 // Searching for the debian_version file is the last resort. It contains
2209 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2210 // "Debian " is printed before the contents of the debian_version file.
2211 void os::Linux::print_distro_info(outputStream* st) {
2212 if (!_print_ascii_file("/etc/oracle-release", st) &&
2213 !_print_ascii_file("/etc/mandriva-release", st) &&
2214 !_print_ascii_file("/etc/mandrake-release", st) &&
2215 !_print_ascii_file("/etc/sun-release", st) &&
2216 !_print_ascii_file("/etc/redhat-release", st) &&
2217 !_print_ascii_file("/etc/lsb-release", st) &&
2218 !_print_ascii_file("/etc/SuSE-release", st) &&
2219 !_print_ascii_file("/etc/turbolinux-release", st) &&
2220 !_print_ascii_file("/etc/gentoo-release", st) &&
2221 !_print_ascii_file("/etc/ltib-release", st) &&
2222 !_print_ascii_file("/etc/angstrom-version", st) &&
2223 !_print_ascii_file("/etc/system-release", st) &&
2224 !_print_ascii_file("/etc/os-release", st)) {
2225
2226 if (file_exists("/etc/debian_version")) {
2227 st->print("Debian ");
2228 _print_ascii_file("/etc/debian_version", st);
2229 } else {
2230 st->print("Linux");
2231 }
2232 }
2233 st->cr();
2234 }
2235
2236 void os::Linux::print_libversion_info(outputStream* st) {
2237 // libc, pthread
2238 st->print("libc:");
2239 st->print("%s ", os::Linux::glibc_version());
2240 st->print("%s ", os::Linux::libpthread_version());
2241 if (os::Linux::is_LinuxThreads()) {
2242 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2243 }
2244 st->cr();
2245 }
2246
2247 void os::Linux::print_full_memory_info(outputStream* st) {
2248 st->print("\n/proc/meminfo:\n");
2249 _print_ascii_file("/proc/meminfo", st);
2250 st->cr();
2251 }
2252
2253 void os::print_memory_info(outputStream* st) {
2254
2255 st->print("Memory:");
2256 st->print(" %dk page", os::vm_page_size()>>10);
2257
2258 // values in struct sysinfo are "unsigned long"
2259 struct sysinfo si;
2260 sysinfo(&si);
2261
2262 st->print(", physical " UINT64_FORMAT "k",
2263 os::physical_memory() >> 10);
2264 st->print("(" UINT64_FORMAT "k free)",
2265 os::available_memory() >> 10);
2266 st->print(", swap " UINT64_FORMAT "k",
2267 ((jlong)si.totalswap * si.mem_unit) >> 10);
2268 st->print("(" UINT64_FORMAT "k free)",
2269 ((jlong)si.freeswap * si.mem_unit) >> 10);
2270 st->cr();
2271 }
2272
2273 void os::pd_print_cpu_info(outputStream* st) {
2274 st->print("\n/proc/cpuinfo:\n");
2275 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2276 st->print(" <Not Available>");
2277 }
2278 st->cr();
2279 }
2280
2281 void os::print_siginfo(outputStream* st, void* siginfo) {
2282 const siginfo_t* si = (const siginfo_t*)siginfo;
2283
2284 os::Posix::print_siginfo_brief(st, si);
2285 #if INCLUDE_CDS
2286 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2287 UseSharedSpaces) {
2288 FileMapInfo* mapinfo = FileMapInfo::current_info();
2289 if (mapinfo->is_in_shared_space(si->si_addr)) {
2290 st->print("\n\nError accessing class data sharing archive." \
2291 " Mapped file inaccessible during execution, " \
2292 " possible disk/network problem.");
2293 }
2294 }
2295 #endif
2296 st->cr();
2297 }
2298
2299
2300 static void print_signal_handler(outputStream* st, int sig,
2301 char* buf, size_t buflen);
2302
2303 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2304 st->print_cr("Signal Handlers:");
2305 print_signal_handler(st, SIGSEGV, buf, buflen);
2306 print_signal_handler(st, SIGBUS , buf, buflen);
2307 print_signal_handler(st, SIGFPE , buf, buflen);
2308 print_signal_handler(st, SIGPIPE, buf, buflen);
2309 print_signal_handler(st, SIGXFSZ, buf, buflen);
2310 print_signal_handler(st, SIGILL , buf, buflen);
2311 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2312 print_signal_handler(st, SR_signum, buf, buflen);
2313 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2314 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2315 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2316 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2317 #if defined(PPC64)
2318 print_signal_handler(st, SIGTRAP, buf, buflen);
2319 #endif
2320 }
2321
2322 static char saved_jvm_path[MAXPATHLEN] = {0};
2323
2324 // Find the full path to the current module, libjvm.so
2325 void os::jvm_path(char *buf, jint buflen) {
2326 // Error checking.
2327 if (buflen < MAXPATHLEN) {
2328 assert(false, "must use a large-enough buffer");
2329 buf[0] = '\0';
2330 return;
2331 }
2332 // Lazy resolve the path to current module.
2333 if (saved_jvm_path[0] != 0) {
2334 strcpy(buf, saved_jvm_path);
2335 return;
2336 }
2337
2338 char dli_fname[MAXPATHLEN];
2339 bool ret = dll_address_to_library_name(
2340 CAST_FROM_FN_PTR(address, os::jvm_path),
2341 dli_fname, sizeof(dli_fname), NULL);
2342 assert(ret, "cannot locate libjvm");
2343 char *rp = NULL;
2344 if (ret && dli_fname[0] != '\0') {
2345 rp = realpath(dli_fname, buf);
2346 }
2347 if (rp == NULL) {
2348 return;
2349 }
2350
2351 if (Arguments::sun_java_launcher_is_altjvm()) {
2352 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2353 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2354 // If "/jre/lib/" appears at the right place in the string, then
2355 // assume we are installed in a JDK and we're done. Otherwise, check
2356 // for a JAVA_HOME environment variable and fix up the path so it
2357 // looks like libjvm.so is installed there (append a fake suffix
2358 // hotspot/libjvm.so).
2359 const char *p = buf + strlen(buf) - 1;
2360 for (int count = 0; p > buf && count < 5; ++count) {
2361 for (--p; p > buf && *p != '/'; --p)
2362 /* empty */ ;
2363 }
2364
2365 if (strncmp(p, "/jre/lib/", 9) != 0) {
2366 // Look for JAVA_HOME in the environment.
2367 char* java_home_var = ::getenv("JAVA_HOME");
2368 if (java_home_var != NULL && java_home_var[0] != 0) {
2369 char* jrelib_p;
2370 int len;
2371
2372 // Check the current module name "libjvm.so".
2373 p = strrchr(buf, '/');
2374 assert(strstr(p, "/libjvm") == p, "invalid library name");
2375
2376 rp = realpath(java_home_var, buf);
2377 if (rp == NULL) {
2378 return;
2379 }
2380
2381 // determine if this is a legacy image or modules image
2382 // modules image doesn't have "jre" subdirectory
2383 len = strlen(buf);
2384 assert(len < buflen, "Ran out of buffer room");
2385 jrelib_p = buf + len;
2386 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2387 if (0 != access(buf, F_OK)) {
2388 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2389 }
2390
2391 if (0 == access(buf, F_OK)) {
2392 // Use current module name "libjvm.so"
2393 len = strlen(buf);
2394 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2395 } else {
2396 // Go back to path of .so
2397 rp = realpath(dli_fname, buf);
2398 if (rp == NULL) {
2399 return;
2400 }
2401 }
2402 }
2403 }
2404 }
2405
2406 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2407 }
2408
2409 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2410 // no prefix required, not even "_"
2411 }
2412
2413 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2414 // no suffix required
2415 }
2416
2417 ////////////////////////////////////////////////////////////////////////////////
2418 // sun.misc.Signal support
2419
2420 static volatile jint sigint_count = 0;
2421
2422 static void UserHandler(int sig, void *siginfo, void *context) {
2423 // 4511530 - sem_post is serialized and handled by the manager thread. When
2424 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2425 // don't want to flood the manager thread with sem_post requests.
2426 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2427 return;
2428 }
2429
2430 // Ctrl-C is pressed during error reporting, likely because the error
2431 // handler fails to abort. Let VM die immediately.
2432 if (sig == SIGINT && is_error_reported()) {
2433 os::die();
2434 }
2435
2436 os::signal_notify(sig);
2437 }
2438
2439 void* os::user_handler() {
2440 return CAST_FROM_FN_PTR(void*, UserHandler);
2441 }
2442
2443 class Semaphore : public StackObj {
2444 public:
2445 Semaphore();
2446 ~Semaphore();
2447 void signal();
2448 void wait();
2449 bool trywait();
2450 bool timedwait(unsigned int sec, int nsec);
2451 private:
2452 sem_t _semaphore;
2453 };
2454
2455 Semaphore::Semaphore() {
2456 sem_init(&_semaphore, 0, 0);
2457 }
2458
2459 Semaphore::~Semaphore() {
2460 sem_destroy(&_semaphore);
2461 }
2462
2463 void Semaphore::signal() {
2464 sem_post(&_semaphore);
2465 }
2466
2467 void Semaphore::wait() {
2468 sem_wait(&_semaphore);
2469 }
2470
2471 bool Semaphore::trywait() {
2472 return sem_trywait(&_semaphore) == 0;
2473 }
2474
2475 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2476
2477 struct timespec ts;
2478 // Semaphore's are always associated with CLOCK_REALTIME
2479 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2480 // see unpackTime for discussion on overflow checking
2481 if (sec >= MAX_SECS) {
2482 ts.tv_sec += MAX_SECS;
2483 ts.tv_nsec = 0;
2484 } else {
2485 ts.tv_sec += sec;
2486 ts.tv_nsec += nsec;
2487 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2488 ts.tv_nsec -= NANOSECS_PER_SEC;
2489 ++ts.tv_sec; // note: this must be <= max_secs
2490 }
2491 }
2492
2493 while (1) {
2494 int result = sem_timedwait(&_semaphore, &ts);
2495 if (result == 0) {
2496 return true;
2497 } else if (errno == EINTR) {
2498 continue;
2499 } else if (errno == ETIMEDOUT) {
2500 return false;
2501 } else {
2502 return false;
2503 }
2504 }
2505 }
2506
2507 extern "C" {
2508 typedef void (*sa_handler_t)(int);
2509 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2510 }
2511
2512 void* os::signal(int signal_number, void* handler) {
2513 struct sigaction sigAct, oldSigAct;
2514
2515 sigfillset(&(sigAct.sa_mask));
2516 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2517 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2518
2519 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2520 // -1 means registration failed
2521 return (void *)-1;
2522 }
2523
2524 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2525 }
2526
2527 void os::signal_raise(int signal_number) {
2528 ::raise(signal_number);
2529 }
2530
2531 // The following code is moved from os.cpp for making this
2532 // code platform specific, which it is by its very nature.
2533
2534 // Will be modified when max signal is changed to be dynamic
2535 int os::sigexitnum_pd() {
2536 return NSIG;
2537 }
2538
2539 // a counter for each possible signal value
2540 static volatile jint pending_signals[NSIG+1] = { 0 };
2541
2542 // Linux(POSIX) specific hand shaking semaphore.
2543 static sem_t sig_sem;
2544 static Semaphore sr_semaphore;
2545
2546 void os::signal_init_pd() {
2547 // Initialize signal structures
2548 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2549
2550 // Initialize signal semaphore
2551 ::sem_init(&sig_sem, 0, 0);
2552 }
2553
2554 void os::signal_notify(int sig) {
2555 Atomic::inc(&pending_signals[sig]);
2556 ::sem_post(&sig_sem);
2557 }
2558
2559 static int check_pending_signals(bool wait) {
2560 Atomic::store(0, &sigint_count);
2561 for (;;) {
2562 for (int i = 0; i < NSIG + 1; i++) {
2563 jint n = pending_signals[i];
2564 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2565 return i;
2566 }
2567 }
2568 if (!wait) {
2569 return -1;
2570 }
2571 JavaThread *thread = JavaThread::current();
2572 ThreadBlockInVM tbivm(thread);
2573
2574 bool threadIsSuspended;
2575 do {
2576 thread->set_suspend_equivalent();
2577 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2578 ::sem_wait(&sig_sem);
2579
2580 // were we externally suspended while we were waiting?
2581 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2582 if (threadIsSuspended) {
2583 // The semaphore has been incremented, but while we were waiting
2584 // another thread suspended us. We don't want to continue running
2585 // while suspended because that would surprise the thread that
2586 // suspended us.
2587 ::sem_post(&sig_sem);
2588
2589 thread->java_suspend_self();
2590 }
2591 } while (threadIsSuspended);
2592 }
2593 }
2594
2595 int os::signal_lookup() {
2596 return check_pending_signals(false);
2597 }
2598
2599 int os::signal_wait() {
2600 return check_pending_signals(true);
2601 }
2602
2603 ////////////////////////////////////////////////////////////////////////////////
2604 // Virtual Memory
2605
2606 int os::vm_page_size() {
2607 // Seems redundant as all get out
2608 assert(os::Linux::page_size() != -1, "must call os::init");
2609 return os::Linux::page_size();
2610 }
2611
2612 // Solaris allocates memory by pages.
2613 int os::vm_allocation_granularity() {
2614 assert(os::Linux::page_size() != -1, "must call os::init");
2615 return os::Linux::page_size();
2616 }
2617
2618 // Rationale behind this function:
2619 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2620 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2621 // samples for JITted code. Here we create private executable mapping over the code cache
2622 // and then we can use standard (well, almost, as mapping can change) way to provide
2623 // info for the reporting script by storing timestamp and location of symbol
2624 void linux_wrap_code(char* base, size_t size) {
2625 static volatile jint cnt = 0;
2626
2627 if (!UseOprofile) {
2628 return;
2629 }
2630
2631 char buf[PATH_MAX+1];
2632 int num = Atomic::add(1, &cnt);
2633
2634 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2635 os::get_temp_directory(), os::current_process_id(), num);
2636 unlink(buf);
2637
2638 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2639
2640 if (fd != -1) {
2641 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2642 if (rv != (off_t)-1) {
2643 if (::write(fd, "", 1) == 1) {
2644 mmap(base, size,
2645 PROT_READ|PROT_WRITE|PROT_EXEC,
2646 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2647 }
2648 }
2649 ::close(fd);
2650 unlink(buf);
2651 }
2652 }
2653
2654 static bool recoverable_mmap_error(int err) {
2655 // See if the error is one we can let the caller handle. This
2656 // list of errno values comes from JBS-6843484. I can't find a
2657 // Linux man page that documents this specific set of errno
2658 // values so while this list currently matches Solaris, it may
2659 // change as we gain experience with this failure mode.
2660 switch (err) {
2661 case EBADF:
2662 case EINVAL:
2663 case ENOTSUP:
2664 // let the caller deal with these errors
2665 return true;
2666
2667 default:
2668 // Any remaining errors on this OS can cause our reserved mapping
2669 // to be lost. That can cause confusion where different data
2670 // structures think they have the same memory mapped. The worst
2671 // scenario is if both the VM and a library think they have the
2672 // same memory mapped.
2673 return false;
2674 }
2675 }
2676
2677 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2678 int err) {
2679 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2680 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2681 strerror(err), err);
2682 }
2683
2684 static void warn_fail_commit_memory(char* addr, size_t size,
2685 size_t alignment_hint, bool exec,
2686 int err) {
2687 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2688 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2689 alignment_hint, exec, strerror(err), err);
2690 }
2691
2692 // NOTE: Linux kernel does not really reserve the pages for us.
2693 // All it does is to check if there are enough free pages
2694 // left at the time of mmap(). This could be a potential
2695 // problem.
2696 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2697 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2698 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2699 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2700 if (res != (uintptr_t) MAP_FAILED) {
2701 if (UseNUMAInterleaving) {
2702 numa_make_global(addr, size);
2703 }
2704 return 0;
2705 }
2706
2707 int err = errno; // save errno from mmap() call above
2708
2709 if (!recoverable_mmap_error(err)) {
2710 warn_fail_commit_memory(addr, size, exec, err);
2711 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2712 }
2713
2714 return err;
2715 }
2716
2717 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2718 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2719 }
2720
2721 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2722 const char* mesg) {
2723 assert(mesg != NULL, "mesg must be specified");
2724 int err = os::Linux::commit_memory_impl(addr, size, exec);
2725 if (err != 0) {
2726 // the caller wants all commit errors to exit with the specified mesg:
2727 warn_fail_commit_memory(addr, size, exec, err);
2728 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2729 }
2730 }
2731
2732 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2733 #ifndef MAP_HUGETLB
2734 #define MAP_HUGETLB 0x40000
2735 #endif
2736
2737 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2738 #ifndef MADV_HUGEPAGE
2739 #define MADV_HUGEPAGE 14
2740 #endif
2741
2742 int os::Linux::commit_memory_impl(char* addr, size_t size,
2743 size_t alignment_hint, bool exec) {
2744 int err = os::Linux::commit_memory_impl(addr, size, exec);
2745 if (err == 0) {
2746 realign_memory(addr, size, alignment_hint);
2747 }
2748 return err;
2749 }
2750
2751 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2752 bool exec) {
2753 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2754 }
2755
2756 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2757 size_t alignment_hint, bool exec,
2758 const char* mesg) {
2759 assert(mesg != NULL, "mesg must be specified");
2760 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2761 if (err != 0) {
2762 // the caller wants all commit errors to exit with the specified mesg:
2763 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2764 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2765 }
2766 }
2767
2768 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2769 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2770 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2771 // be supported or the memory may already be backed by huge pages.
2772 ::madvise(addr, bytes, MADV_HUGEPAGE);
2773 }
2774 }
2775
2776 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2777 // This method works by doing an mmap over an existing mmaping and effectively discarding
2778 // the existing pages. However it won't work for SHM-based large pages that cannot be
2779 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2780 // small pages on top of the SHM segment. This method always works for small pages, so we
2781 // allow that in any case.
2782 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2783 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2784 }
2785 }
2786
2787 void os::numa_make_global(char *addr, size_t bytes) {
2788 Linux::numa_interleave_memory(addr, bytes);
2789 }
2790
2791 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2792 // bind policy to MPOL_PREFERRED for the current thread.
2793 #define USE_MPOL_PREFERRED 0
2794
2795 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2796 // To make NUMA and large pages more robust when both enabled, we need to ease
2797 // the requirements on where the memory should be allocated. MPOL_BIND is the
2798 // default policy and it will force memory to be allocated on the specified
2799 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2800 // the specified node, but will not force it. Using this policy will prevent
2801 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2802 // free large pages.
2803 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2804 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2805 }
2806
2807 bool os::numa_topology_changed() { return false; }
2808
2809 size_t os::numa_get_groups_num() {
2810 int max_node = Linux::numa_max_node();
2811 return max_node > 0 ? max_node + 1 : 1;
2812 }
2813
2814 int os::numa_get_group_id() {
2815 int cpu_id = Linux::sched_getcpu();
2816 if (cpu_id != -1) {
2817 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2818 if (lgrp_id != -1) {
2819 return lgrp_id;
2820 }
2821 }
2822 return 0;
2823 }
2824
2825 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2826 for (size_t i = 0; i < size; i++) {
2827 ids[i] = i;
2828 }
2829 return size;
2830 }
2831
2832 bool os::get_page_info(char *start, page_info* info) {
2833 return false;
2834 }
2835
2836 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2837 page_info* page_found) {
2838 return end;
2839 }
2840
2841
2842 int os::Linux::sched_getcpu_syscall(void) {
2843 unsigned int cpu;
2844 int retval = -1;
2845
2846 #if defined(IA32)
2847 #ifndef SYS_getcpu
2848 #define SYS_getcpu 318
2849 #endif
2850 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2851 #elif defined(AMD64)
2852 // Unfortunately we have to bring all these macros here from vsyscall.h
2853 // to be able to compile on old linuxes.
2854 #define __NR_vgetcpu 2
2855 #define VSYSCALL_START (-10UL << 20)
2856 #define VSYSCALL_SIZE 1024
2857 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2858 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2859 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2860 retval = vgetcpu(&cpu, NULL, NULL);
2861 #endif
2862
2863 return (retval == -1) ? retval : cpu;
2864 }
2865
2866 // Something to do with the numa-aware allocator needs these symbols
2867 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2868 extern "C" JNIEXPORT void numa_error(char *where) { }
2869 extern "C" JNIEXPORT int fork1() { return fork(); }
2870
2871
2872 // If we are running with libnuma version > 2, then we should
2873 // be trying to use symbols with versions 1.1
2874 // If we are running with earlier version, which did not have symbol versions,
2875 // we should use the base version.
2876 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2877 void *f = dlvsym(handle, name, "libnuma_1.1");
2878 if (f == NULL) {
2879 f = dlsym(handle, name);
2880 }
2881 return f;
2882 }
2883
2884 bool os::Linux::libnuma_init() {
2885 // sched_getcpu() should be in libc.
2886 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2887 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2888
2889 // If it's not, try a direct syscall.
2890 if (sched_getcpu() == -1) {
2891 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2892 (void*)&sched_getcpu_syscall));
2893 }
2894
2895 if (sched_getcpu() != -1) { // Does it work?
2896 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2897 if (handle != NULL) {
2898 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2899 libnuma_dlsym(handle, "numa_node_to_cpus")));
2900 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2901 libnuma_dlsym(handle, "numa_max_node")));
2902 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2903 libnuma_dlsym(handle, "numa_available")));
2904 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2905 libnuma_dlsym(handle, "numa_tonode_memory")));
2906 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2907 libnuma_dlsym(handle, "numa_interleave_memory")));
2908 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2909 libnuma_dlsym(handle, "numa_set_bind_policy")));
2910
2911
2912 if (numa_available() != -1) {
2913 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2914 // Create a cpu -> node mapping
2915 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2916 rebuild_cpu_to_node_map();
2917 return true;
2918 }
2919 }
2920 }
2921 return false;
2922 }
2923
2924 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2925 // The table is later used in get_node_by_cpu().
2926 void os::Linux::rebuild_cpu_to_node_map() {
2927 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2928 // in libnuma (possible values are starting from 16,
2929 // and continuing up with every other power of 2, but less
2930 // than the maximum number of CPUs supported by kernel), and
2931 // is a subject to change (in libnuma version 2 the requirements
2932 // are more reasonable) we'll just hardcode the number they use
2933 // in the library.
2934 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2935
2936 size_t cpu_num = os::active_processor_count();
2937 size_t cpu_map_size = NCPUS / BitsPerCLong;
2938 size_t cpu_map_valid_size =
2939 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2940
2941 cpu_to_node()->clear();
2942 cpu_to_node()->at_grow(cpu_num - 1);
2943 size_t node_num = numa_get_groups_num();
2944
2945 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2946 for (size_t i = 0; i < node_num; i++) {
2947 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2948 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2949 if (cpu_map[j] != 0) {
2950 for (size_t k = 0; k < BitsPerCLong; k++) {
2951 if (cpu_map[j] & (1UL << k)) {
2952 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2953 }
2954 }
2955 }
2956 }
2957 }
2958 }
2959 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2960 }
2961
2962 int os::Linux::get_node_by_cpu(int cpu_id) {
2963 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2964 return cpu_to_node()->at(cpu_id);
2965 }
2966 return -1;
2967 }
2968
2969 GrowableArray<int>* os::Linux::_cpu_to_node;
2970 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2971 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2972 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2973 os::Linux::numa_available_func_t os::Linux::_numa_available;
2974 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2975 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2976 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2977 unsigned long* os::Linux::_numa_all_nodes;
2978
2979 bool os::pd_uncommit_memory(char* addr, size_t size) {
2980 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2981 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2982 return res != (uintptr_t) MAP_FAILED;
2983 }
2984
2985 static address get_stack_commited_bottom(address bottom, size_t size) {
2986 address nbot = bottom;
2987 address ntop = bottom + size;
2988
2989 size_t page_sz = os::vm_page_size();
2990 unsigned pages = size / page_sz;
2991
2992 unsigned char vec[1];
2993 unsigned imin = 1, imax = pages + 1, imid;
2994 int mincore_return_value = 0;
2995
2996 assert(imin <= imax, "Unexpected page size");
2997
2998 while (imin < imax) {
2999 imid = (imax + imin) / 2;
3000 nbot = ntop - (imid * page_sz);
3001
3002 // Use a trick with mincore to check whether the page is mapped or not.
3003 // mincore sets vec to 1 if page resides in memory and to 0 if page
3004 // is swapped output but if page we are asking for is unmapped
3005 // it returns -1,ENOMEM
3006 mincore_return_value = mincore(nbot, page_sz, vec);
3007
3008 if (mincore_return_value == -1) {
3009 // Page is not mapped go up
3010 // to find first mapped page
3011 if (errno != EAGAIN) {
3012 assert(errno == ENOMEM, "Unexpected mincore errno");
3013 imax = imid;
3014 }
3015 } else {
3016 // Page is mapped go down
3017 // to find first not mapped page
3018 imin = imid + 1;
3019 }
3020 }
3021
3022 nbot = nbot + page_sz;
3023
3024 // Adjust stack bottom one page up if last checked page is not mapped
3025 if (mincore_return_value == -1) {
3026 nbot = nbot + page_sz;
3027 }
3028
3029 return nbot;
3030 }
3031
3032
3033 // Linux uses a growable mapping for the stack, and if the mapping for
3034 // the stack guard pages is not removed when we detach a thread the
3035 // stack cannot grow beyond the pages where the stack guard was
3036 // mapped. If at some point later in the process the stack expands to
3037 // that point, the Linux kernel cannot expand the stack any further
3038 // because the guard pages are in the way, and a segfault occurs.
3039 //
3040 // However, it's essential not to split the stack region by unmapping
3041 // a region (leaving a hole) that's already part of the stack mapping,
3042 // so if the stack mapping has already grown beyond the guard pages at
3043 // the time we create them, we have to truncate the stack mapping.
3044 // So, we need to know the extent of the stack mapping when
3045 // create_stack_guard_pages() is called.
3046
3047 // We only need this for stacks that are growable: at the time of
3048 // writing thread stacks don't use growable mappings (i.e. those
3049 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3050 // only applies to the main thread.
3051
3052 // If the (growable) stack mapping already extends beyond the point
3053 // where we're going to put our guard pages, truncate the mapping at
3054 // that point by munmap()ping it. This ensures that when we later
3055 // munmap() the guard pages we don't leave a hole in the stack
3056 // mapping. This only affects the main/initial thread
3057
3058 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3059 if (os::Linux::is_initial_thread()) {
3060 // As we manually grow stack up to bottom inside create_attached_thread(),
3061 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3062 // we don't need to do anything special.
3063 // Check it first, before calling heavy function.
3064 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3065 unsigned char vec[1];
3066
3067 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3068 // Fallback to slow path on all errors, including EAGAIN
3069 stack_extent = (uintptr_t) get_stack_commited_bottom(
3070 os::Linux::initial_thread_stack_bottom(),
3071 (size_t)addr - stack_extent);
3072 }
3073
3074 if (stack_extent < (uintptr_t)addr) {
3075 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3076 }
3077 }
3078
3079 return os::commit_memory(addr, size, !ExecMem);
3080 }
3081
3082 // If this is a growable mapping, remove the guard pages entirely by
3083 // munmap()ping them. If not, just call uncommit_memory(). This only
3084 // affects the main/initial thread, but guard against future OS changes
3085 // It's safe to always unmap guard pages for initial thread because we
3086 // always place it right after end of the mapped region
3087
3088 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3089 uintptr_t stack_extent, stack_base;
3090
3091 if (os::Linux::is_initial_thread()) {
3092 return ::munmap(addr, size) == 0;
3093 }
3094
3095 return os::uncommit_memory(addr, size);
3096 }
3097
3098 static address _highest_vm_reserved_address = NULL;
3099
3100 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3101 // at 'requested_addr'. If there are existing memory mappings at the same
3102 // location, however, they will be overwritten. If 'fixed' is false,
3103 // 'requested_addr' is only treated as a hint, the return value may or
3104 // may not start from the requested address. Unlike Linux mmap(), this
3105 // function returns NULL to indicate failure.
3106 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3107 char * addr;
3108 int flags;
3109
3110 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3111 if (fixed) {
3112 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3113 flags |= MAP_FIXED;
3114 }
3115
3116 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3117 // touch an uncommitted page. Otherwise, the read/write might
3118 // succeed if we have enough swap space to back the physical page.
3119 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3120 flags, -1, 0);
3121
3122 if (addr != MAP_FAILED) {
3123 // anon_mmap() should only get called during VM initialization,
3124 // don't need lock (actually we can skip locking even it can be called
3125 // from multiple threads, because _highest_vm_reserved_address is just a
3126 // hint about the upper limit of non-stack memory regions.)
3127 if ((address)addr + bytes > _highest_vm_reserved_address) {
3128 _highest_vm_reserved_address = (address)addr + bytes;
3129 }
3130 }
3131
3132 return addr == MAP_FAILED ? NULL : addr;
3133 }
3134
3135 // Don't update _highest_vm_reserved_address, because there might be memory
3136 // regions above addr + size. If so, releasing a memory region only creates
3137 // a hole in the address space, it doesn't help prevent heap-stack collision.
3138 //
3139 static int anon_munmap(char * addr, size_t size) {
3140 return ::munmap(addr, size) == 0;
3141 }
3142
3143 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3144 size_t alignment_hint) {
3145 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3146 }
3147
3148 bool os::pd_release_memory(char* addr, size_t size) {
3149 return anon_munmap(addr, size);
3150 }
3151
3152 static address highest_vm_reserved_address() {
3153 return _highest_vm_reserved_address;
3154 }
3155
3156 static bool linux_mprotect(char* addr, size_t size, int prot) {
3157 // Linux wants the mprotect address argument to be page aligned.
3158 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3159
3160 // According to SUSv3, mprotect() should only be used with mappings
3161 // established by mmap(), and mmap() always maps whole pages. Unaligned
3162 // 'addr' likely indicates problem in the VM (e.g. trying to change
3163 // protection of malloc'ed or statically allocated memory). Check the
3164 // caller if you hit this assert.
3165 assert(addr == bottom, "sanity check");
3166
3167 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3168 return ::mprotect(bottom, size, prot) == 0;
3169 }
3170
3171 // Set protections specified
3172 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3173 bool is_committed) {
3174 unsigned int p = 0;
3175 switch (prot) {
3176 case MEM_PROT_NONE: p = PROT_NONE; break;
3177 case MEM_PROT_READ: p = PROT_READ; break;
3178 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3179 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3180 default:
3181 ShouldNotReachHere();
3182 }
3183 // is_committed is unused.
3184 return linux_mprotect(addr, bytes, p);
3185 }
3186
3187 bool os::guard_memory(char* addr, size_t size) {
3188 return linux_mprotect(addr, size, PROT_NONE);
3189 }
3190
3191 bool os::unguard_memory(char* addr, size_t size) {
3192 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3193 }
3194
3195 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3196 size_t page_size) {
3197 bool result = false;
3198 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3199 MAP_ANONYMOUS|MAP_PRIVATE,
3200 -1, 0);
3201 if (p != MAP_FAILED) {
3202 void *aligned_p = align_ptr_up(p, page_size);
3203
3204 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3205
3206 munmap(p, page_size * 2);
3207 }
3208
3209 if (warn && !result) {
3210 warning("TransparentHugePages is not supported by the operating system.");
3211 }
3212
3213 return result;
3214 }
3215
3216 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3217 bool result = false;
3218 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3219 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3220 -1, 0);
3221
3222 if (p != MAP_FAILED) {
3223 // We don't know if this really is a huge page or not.
3224 FILE *fp = fopen("/proc/self/maps", "r");
3225 if (fp) {
3226 while (!feof(fp)) {
3227 char chars[257];
3228 long x = 0;
3229 if (fgets(chars, sizeof(chars), fp)) {
3230 if (sscanf(chars, "%lx-%*x", &x) == 1
3231 && x == (long)p) {
3232 if (strstr (chars, "hugepage")) {
3233 result = true;
3234 break;
3235 }
3236 }
3237 }
3238 }
3239 fclose(fp);
3240 }
3241 munmap(p, page_size);
3242 }
3243
3244 if (warn && !result) {
3245 warning("HugeTLBFS is not supported by the operating system.");
3246 }
3247
3248 return result;
3249 }
3250
3251 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3252 //
3253 // From the coredump_filter documentation:
3254 //
3255 // - (bit 0) anonymous private memory
3256 // - (bit 1) anonymous shared memory
3257 // - (bit 2) file-backed private memory
3258 // - (bit 3) file-backed shared memory
3259 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3260 // effective only if the bit 2 is cleared)
3261 // - (bit 5) hugetlb private memory
3262 // - (bit 6) hugetlb shared memory
3263 //
3264 static void set_coredump_filter(void) {
3265 FILE *f;
3266 long cdm;
3267
3268 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3269 return;
3270 }
3271
3272 if (fscanf(f, "%lx", &cdm) != 1) {
3273 fclose(f);
3274 return;
3275 }
3276
3277 rewind(f);
3278
3279 if ((cdm & LARGEPAGES_BIT) == 0) {
3280 cdm |= LARGEPAGES_BIT;
3281 fprintf(f, "%#lx", cdm);
3282 }
3283
3284 fclose(f);
3285 }
3286
3287 // Large page support
3288
3289 static size_t _large_page_size = 0;
3290
3291 size_t os::Linux::find_large_page_size() {
3292 size_t large_page_size = 0;
3293
3294 // large_page_size on Linux is used to round up heap size. x86 uses either
3295 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3296 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3297 // page as large as 256M.
3298 //
3299 // Here we try to figure out page size by parsing /proc/meminfo and looking
3300 // for a line with the following format:
3301 // Hugepagesize: 2048 kB
3302 //
3303 // If we can't determine the value (e.g. /proc is not mounted, or the text
3304 // format has been changed), we'll use the largest page size supported by
3305 // the processor.
3306
3307 #ifndef ZERO
3308 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3309 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3310 #endif // ZERO
3311
3312 FILE *fp = fopen("/proc/meminfo", "r");
3313 if (fp) {
3314 while (!feof(fp)) {
3315 int x = 0;
3316 char buf[16];
3317 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3318 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3319 large_page_size = x * K;
3320 break;
3321 }
3322 } else {
3323 // skip to next line
3324 for (;;) {
3325 int ch = fgetc(fp);
3326 if (ch == EOF || ch == (int)'\n') break;
3327 }
3328 }
3329 }
3330 fclose(fp);
3331 }
3332
3333 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3334 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3335 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3336 proper_unit_for_byte_size(large_page_size));
3337 }
3338
3339 return large_page_size;
3340 }
3341
3342 size_t os::Linux::setup_large_page_size() {
3343 _large_page_size = Linux::find_large_page_size();
3344 const size_t default_page_size = (size_t)Linux::page_size();
3345 if (_large_page_size > default_page_size) {
3346 _page_sizes[0] = _large_page_size;
3347 _page_sizes[1] = default_page_size;
3348 _page_sizes[2] = 0;
3349 }
3350
3351 return _large_page_size;
3352 }
3353
3354 bool os::Linux::setup_large_page_type(size_t page_size) {
3355 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3356 FLAG_IS_DEFAULT(UseSHM) &&
3357 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3358
3359 // The type of large pages has not been specified by the user.
3360
3361 // Try UseHugeTLBFS and then UseSHM.
3362 UseHugeTLBFS = UseSHM = true;
3363
3364 // Don't try UseTransparentHugePages since there are known
3365 // performance issues with it turned on. This might change in the future.
3366 UseTransparentHugePages = false;
3367 }
3368
3369 if (UseTransparentHugePages) {
3370 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3371 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3372 UseHugeTLBFS = false;
3373 UseSHM = false;
3374 return true;
3375 }
3376 UseTransparentHugePages = false;
3377 }
3378
3379 if (UseHugeTLBFS) {
3380 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3381 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3382 UseSHM = false;
3383 return true;
3384 }
3385 UseHugeTLBFS = false;
3386 }
3387
3388 return UseSHM;
3389 }
3390
3391 void os::large_page_init() {
3392 if (!UseLargePages &&
3393 !UseTransparentHugePages &&
3394 !UseHugeTLBFS &&
3395 !UseSHM) {
3396 // Not using large pages.
3397 return;
3398 }
3399
3400 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3401 // The user explicitly turned off large pages.
3402 // Ignore the rest of the large pages flags.
3403 UseTransparentHugePages = false;
3404 UseHugeTLBFS = false;
3405 UseSHM = false;
3406 return;
3407 }
3408
3409 size_t large_page_size = Linux::setup_large_page_size();
3410 UseLargePages = Linux::setup_large_page_type(large_page_size);
3411
3412 set_coredump_filter();
3413 }
3414
3415 #ifndef SHM_HUGETLB
3416 #define SHM_HUGETLB 04000
3417 #endif
3418
3419 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3420 char* req_addr, bool exec) {
3421 // "exec" is passed in but not used. Creating the shared image for
3422 // the code cache doesn't have an SHM_X executable permission to check.
3423 assert(UseLargePages && UseSHM, "only for SHM large pages");
3424 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3425
3426 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3427 return NULL; // Fallback to small pages.
3428 }
3429
3430 key_t key = IPC_PRIVATE;
3431 char *addr;
3432
3433 bool warn_on_failure = UseLargePages &&
3434 (!FLAG_IS_DEFAULT(UseLargePages) ||
3435 !FLAG_IS_DEFAULT(UseSHM) ||
3436 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3437 char msg[128];
3438
3439 // Create a large shared memory region to attach to based on size.
3440 // Currently, size is the total size of the heap
3441 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3442 if (shmid == -1) {
3443 // Possible reasons for shmget failure:
3444 // 1. shmmax is too small for Java heap.
3445 // > check shmmax value: cat /proc/sys/kernel/shmmax
3446 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3447 // 2. not enough large page memory.
3448 // > check available large pages: cat /proc/meminfo
3449 // > increase amount of large pages:
3450 // echo new_value > /proc/sys/vm/nr_hugepages
3451 // Note 1: different Linux may use different name for this property,
3452 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3453 // Note 2: it's possible there's enough physical memory available but
3454 // they are so fragmented after a long run that they can't
3455 // coalesce into large pages. Try to reserve large pages when
3456 // the system is still "fresh".
3457 if (warn_on_failure) {
3458 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3459 warning("%s", msg);
3460 }
3461 return NULL;
3462 }
3463
3464 // attach to the region
3465 addr = (char*)shmat(shmid, req_addr, 0);
3466 int err = errno;
3467
3468 // Remove shmid. If shmat() is successful, the actual shared memory segment
3469 // will be deleted when it's detached by shmdt() or when the process
3470 // terminates. If shmat() is not successful this will remove the shared
3471 // segment immediately.
3472 shmctl(shmid, IPC_RMID, NULL);
3473
3474 if ((intptr_t)addr == -1) {
3475 if (warn_on_failure) {
3476 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3477 warning("%s", msg);
3478 }
3479 return NULL;
3480 }
3481
3482 return addr;
3483 }
3484
3485 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3486 int error) {
3487 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3488
3489 bool warn_on_failure = UseLargePages &&
3490 (!FLAG_IS_DEFAULT(UseLargePages) ||
3491 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3492 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3493
3494 if (warn_on_failure) {
3495 char msg[128];
3496 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3497 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3498 warning("%s", msg);
3499 }
3500 }
3501
3502 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3503 char* req_addr,
3504 bool exec) {
3505 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3506 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3507 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3508
3509 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3510 char* addr = (char*)::mmap(req_addr, bytes, prot,
3511 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3512 -1, 0);
3513
3514 if (addr == MAP_FAILED) {
3515 warn_on_large_pages_failure(req_addr, bytes, errno);
3516 return NULL;
3517 }
3518
3519 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3520
3521 return addr;
3522 }
3523
3524 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3525 size_t alignment,
3526 char* req_addr,
3527 bool exec) {
3528 size_t large_page_size = os::large_page_size();
3529
3530 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3531
3532 // Allocate small pages.
3533
3534 char* start;
3535 if (req_addr != NULL) {
3536 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3537 assert(is_size_aligned(bytes, alignment), "Must be");
3538 start = os::reserve_memory(bytes, req_addr);
3539 assert(start == NULL || start == req_addr, "Must be");
3540 } else {
3541 start = os::reserve_memory_aligned(bytes, alignment);
3542 }
3543
3544 if (start == NULL) {
3545 return NULL;
3546 }
3547
3548 assert(is_ptr_aligned(start, alignment), "Must be");
3549
3550 if (MemTracker::tracking_level() > NMT_minimal) {
3551 // os::reserve_memory_special will record this memory area.
3552 // Need to release it here to prevent overlapping reservations.
3553 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3554 tkr.record((address)start, bytes);
3555 }
3556
3557 char* end = start + bytes;
3558
3559 // Find the regions of the allocated chunk that can be promoted to large pages.
3560 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3561 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3562
3563 size_t lp_bytes = lp_end - lp_start;
3564
3565 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3566
3567 if (lp_bytes == 0) {
3568 // The mapped region doesn't even span the start and the end of a large page.
3569 // Fall back to allocate a non-special area.
3570 ::munmap(start, end - start);
3571 return NULL;
3572 }
3573
3574 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3575
3576
3577 void* result;
3578
3579 if (start != lp_start) {
3580 result = ::mmap(start, lp_start - start, prot,
3581 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3582 -1, 0);
3583 if (result == MAP_FAILED) {
3584 ::munmap(lp_start, end - lp_start);
3585 return NULL;
3586 }
3587 }
3588
3589 result = ::mmap(lp_start, lp_bytes, prot,
3590 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3591 -1, 0);
3592 if (result == MAP_FAILED) {
3593 warn_on_large_pages_failure(req_addr, bytes, errno);
3594 // If the mmap above fails, the large pages region will be unmapped and we
3595 // have regions before and after with small pages. Release these regions.
3596 //
3597 // | mapped | unmapped | mapped |
3598 // ^ ^ ^ ^
3599 // start lp_start lp_end end
3600 //
3601 ::munmap(start, lp_start - start);
3602 ::munmap(lp_end, end - lp_end);
3603 return NULL;
3604 }
3605
3606 if (lp_end != end) {
3607 result = ::mmap(lp_end, end - lp_end, prot,
3608 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3609 -1, 0);
3610 if (result == MAP_FAILED) {
3611 ::munmap(start, lp_end - start);
3612 return NULL;
3613 }
3614 }
3615
3616 return start;
3617 }
3618
3619 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3620 size_t alignment,
3621 char* req_addr,
3622 bool exec) {
3623 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3624 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3625 assert(is_power_of_2(alignment), "Must be");
3626 assert(is_power_of_2(os::large_page_size()), "Must be");
3627 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3628
3629 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3630 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3631 } else {
3632 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3633 }
3634 }
3635
3636 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3637 char* req_addr, bool exec) {
3638 assert(UseLargePages, "only for large pages");
3639
3640 char* addr;
3641 if (UseSHM) {
3642 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3643 } else {
3644 assert(UseHugeTLBFS, "must be");
3645 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3646 }
3647
3648 if (addr != NULL) {
3649 if (UseNUMAInterleaving) {
3650 numa_make_global(addr, bytes);
3651 }
3652
3653 // The memory is committed
3654 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3655 }
3656
3657 return addr;
3658 }
3659
3660 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3661 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3662 return shmdt(base) == 0;
3663 }
3664
3665 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3666 return pd_release_memory(base, bytes);
3667 }
3668
3669 bool os::release_memory_special(char* base, size_t bytes) {
3670 bool res;
3671 if (MemTracker::tracking_level() > NMT_minimal) {
3672 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3673 res = os::Linux::release_memory_special_impl(base, bytes);
3674 if (res) {
3675 tkr.record((address)base, bytes);
3676 }
3677
3678 } else {
3679 res = os::Linux::release_memory_special_impl(base, bytes);
3680 }
3681 return res;
3682 }
3683
3684 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3685 assert(UseLargePages, "only for large pages");
3686 bool res;
3687
3688 if (UseSHM) {
3689 res = os::Linux::release_memory_special_shm(base, bytes);
3690 } else {
3691 assert(UseHugeTLBFS, "must be");
3692 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3693 }
3694 return res;
3695 }
3696
3697 size_t os::large_page_size() {
3698 return _large_page_size;
3699 }
3700
3701 // With SysV SHM the entire memory region must be allocated as shared
3702 // memory.
3703 // HugeTLBFS allows application to commit large page memory on demand.
3704 // However, when committing memory with HugeTLBFS fails, the region
3705 // that was supposed to be committed will lose the old reservation
3706 // and allow other threads to steal that memory region. Because of this
3707 // behavior we can't commit HugeTLBFS memory.
3708 bool os::can_commit_large_page_memory() {
3709 return UseTransparentHugePages;
3710 }
3711
3712 bool os::can_execute_large_page_memory() {
3713 return UseTransparentHugePages || UseHugeTLBFS;
3714 }
3715
3716 // Reserve memory at an arbitrary address, only if that area is
3717 // available (and not reserved for something else).
3718
3719 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3720 const int max_tries = 10;
3721 char* base[max_tries];
3722 size_t size[max_tries];
3723 const size_t gap = 0x000000;
3724
3725 // Assert only that the size is a multiple of the page size, since
3726 // that's all that mmap requires, and since that's all we really know
3727 // about at this low abstraction level. If we need higher alignment,
3728 // we can either pass an alignment to this method or verify alignment
3729 // in one of the methods further up the call chain. See bug 5044738.
3730 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3731
3732 // Repeatedly allocate blocks until the block is allocated at the
3733 // right spot. Give up after max_tries. Note that reserve_memory() will
3734 // automatically update _highest_vm_reserved_address if the call is
3735 // successful. The variable tracks the highest memory address every reserved
3736 // by JVM. It is used to detect heap-stack collision if running with
3737 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3738 // space than needed, it could confuse the collision detecting code. To
3739 // solve the problem, save current _highest_vm_reserved_address and
3740 // calculate the correct value before return.
3741 address old_highest = _highest_vm_reserved_address;
3742
3743 // Linux mmap allows caller to pass an address as hint; give it a try first,
3744 // if kernel honors the hint then we can return immediately.
3745 char * addr = anon_mmap(requested_addr, bytes, false);
3746 if (addr == requested_addr) {
3747 return requested_addr;
3748 }
3749
3750 if (addr != NULL) {
3751 // mmap() is successful but it fails to reserve at the requested address
3752 anon_munmap(addr, bytes);
3753 }
3754
3755 int i;
3756 for (i = 0; i < max_tries; ++i) {
3757 base[i] = reserve_memory(bytes);
3758
3759 if (base[i] != NULL) {
3760 // Is this the block we wanted?
3761 if (base[i] == requested_addr) {
3762 size[i] = bytes;
3763 break;
3764 }
3765
3766 // Does this overlap the block we wanted? Give back the overlapped
3767 // parts and try again.
3768
3769 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3770 if (top_overlap >= 0 && top_overlap < bytes) {
3771 unmap_memory(base[i], top_overlap);
3772 base[i] += top_overlap;
3773 size[i] = bytes - top_overlap;
3774 } else {
3775 size_t bottom_overlap = base[i] + bytes - requested_addr;
3776 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3777 unmap_memory(requested_addr, bottom_overlap);
3778 size[i] = bytes - bottom_overlap;
3779 } else {
3780 size[i] = bytes;
3781 }
3782 }
3783 }
3784 }
3785
3786 // Give back the unused reserved pieces.
3787
3788 for (int j = 0; j < i; ++j) {
3789 if (base[j] != NULL) {
3790 unmap_memory(base[j], size[j]);
3791 }
3792 }
3793
3794 if (i < max_tries) {
3795 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3796 return requested_addr;
3797 } else {
3798 _highest_vm_reserved_address = old_highest;
3799 return NULL;
3800 }
3801 }
3802
3803 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3804 return ::read(fd, buf, nBytes);
3805 }
3806
3807 // Short sleep, direct OS call.
3808 //
3809 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3810 // sched_yield(2) will actually give up the CPU:
3811 //
3812 // * Alone on this pariticular CPU, keeps running.
3813 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3814 // (pre 2.6.39).
3815 //
3816 // So calling this with 0 is an alternative.
3817 //
3818 void os::naked_short_sleep(jlong ms) {
3819 struct timespec req;
3820
3821 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3822 req.tv_sec = 0;
3823 if (ms > 0) {
3824 req.tv_nsec = (ms % 1000) * 1000000;
3825 } else {
3826 req.tv_nsec = 1;
3827 }
3828
3829 nanosleep(&req, NULL);
3830
3831 return;
3832 }
3833
3834 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3835 void os::infinite_sleep() {
3836 while (true) { // sleep forever ...
3837 ::sleep(100); // ... 100 seconds at a time
3838 }
3839 }
3840
3841 // Used to convert frequent JVM_Yield() to nops
3842 bool os::dont_yield() {
3843 return DontYieldALot;
3844 }
3845
3846 void os::naked_yield() {
3847 sched_yield();
3848 }
3849
3850 ////////////////////////////////////////////////////////////////////////////////
3851 // thread priority support
3852
3853 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3854 // only supports dynamic priority, static priority must be zero. For real-time
3855 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3856 // However, for large multi-threaded applications, SCHED_RR is not only slower
3857 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3858 // of 5 runs - Sep 2005).
3859 //
3860 // The following code actually changes the niceness of kernel-thread/LWP. It
3861 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3862 // not the entire user process, and user level threads are 1:1 mapped to kernel
3863 // threads. It has always been the case, but could change in the future. For
3864 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3865 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3866
3867 int os::java_to_os_priority[CriticalPriority + 1] = {
3868 19, // 0 Entry should never be used
3869
3870 4, // 1 MinPriority
3871 3, // 2
3872 2, // 3
3873
3874 1, // 4
3875 0, // 5 NormPriority
3876 -1, // 6
3877
3878 -2, // 7
3879 -3, // 8
3880 -4, // 9 NearMaxPriority
3881
3882 -5, // 10 MaxPriority
3883
3884 -5 // 11 CriticalPriority
3885 };
3886
3887 static int prio_init() {
3888 if (ThreadPriorityPolicy == 1) {
3889 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3890 // if effective uid is not root. Perhaps, a more elegant way of doing
3891 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3892 if (geteuid() != 0) {
3893 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3894 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3895 }
3896 ThreadPriorityPolicy = 0;
3897 }
3898 }
3899 if (UseCriticalJavaThreadPriority) {
3900 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3901 }
3902 return 0;
3903 }
3904
3905 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3906 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3907
3908 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3909 return (ret == 0) ? OS_OK : OS_ERR;
3910 }
3911
3912 OSReturn os::get_native_priority(const Thread* const thread,
3913 int *priority_ptr) {
3914 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3915 *priority_ptr = java_to_os_priority[NormPriority];
3916 return OS_OK;
3917 }
3918
3919 errno = 0;
3920 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3921 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3922 }
3923
3924 // Hint to the underlying OS that a task switch would not be good.
3925 // Void return because it's a hint and can fail.
3926 void os::hint_no_preempt() {}
3927
3928 ////////////////////////////////////////////////////////////////////////////////
3929 // suspend/resume support
3930
3931 // the low-level signal-based suspend/resume support is a remnant from the
3932 // old VM-suspension that used to be for java-suspension, safepoints etc,
3933 // within hotspot. Now there is a single use-case for this:
3934 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3935 // that runs in the watcher thread.
3936 // The remaining code is greatly simplified from the more general suspension
3937 // code that used to be used.
3938 //
3939 // The protocol is quite simple:
3940 // - suspend:
3941 // - sends a signal to the target thread
3942 // - polls the suspend state of the osthread using a yield loop
3943 // - target thread signal handler (SR_handler) sets suspend state
3944 // and blocks in sigsuspend until continued
3945 // - resume:
3946 // - sets target osthread state to continue
3947 // - sends signal to end the sigsuspend loop in the SR_handler
3948 //
3949 // Note that the SR_lock plays no role in this suspend/resume protocol.
3950
3951 static void resume_clear_context(OSThread *osthread) {
3952 osthread->set_ucontext(NULL);
3953 osthread->set_siginfo(NULL);
3954 }
3955
3956 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3957 ucontext_t* context) {
3958 osthread->set_ucontext(context);
3959 osthread->set_siginfo(siginfo);
3960 }
3961
3962 // Handler function invoked when a thread's execution is suspended or
3963 // resumed. We have to be careful that only async-safe functions are
3964 // called here (Note: most pthread functions are not async safe and
3965 // should be avoided.)
3966 //
3967 // Note: sigwait() is a more natural fit than sigsuspend() from an
3968 // interface point of view, but sigwait() prevents the signal hander
3969 // from being run. libpthread would get very confused by not having
3970 // its signal handlers run and prevents sigwait()'s use with the
3971 // mutex granting granting signal.
3972 //
3973 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3974 //
3975 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3976 // Save and restore errno to avoid confusing native code with EINTR
3977 // after sigsuspend.
3978 int old_errno = errno;
3979
3980 Thread* thread = Thread::current();
3981 OSThread* osthread = thread->osthread();
3982 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3983
3984 os::SuspendResume::State current = osthread->sr.state();
3985 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3986 suspend_save_context(osthread, siginfo, context);
3987
3988 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3989 os::SuspendResume::State state = osthread->sr.suspended();
3990 if (state == os::SuspendResume::SR_SUSPENDED) {
3991 sigset_t suspend_set; // signals for sigsuspend()
3992
3993 // get current set of blocked signals and unblock resume signal
3994 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3995 sigdelset(&suspend_set, SR_signum);
3996
3997 sr_semaphore.signal();
3998 // wait here until we are resumed
3999 while (1) {
4000 sigsuspend(&suspend_set);
4001
4002 os::SuspendResume::State result = osthread->sr.running();
4003 if (result == os::SuspendResume::SR_RUNNING) {
4004 sr_semaphore.signal();
4005 break;
4006 }
4007 }
4008
4009 } else if (state == os::SuspendResume::SR_RUNNING) {
4010 // request was cancelled, continue
4011 } else {
4012 ShouldNotReachHere();
4013 }
4014
4015 resume_clear_context(osthread);
4016 } else if (current == os::SuspendResume::SR_RUNNING) {
4017 // request was cancelled, continue
4018 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4019 // ignore
4020 } else {
4021 // ignore
4022 }
4023
4024 errno = old_errno;
4025 }
4026
4027
4028 static int SR_initialize() {
4029 struct sigaction act;
4030 char *s;
4031 // Get signal number to use for suspend/resume
4032 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4033 int sig = ::strtol(s, 0, 10);
4034 if (sig > 0 || sig < _NSIG) {
4035 SR_signum = sig;
4036 }
4037 }
4038
4039 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4040 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4041
4042 sigemptyset(&SR_sigset);
4043 sigaddset(&SR_sigset, SR_signum);
4044
4045 // Set up signal handler for suspend/resume
4046 act.sa_flags = SA_RESTART|SA_SIGINFO;
4047 act.sa_handler = (void (*)(int)) SR_handler;
4048
4049 // SR_signum is blocked by default.
4050 // 4528190 - We also need to block pthread restart signal (32 on all
4051 // supported Linux platforms). Note that LinuxThreads need to block
4052 // this signal for all threads to work properly. So we don't have
4053 // to use hard-coded signal number when setting up the mask.
4054 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4055
4056 if (sigaction(SR_signum, &act, 0) == -1) {
4057 return -1;
4058 }
4059
4060 // Save signal flag
4061 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4062 return 0;
4063 }
4064
4065 static int sr_notify(OSThread* osthread) {
4066 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4067 assert_status(status == 0, status, "pthread_kill");
4068 return status;
4069 }
4070
4071 // "Randomly" selected value for how long we want to spin
4072 // before bailing out on suspending a thread, also how often
4073 // we send a signal to a thread we want to resume
4074 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4075 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4076
4077 // returns true on success and false on error - really an error is fatal
4078 // but this seems the normal response to library errors
4079 static bool do_suspend(OSThread* osthread) {
4080 assert(osthread->sr.is_running(), "thread should be running");
4081 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4082
4083 // mark as suspended and send signal
4084 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4085 // failed to switch, state wasn't running?
4086 ShouldNotReachHere();
4087 return false;
4088 }
4089
4090 if (sr_notify(osthread) != 0) {
4091 ShouldNotReachHere();
4092 }
4093
4094 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4095 while (true) {
4096 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4097 break;
4098 } else {
4099 // timeout
4100 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4101 if (cancelled == os::SuspendResume::SR_RUNNING) {
4102 return false;
4103 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4104 // make sure that we consume the signal on the semaphore as well
4105 sr_semaphore.wait();
4106 break;
4107 } else {
4108 ShouldNotReachHere();
4109 return false;
4110 }
4111 }
4112 }
4113
4114 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4115 return true;
4116 }
4117
4118 static void do_resume(OSThread* osthread) {
4119 assert(osthread->sr.is_suspended(), "thread should be suspended");
4120 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4121
4122 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4123 // failed to switch to WAKEUP_REQUEST
4124 ShouldNotReachHere();
4125 return;
4126 }
4127
4128 while (true) {
4129 if (sr_notify(osthread) == 0) {
4130 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4131 if (osthread->sr.is_running()) {
4132 return;
4133 }
4134 }
4135 } else {
4136 ShouldNotReachHere();
4137 }
4138 }
4139
4140 guarantee(osthread->sr.is_running(), "Must be running!");
4141 }
4142
4143 ///////////////////////////////////////////////////////////////////////////////////
4144 // signal handling (except suspend/resume)
4145
4146 // This routine may be used by user applications as a "hook" to catch signals.
4147 // The user-defined signal handler must pass unrecognized signals to this
4148 // routine, and if it returns true (non-zero), then the signal handler must
4149 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4150 // routine will never retun false (zero), but instead will execute a VM panic
4151 // routine kill the process.
4152 //
4153 // If this routine returns false, it is OK to call it again. This allows
4154 // the user-defined signal handler to perform checks either before or after
4155 // the VM performs its own checks. Naturally, the user code would be making
4156 // a serious error if it tried to handle an exception (such as a null check
4157 // or breakpoint) that the VM was generating for its own correct operation.
4158 //
4159 // This routine may recognize any of the following kinds of signals:
4160 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4161 // It should be consulted by handlers for any of those signals.
4162 //
4163 // The caller of this routine must pass in the three arguments supplied
4164 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4165 // field of the structure passed to sigaction(). This routine assumes that
4166 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4167 //
4168 // Note that the VM will print warnings if it detects conflicting signal
4169 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4170 //
4171 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4172 siginfo_t* siginfo,
4173 void* ucontext,
4174 int abort_if_unrecognized);
4175
4176 void signalHandler(int sig, siginfo_t* info, void* uc) {
4177 assert(info != NULL && uc != NULL, "it must be old kernel");
4178 int orig_errno = errno; // Preserve errno value over signal handler.
4179 JVM_handle_linux_signal(sig, info, uc, true);
4180 errno = orig_errno;
4181 }
4182
4183
4184 // This boolean allows users to forward their own non-matching signals
4185 // to JVM_handle_linux_signal, harmlessly.
4186 bool os::Linux::signal_handlers_are_installed = false;
4187
4188 // For signal-chaining
4189 struct sigaction os::Linux::sigact[MAXSIGNUM];
4190 unsigned int os::Linux::sigs = 0;
4191 bool os::Linux::libjsig_is_loaded = false;
4192 typedef struct sigaction *(*get_signal_t)(int);
4193 get_signal_t os::Linux::get_signal_action = NULL;
4194
4195 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4196 struct sigaction *actp = NULL;
4197
4198 if (libjsig_is_loaded) {
4199 // Retrieve the old signal handler from libjsig
4200 actp = (*get_signal_action)(sig);
4201 }
4202 if (actp == NULL) {
4203 // Retrieve the preinstalled signal handler from jvm
4204 actp = get_preinstalled_handler(sig);
4205 }
4206
4207 return actp;
4208 }
4209
4210 static bool call_chained_handler(struct sigaction *actp, int sig,
4211 siginfo_t *siginfo, void *context) {
4212 // Call the old signal handler
4213 if (actp->sa_handler == SIG_DFL) {
4214 // It's more reasonable to let jvm treat it as an unexpected exception
4215 // instead of taking the default action.
4216 return false;
4217 } else if (actp->sa_handler != SIG_IGN) {
4218 if ((actp->sa_flags & SA_NODEFER) == 0) {
4219 // automaticlly block the signal
4220 sigaddset(&(actp->sa_mask), sig);
4221 }
4222
4223 sa_handler_t hand;
4224 sa_sigaction_t sa;
4225 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4226 // retrieve the chained handler
4227 if (siginfo_flag_set) {
4228 sa = actp->sa_sigaction;
4229 } else {
4230 hand = actp->sa_handler;
4231 }
4232
4233 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4234 actp->sa_handler = SIG_DFL;
4235 }
4236
4237 // try to honor the signal mask
4238 sigset_t oset;
4239 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4240
4241 // call into the chained handler
4242 if (siginfo_flag_set) {
4243 (*sa)(sig, siginfo, context);
4244 } else {
4245 (*hand)(sig);
4246 }
4247
4248 // restore the signal mask
4249 pthread_sigmask(SIG_SETMASK, &oset, 0);
4250 }
4251 // Tell jvm's signal handler the signal is taken care of.
4252 return true;
4253 }
4254
4255 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4256 bool chained = false;
4257 // signal-chaining
4258 if (UseSignalChaining) {
4259 struct sigaction *actp = get_chained_signal_action(sig);
4260 if (actp != NULL) {
4261 chained = call_chained_handler(actp, sig, siginfo, context);
4262 }
4263 }
4264 return chained;
4265 }
4266
4267 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4268 if ((((unsigned int)1 << sig) & sigs) != 0) {
4269 return &sigact[sig];
4270 }
4271 return NULL;
4272 }
4273
4274 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4275 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4276 sigact[sig] = oldAct;
4277 sigs |= (unsigned int)1 << sig;
4278 }
4279
4280 // for diagnostic
4281 int os::Linux::sigflags[MAXSIGNUM];
4282
4283 int os::Linux::get_our_sigflags(int sig) {
4284 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4285 return sigflags[sig];
4286 }
4287
4288 void os::Linux::set_our_sigflags(int sig, int flags) {
4289 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4290 sigflags[sig] = flags;
4291 }
4292
4293 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4294 // Check for overwrite.
4295 struct sigaction oldAct;
4296 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4297
4298 void* oldhand = oldAct.sa_sigaction
4299 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4300 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4301 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4302 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4303 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4304 if (AllowUserSignalHandlers || !set_installed) {
4305 // Do not overwrite; user takes responsibility to forward to us.
4306 return;
4307 } else if (UseSignalChaining) {
4308 // save the old handler in jvm
4309 save_preinstalled_handler(sig, oldAct);
4310 // libjsig also interposes the sigaction() call below and saves the
4311 // old sigaction on it own.
4312 } else {
4313 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4314 "%#lx for signal %d.", (long)oldhand, sig));
4315 }
4316 }
4317
4318 struct sigaction sigAct;
4319 sigfillset(&(sigAct.sa_mask));
4320 sigAct.sa_handler = SIG_DFL;
4321 if (!set_installed) {
4322 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4323 } else {
4324 sigAct.sa_sigaction = signalHandler;
4325 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4326 }
4327 // Save flags, which are set by ours
4328 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4329 sigflags[sig] = sigAct.sa_flags;
4330
4331 int ret = sigaction(sig, &sigAct, &oldAct);
4332 assert(ret == 0, "check");
4333
4334 void* oldhand2 = oldAct.sa_sigaction
4335 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4336 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4337 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4338 }
4339
4340 // install signal handlers for signals that HotSpot needs to
4341 // handle in order to support Java-level exception handling.
4342
4343 void os::Linux::install_signal_handlers() {
4344 if (!signal_handlers_are_installed) {
4345 signal_handlers_are_installed = true;
4346
4347 // signal-chaining
4348 typedef void (*signal_setting_t)();
4349 signal_setting_t begin_signal_setting = NULL;
4350 signal_setting_t end_signal_setting = NULL;
4351 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4352 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4353 if (begin_signal_setting != NULL) {
4354 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4355 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4356 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4357 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4358 libjsig_is_loaded = true;
4359 assert(UseSignalChaining, "should enable signal-chaining");
4360 }
4361 if (libjsig_is_loaded) {
4362 // Tell libjsig jvm is setting signal handlers
4363 (*begin_signal_setting)();
4364 }
4365
4366 set_signal_handler(SIGSEGV, true);
4367 set_signal_handler(SIGPIPE, true);
4368 set_signal_handler(SIGBUS, true);
4369 set_signal_handler(SIGILL, true);
4370 set_signal_handler(SIGFPE, true);
4371 #if defined(PPC64)
4372 set_signal_handler(SIGTRAP, true);
4373 #endif
4374 set_signal_handler(SIGXFSZ, true);
4375
4376 if (libjsig_is_loaded) {
4377 // Tell libjsig jvm finishes setting signal handlers
4378 (*end_signal_setting)();
4379 }
4380
4381 // We don't activate signal checker if libjsig is in place, we trust ourselves
4382 // and if UserSignalHandler is installed all bets are off.
4383 // Log that signal checking is off only if -verbose:jni is specified.
4384 if (CheckJNICalls) {
4385 if (libjsig_is_loaded) {
4386 if (PrintJNIResolving) {
4387 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4388 }
4389 check_signals = false;
4390 }
4391 if (AllowUserSignalHandlers) {
4392 if (PrintJNIResolving) {
4393 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4394 }
4395 check_signals = false;
4396 }
4397 }
4398 }
4399 }
4400
4401 // This is the fastest way to get thread cpu time on Linux.
4402 // Returns cpu time (user+sys) for any thread, not only for current.
4403 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4404 // It might work on 2.6.10+ with a special kernel/glibc patch.
4405 // For reference, please, see IEEE Std 1003.1-2004:
4406 // http://www.unix.org/single_unix_specification
4407
4408 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4409 struct timespec tp;
4410 int rc = os::Linux::clock_gettime(clockid, &tp);
4411 assert(rc == 0, "clock_gettime is expected to return 0 code");
4412
4413 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4414 }
4415
4416 /////
4417 // glibc on Linux platform uses non-documented flag
4418 // to indicate, that some special sort of signal
4419 // trampoline is used.
4420 // We will never set this flag, and we should
4421 // ignore this flag in our diagnostic
4422 #ifdef SIGNIFICANT_SIGNAL_MASK
4423 #undef SIGNIFICANT_SIGNAL_MASK
4424 #endif
4425 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4426
4427 static const char* get_signal_handler_name(address handler,
4428 char* buf, int buflen) {
4429 int offset;
4430 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4431 if (found) {
4432 // skip directory names
4433 const char *p1, *p2;
4434 p1 = buf;
4435 size_t len = strlen(os::file_separator());
4436 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4437 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4438 } else {
4439 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4440 }
4441 return buf;
4442 }
4443
4444 static void print_signal_handler(outputStream* st, int sig,
4445 char* buf, size_t buflen) {
4446 struct sigaction sa;
4447
4448 sigaction(sig, NULL, &sa);
4449
4450 // See comment for SIGNIFICANT_SIGNAL_MASK define
4451 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4452
4453 st->print("%s: ", os::exception_name(sig, buf, buflen));
4454
4455 address handler = (sa.sa_flags & SA_SIGINFO)
4456 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4457 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4458
4459 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4460 st->print("SIG_DFL");
4461 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4462 st->print("SIG_IGN");
4463 } else {
4464 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4465 }
4466
4467 st->print(", sa_mask[0]=");
4468 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4469
4470 address rh = VMError::get_resetted_sighandler(sig);
4471 // May be, handler was resetted by VMError?
4472 if (rh != NULL) {
4473 handler = rh;
4474 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4475 }
4476
4477 st->print(", sa_flags=");
4478 os::Posix::print_sa_flags(st, sa.sa_flags);
4479
4480 // Check: is it our handler?
4481 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4482 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4483 // It is our signal handler
4484 // check for flags, reset system-used one!
4485 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4486 st->print(
4487 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4488 os::Linux::get_our_sigflags(sig));
4489 }
4490 }
4491 st->cr();
4492 }
4493
4494
4495 #define DO_SIGNAL_CHECK(sig) \
4496 do { \
4497 if (!sigismember(&check_signal_done, sig)) { \
4498 os::Linux::check_signal_handler(sig); \
4499 } \
4500 } while (0)
4501
4502 // This method is a periodic task to check for misbehaving JNI applications
4503 // under CheckJNI, we can add any periodic checks here
4504
4505 void os::run_periodic_checks() {
4506 if (check_signals == false) return;
4507
4508 // SEGV and BUS if overridden could potentially prevent
4509 // generation of hs*.log in the event of a crash, debugging
4510 // such a case can be very challenging, so we absolutely
4511 // check the following for a good measure:
4512 DO_SIGNAL_CHECK(SIGSEGV);
4513 DO_SIGNAL_CHECK(SIGILL);
4514 DO_SIGNAL_CHECK(SIGFPE);
4515 DO_SIGNAL_CHECK(SIGBUS);
4516 DO_SIGNAL_CHECK(SIGPIPE);
4517 DO_SIGNAL_CHECK(SIGXFSZ);
4518 #if defined(PPC64)
4519 DO_SIGNAL_CHECK(SIGTRAP);
4520 #endif
4521
4522 // ReduceSignalUsage allows the user to override these handlers
4523 // see comments at the very top and jvm_solaris.h
4524 if (!ReduceSignalUsage) {
4525 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4526 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4527 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4528 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4529 }
4530
4531 DO_SIGNAL_CHECK(SR_signum);
4532 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4533 }
4534
4535 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4536
4537 static os_sigaction_t os_sigaction = NULL;
4538
4539 void os::Linux::check_signal_handler(int sig) {
4540 char buf[O_BUFLEN];
4541 address jvmHandler = NULL;
4542
4543
4544 struct sigaction act;
4545 if (os_sigaction == NULL) {
4546 // only trust the default sigaction, in case it has been interposed
4547 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4548 if (os_sigaction == NULL) return;
4549 }
4550
4551 os_sigaction(sig, (struct sigaction*)NULL, &act);
4552
4553
4554 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4555
4556 address thisHandler = (act.sa_flags & SA_SIGINFO)
4557 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4558 : CAST_FROM_FN_PTR(address, act.sa_handler);
4559
4560
4561 switch (sig) {
4562 case SIGSEGV:
4563 case SIGBUS:
4564 case SIGFPE:
4565 case SIGPIPE:
4566 case SIGILL:
4567 case SIGXFSZ:
4568 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4569 break;
4570
4571 case SHUTDOWN1_SIGNAL:
4572 case SHUTDOWN2_SIGNAL:
4573 case SHUTDOWN3_SIGNAL:
4574 case BREAK_SIGNAL:
4575 jvmHandler = (address)user_handler();
4576 break;
4577
4578 case INTERRUPT_SIGNAL:
4579 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4580 break;
4581
4582 default:
4583 if (sig == SR_signum) {
4584 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4585 } else {
4586 return;
4587 }
4588 break;
4589 }
4590
4591 if (thisHandler != jvmHandler) {
4592 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4593 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4594 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4595 // No need to check this sig any longer
4596 sigaddset(&check_signal_done, sig);
4597 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4598 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4599 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4600 exception_name(sig, buf, O_BUFLEN));
4601 }
4602 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4603 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4604 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4605 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4606 // No need to check this sig any longer
4607 sigaddset(&check_signal_done, sig);
4608 }
4609
4610 // Dump all the signal
4611 if (sigismember(&check_signal_done, sig)) {
4612 print_signal_handlers(tty, buf, O_BUFLEN);
4613 }
4614 }
4615
4616 extern void report_error(char* file_name, int line_no, char* title,
4617 char* format, ...);
4618
4619 extern bool signal_name(int signo, char* buf, size_t len);
4620
4621 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4622 if (0 < exception_code && exception_code <= SIGRTMAX) {
4623 // signal
4624 if (!signal_name(exception_code, buf, size)) {
4625 jio_snprintf(buf, size, "SIG%d", exception_code);
4626 }
4627 return buf;
4628 } else {
4629 return NULL;
4630 }
4631 }
4632
4633 // this is called _before_ the most of global arguments have been parsed
4634 void os::init(void) {
4635 char dummy; // used to get a guess on initial stack address
4636 // first_hrtime = gethrtime();
4637
4638 // With LinuxThreads the JavaMain thread pid (primordial thread)
4639 // is different than the pid of the java launcher thread.
4640 // So, on Linux, the launcher thread pid is passed to the VM
4641 // via the sun.java.launcher.pid property.
4642 // Use this property instead of getpid() if it was correctly passed.
4643 // See bug 6351349.
4644 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4645
4646 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4647
4648 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4649
4650 init_random(1234567);
4651
4652 ThreadCritical::initialize();
4653
4654 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4655 if (Linux::page_size() == -1) {
4656 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4657 strerror(errno)));
4658 }
4659 init_page_sizes((size_t) Linux::page_size());
4660
4661 Linux::initialize_system_info();
4662
4663 // main_thread points to the aboriginal thread
4664 Linux::_main_thread = pthread_self();
4665
4666 Linux::clock_init();
4667 initial_time_count = javaTimeNanos();
4668
4669 // pthread_condattr initialization for monotonic clock
4670 int status;
4671 pthread_condattr_t* _condattr = os::Linux::condAttr();
4672 if ((status = pthread_condattr_init(_condattr)) != 0) {
4673 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4674 }
4675 // Only set the clock if CLOCK_MONOTONIC is available
4676 if (os::supports_monotonic_clock()) {
4677 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4678 if (status == EINVAL) {
4679 warning("Unable to use monotonic clock with relative timed-waits" \
4680 " - changes to the time-of-day clock may have adverse affects");
4681 } else {
4682 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4683 }
4684 }
4685 }
4686 // else it defaults to CLOCK_REALTIME
4687
4688 pthread_mutex_init(&dl_mutex, NULL);
4689
4690 // If the pagesize of the VM is greater than 8K determine the appropriate
4691 // number of initial guard pages. The user can change this with the
4692 // command line arguments, if needed.
4693 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4694 StackYellowPages = 1;
4695 StackRedPages = 1;
4696 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4697 }
4698 }
4699
4700 // To install functions for atexit system call
4701 extern "C" {
4702 static void perfMemory_exit_helper() {
4703 perfMemory_exit();
4704 }
4705 }
4706
4707 // this is called _after_ the global arguments have been parsed
4708 jint os::init_2(void) {
4709 Linux::fast_thread_clock_init();
4710
4711 // Allocate a single page and mark it as readable for safepoint polling
4712 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4713 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4714
4715 os::set_polling_page(polling_page);
4716
4717 #ifndef PRODUCT
4718 if (Verbose && PrintMiscellaneous) {
4719 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4720 (intptr_t)polling_page);
4721 }
4722 #endif
4723
4724 if (!UseMembar) {
4725 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4726 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4727 os::set_memory_serialize_page(mem_serialize_page);
4728
4729 #ifndef PRODUCT
4730 if (Verbose && PrintMiscellaneous) {
4731 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4732 (intptr_t)mem_serialize_page);
4733 }
4734 #endif
4735 }
4736
4737 // initialize suspend/resume support - must do this before signal_sets_init()
4738 if (SR_initialize() != 0) {
4739 perror("SR_initialize failed");
4740 return JNI_ERR;
4741 }
4742
4743 Linux::signal_sets_init();
4744 Linux::install_signal_handlers();
4745
4746 // Check minimum allowable stack size for thread creation and to initialize
4747 // the java system classes, including StackOverflowError - depends on page
4748 // size. Add a page for compiler2 recursion in main thread.
4749 // Add in 2*BytesPerWord times page size to account for VM stack during
4750 // class initialization depending on 32 or 64 bit VM.
4751 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4752 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4753 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4754
4755 size_t threadStackSizeInBytes = ThreadStackSize * K;
4756 if (threadStackSizeInBytes != 0 &&
4757 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4758 tty->print_cr("\nThe stack size specified is too small, "
4759 "Specify at least %dk",
4760 os::Linux::min_stack_allowed/ K);
4761 return JNI_ERR;
4762 }
4763
4764 // Make the stack size a multiple of the page size so that
4765 // the yellow/red zones can be guarded.
4766 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4767 vm_page_size()));
4768
4769 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4770
4771 #if defined(IA32)
4772 workaround_expand_exec_shield_cs_limit();
4773 #endif
4774
4775 Linux::libpthread_init();
4776 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4777 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4778 Linux::glibc_version(), Linux::libpthread_version(),
4779 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4780 }
4781
4782 if (UseNUMA) {
4783 if (!Linux::libnuma_init()) {
4784 UseNUMA = false;
4785 } else {
4786 if ((Linux::numa_max_node() < 1)) {
4787 // There's only one node(they start from 0), disable NUMA.
4788 UseNUMA = false;
4789 }
4790 }
4791 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4792 // we can make the adaptive lgrp chunk resizing work. If the user specified
4793 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4794 // disable adaptive resizing.
4795 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4796 if (FLAG_IS_DEFAULT(UseNUMA)) {
4797 UseNUMA = false;
4798 } else {
4799 if (FLAG_IS_DEFAULT(UseLargePages) &&
4800 FLAG_IS_DEFAULT(UseSHM) &&
4801 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4802 UseLargePages = false;
4803 } else {
4804 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4805 UseAdaptiveSizePolicy = false;
4806 UseAdaptiveNUMAChunkSizing = false;
4807 }
4808 }
4809 }
4810 if (!UseNUMA && ForceNUMA) {
4811 UseNUMA = true;
4812 }
4813 }
4814
4815 if (MaxFDLimit) {
4816 // set the number of file descriptors to max. print out error
4817 // if getrlimit/setrlimit fails but continue regardless.
4818 struct rlimit nbr_files;
4819 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4820 if (status != 0) {
4821 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4822 perror("os::init_2 getrlimit failed");
4823 }
4824 } else {
4825 nbr_files.rlim_cur = nbr_files.rlim_max;
4826 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4827 if (status != 0) {
4828 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4829 perror("os::init_2 setrlimit failed");
4830 }
4831 }
4832 }
4833 }
4834
4835 // Initialize lock used to serialize thread creation (see os::create_thread)
4836 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4837
4838 // at-exit methods are called in the reverse order of their registration.
4839 // atexit functions are called on return from main or as a result of a
4840 // call to exit(3C). There can be only 32 of these functions registered
4841 // and atexit() does not set errno.
4842
4843 if (PerfAllowAtExitRegistration) {
4844 // only register atexit functions if PerfAllowAtExitRegistration is set.
4845 // atexit functions can be delayed until process exit time, which
4846 // can be problematic for embedded VM situations. Embedded VMs should
4847 // call DestroyJavaVM() to assure that VM resources are released.
4848
4849 // note: perfMemory_exit_helper atexit function may be removed in
4850 // the future if the appropriate cleanup code can be added to the
4851 // VM_Exit VMOperation's doit method.
4852 if (atexit(perfMemory_exit_helper) != 0) {
4853 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4854 }
4855 }
4856
4857 // initialize thread priority policy
4858 prio_init();
4859
4860 return JNI_OK;
4861 }
4862
4863 // this is called at the end of vm_initialization
4864 void os::init_3(void) {
4865 #ifdef JAVASE_EMBEDDED
4866 // Start the MemNotifyThread
4867 if (LowMemoryProtection) {
4868 MemNotifyThread::start();
4869 }
4870 return;
4871 #endif
4872 }
4873
4874 // Mark the polling page as unreadable
4875 void os::make_polling_page_unreadable(void) {
4876 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4877 fatal("Could not disable polling page");
4878 }
4879 }
4880
4881 // Mark the polling page as readable
4882 void os::make_polling_page_readable(void) {
4883 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4884 fatal("Could not enable polling page");
4885 }
4886 }
4887
4888 int os::active_processor_count() {
4889 // Linux doesn't yet have a (official) notion of processor sets,
4890 // so just return the number of online processors.
4891 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4892 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4893 return online_cpus;
4894 }
4895
4896 void os::set_native_thread_name(const char *name) {
4897 // Not yet implemented.
4898 return;
4899 }
4900
4901 bool os::distribute_processes(uint length, uint* distribution) {
4902 // Not yet implemented.
4903 return false;
4904 }
4905
4906 bool os::bind_to_processor(uint processor_id) {
4907 // Not yet implemented.
4908 return false;
4909 }
4910
4911 ///
4912
4913 void os::SuspendedThreadTask::internal_do_task() {
4914 if (do_suspend(_thread->osthread())) {
4915 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4916 do_task(context);
4917 do_resume(_thread->osthread());
4918 }
4919 }
4920
4921 class PcFetcher : public os::SuspendedThreadTask {
4922 public:
4923 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4924 ExtendedPC result();
4925 protected:
4926 void do_task(const os::SuspendedThreadTaskContext& context);
4927 private:
4928 ExtendedPC _epc;
4929 };
4930
4931 ExtendedPC PcFetcher::result() {
4932 guarantee(is_done(), "task is not done yet.");
4933 return _epc;
4934 }
4935
4936 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4937 Thread* thread = context.thread();
4938 OSThread* osthread = thread->osthread();
4939 if (osthread->ucontext() != NULL) {
4940 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4941 } else {
4942 // NULL context is unexpected, double-check this is the VMThread
4943 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4944 }
4945 }
4946
4947 // Suspends the target using the signal mechanism and then grabs the PC before
4948 // resuming the target. Used by the flat-profiler only
4949 ExtendedPC os::get_thread_pc(Thread* thread) {
4950 // Make sure that it is called by the watcher for the VMThread
4951 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4952 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4953
4954 PcFetcher fetcher(thread);
4955 fetcher.run();
4956 return fetcher.result();
4957 }
4958
4959 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4960 pthread_mutex_t *_mutex,
4961 const struct timespec *_abstime) {
4962 if (is_NPTL()) {
4963 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4964 } else {
4965 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4966 // word back to default 64bit precision if condvar is signaled. Java
4967 // wants 53bit precision. Save and restore current value.
4968 int fpu = get_fpu_control_word();
4969 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4970 set_fpu_control_word(fpu);
4971 return status;
4972 }
4973 }
4974
4975 ////////////////////////////////////////////////////////////////////////////////
4976 // debug support
4977
4978 bool os::find(address addr, outputStream* st) {
4979 Dl_info dlinfo;
4980 memset(&dlinfo, 0, sizeof(dlinfo));
4981 if (dladdr(addr, &dlinfo) != 0) {
4982 st->print(PTR_FORMAT ": ", addr);
4983 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4984 st->print("%s+%#x", dlinfo.dli_sname,
4985 addr - (intptr_t)dlinfo.dli_saddr);
4986 } else if (dlinfo.dli_fbase != NULL) {
4987 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4988 } else {
4989 st->print("<absolute address>");
4990 }
4991 if (dlinfo.dli_fname != NULL) {
4992 st->print(" in %s", dlinfo.dli_fname);
4993 }
4994 if (dlinfo.dli_fbase != NULL) {
4995 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4996 }
4997 st->cr();
4998
4999 if (Verbose) {
5000 // decode some bytes around the PC
5001 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5002 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5003 address lowest = (address) dlinfo.dli_sname;
5004 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5005 if (begin < lowest) begin = lowest;
5006 Dl_info dlinfo2;
5007 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5008 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5009 end = (address) dlinfo2.dli_saddr;
5010 }
5011 Disassembler::decode(begin, end, st);
5012 }
5013 return true;
5014 }
5015 return false;
5016 }
5017
5018 ////////////////////////////////////////////////////////////////////////////////
5019 // misc
5020
5021 // This does not do anything on Linux. This is basically a hook for being
5022 // able to use structured exception handling (thread-local exception filters)
5023 // on, e.g., Win32.
5024 void
5025 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5026 JavaCallArguments* args, Thread* thread) {
5027 f(value, method, args, thread);
5028 }
5029
5030 void os::print_statistics() {
5031 }
5032
5033 int os::message_box(const char* title, const char* message) {
5034 int i;
5035 fdStream err(defaultStream::error_fd());
5036 for (i = 0; i < 78; i++) err.print_raw("=");
5037 err.cr();
5038 err.print_raw_cr(title);
5039 for (i = 0; i < 78; i++) err.print_raw("-");
5040 err.cr();
5041 err.print_raw_cr(message);
5042 for (i = 0; i < 78; i++) err.print_raw("=");
5043 err.cr();
5044
5045 char buf[16];
5046 // Prevent process from exiting upon "read error" without consuming all CPU
5047 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5048
5049 return buf[0] == 'y' || buf[0] == 'Y';
5050 }
5051
5052 int os::stat(const char *path, struct stat *sbuf) {
5053 char pathbuf[MAX_PATH];
5054 if (strlen(path) > MAX_PATH - 1) {
5055 errno = ENAMETOOLONG;
5056 return -1;
5057 }
5058 os::native_path(strcpy(pathbuf, path));
5059 return ::stat(pathbuf, sbuf);
5060 }
5061
5062 bool os::check_heap(bool force) {
5063 return true;
5064 }
5065
5066 // Is a (classpath) directory empty?
5067 bool os::dir_is_empty(const char* path) {
5068 DIR *dir = NULL;
5069 struct dirent *ptr;
5070
5071 dir = opendir(path);
5072 if (dir == NULL) return true;
5073
5074 // Scan the directory
5075 bool result = true;
5076 char buf[sizeof(struct dirent) + MAX_PATH];
5077 while (result && (ptr = ::readdir(dir)) != NULL) {
5078 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5079 result = false;
5080 }
5081 }
5082 closedir(dir);
5083 return result;
5084 }
5085
5086 // This code originates from JDK's sysOpen and open64_w
5087 // from src/solaris/hpi/src/system_md.c
5088
5089 #ifndef O_DELETE
5090 #define O_DELETE 0x10000
5091 #endif
5092
5093 // Open a file. Unlink the file immediately after open returns
5094 // if the specified oflag has the O_DELETE flag set.
5095 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5096
5097 int os::open(const char *path, int oflag, int mode) {
5098 if (strlen(path) > MAX_PATH - 1) {
5099 errno = ENAMETOOLONG;
5100 return -1;
5101 }
5102 int fd;
5103 int o_delete = (oflag & O_DELETE);
5104 oflag = oflag & ~O_DELETE;
5105
5106 fd = ::open64(path, oflag, mode);
5107 if (fd == -1) return -1;
5108
5109 //If the open succeeded, the file might still be a directory
5110 {
5111 struct stat64 buf64;
5112 int ret = ::fstat64(fd, &buf64);
5113 int st_mode = buf64.st_mode;
5114
5115 if (ret != -1) {
5116 if ((st_mode & S_IFMT) == S_IFDIR) {
5117 errno = EISDIR;
5118 ::close(fd);
5119 return -1;
5120 }
5121 } else {
5122 ::close(fd);
5123 return -1;
5124 }
5125 }
5126
5127 // All file descriptors that are opened in the JVM and not
5128 // specifically destined for a subprocess should have the
5129 // close-on-exec flag set. If we don't set it, then careless 3rd
5130 // party native code might fork and exec without closing all
5131 // appropriate file descriptors (e.g. as we do in closeDescriptors in
5132 // UNIXProcess.c), and this in turn might:
5133 //
5134 // - cause end-of-file to fail to be detected on some file
5135 // descriptors, resulting in mysterious hangs, or
5136 //
5137 // - might cause an fopen in the subprocess to fail on a system
5138 // suffering from bug 1085341.
5139 //
5140 // (Yes, the default setting of the close-on-exec flag is a Unix
5141 // design flaw)
5142 //
5143 // See:
5144 // 1085341: 32-bit stdio routines should support file descriptors >255
5145 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5146 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5147 //
5148 #ifdef FD_CLOEXEC
5149 {
5150 int flags = ::fcntl(fd, F_GETFD);
5151 if (flags != -1) {
5152 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5153 }
5154 }
5155 #endif
5156
5157 if (o_delete != 0) {
5158 ::unlink(path);
5159 }
5160 return fd;
5161 }
5162
5163
5164 // create binary file, rewriting existing file if required
5165 int os::create_binary_file(const char* path, bool rewrite_existing) {
5166 int oflags = O_WRONLY | O_CREAT;
5167 if (!rewrite_existing) {
5168 oflags |= O_EXCL;
5169 }
5170 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5171 }
5172
5173 // return current position of file pointer
5174 jlong os::current_file_offset(int fd) {
5175 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5176 }
5177
5178 // move file pointer to the specified offset
5179 jlong os::seek_to_file_offset(int fd, jlong offset) {
5180 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5181 }
5182
5183 // This code originates from JDK's sysAvailable
5184 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5185
5186 int os::available(int fd, jlong *bytes) {
5187 jlong cur, end;
5188 int mode;
5189 struct stat64 buf64;
5190
5191 if (::fstat64(fd, &buf64) >= 0) {
5192 mode = buf64.st_mode;
5193 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5194 // XXX: is the following call interruptible? If so, this might
5195 // need to go through the INTERRUPT_IO() wrapper as for other
5196 // blocking, interruptible calls in this file.
5197 int n;
5198 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5199 *bytes = n;
5200 return 1;
5201 }
5202 }
5203 }
5204 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5205 return 0;
5206 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5207 return 0;
5208 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5209 return 0;
5210 }
5211 *bytes = end - cur;
5212 return 1;
5213 }
5214
5215 int os::socket_available(int fd, jint *pbytes) {
5216 // Linux doc says EINTR not returned, unlike Solaris
5217 int ret = ::ioctl(fd, FIONREAD, pbytes);
5218
5219 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5220 // is expected to return 0 on failure and 1 on success to the jdk.
5221 return (ret < 0) ? 0 : 1;
5222 }
5223
5224 // Map a block of memory.
5225 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5226 char *addr, size_t bytes, bool read_only,
5227 bool allow_exec) {
5228 int prot;
5229 int flags = MAP_PRIVATE;
5230
5231 if (read_only) {
5232 prot = PROT_READ;
5233 } else {
5234 prot = PROT_READ | PROT_WRITE;
5235 }
5236
5237 if (allow_exec) {
5238 prot |= PROT_EXEC;
5239 }
5240
5241 if (addr != NULL) {
5242 flags |= MAP_FIXED;
5243 }
5244
5245 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5246 fd, file_offset);
5247 if (mapped_address == MAP_FAILED) {
5248 return NULL;
5249 }
5250 return mapped_address;
5251 }
5252
5253
5254 // Remap a block of memory.
5255 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5256 char *addr, size_t bytes, bool read_only,
5257 bool allow_exec) {
5258 // same as map_memory() on this OS
5259 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5260 allow_exec);
5261 }
5262
5263
5264 // Unmap a block of memory.
5265 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5266 return munmap(addr, bytes) == 0;
5267 }
5268
5269 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5270
5271 static clockid_t thread_cpu_clockid(Thread* thread) {
5272 pthread_t tid = thread->osthread()->pthread_id();
5273 clockid_t clockid;
5274
5275 // Get thread clockid
5276 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5277 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5278 return clockid;
5279 }
5280
5281 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5282 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5283 // of a thread.
5284 //
5285 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5286 // the fast estimate available on the platform.
5287
5288 jlong os::current_thread_cpu_time() {
5289 if (os::Linux::supports_fast_thread_cpu_time()) {
5290 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5291 } else {
5292 // return user + sys since the cost is the same
5293 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5294 }
5295 }
5296
5297 jlong os::thread_cpu_time(Thread* thread) {
5298 // consistent with what current_thread_cpu_time() returns
5299 if (os::Linux::supports_fast_thread_cpu_time()) {
5300 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5301 } else {
5302 return slow_thread_cpu_time(thread, true /* user + sys */);
5303 }
5304 }
5305
5306 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5307 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5308 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5309 } else {
5310 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5311 }
5312 }
5313
5314 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5315 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5316 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5317 } else {
5318 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5319 }
5320 }
5321
5322 // -1 on error.
5323 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5324 pid_t tid = thread->osthread()->thread_id();
5325 char *s;
5326 char stat[2048];
5327 int statlen;
5328 char proc_name[64];
5329 int count;
5330 long sys_time, user_time;
5331 char cdummy;
5332 int idummy;
5333 long ldummy;
5334 FILE *fp;
5335
5336 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5337 fp = fopen(proc_name, "r");
5338 if (fp == NULL) return -1;
5339 statlen = fread(stat, 1, 2047, fp);
5340 stat[statlen] = '\0';
5341 fclose(fp);
5342
5343 // Skip pid and the command string. Note that we could be dealing with
5344 // weird command names, e.g. user could decide to rename java launcher
5345 // to "java 1.4.2 :)", then the stat file would look like
5346 // 1234 (java 1.4.2 :)) R ... ...
5347 // We don't really need to know the command string, just find the last
5348 // occurrence of ")" and then start parsing from there. See bug 4726580.
5349 s = strrchr(stat, ')');
5350 if (s == NULL) return -1;
5351
5352 // Skip blank chars
5353 do s++; while (isspace(*s));
5354
5355 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5356 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5357 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5358 &user_time, &sys_time);
5359 if (count != 13) return -1;
5360 if (user_sys_cpu_time) {
5361 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5362 } else {
5363 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5364 }
5365 }
5366
5367 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5368 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5369 info_ptr->may_skip_backward = false; // elapsed time not wall time
5370 info_ptr->may_skip_forward = false; // elapsed time not wall time
5371 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5372 }
5373
5374 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5375 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5376 info_ptr->may_skip_backward = false; // elapsed time not wall time
5377 info_ptr->may_skip_forward = false; // elapsed time not wall time
5378 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5379 }
5380
5381 bool os::is_thread_cpu_time_supported() {
5382 return true;
5383 }
5384
5385 // System loadavg support. Returns -1 if load average cannot be obtained.
5386 // Linux doesn't yet have a (official) notion of processor sets,
5387 // so just return the system wide load average.
5388 int os::loadavg(double loadavg[], int nelem) {
5389 return ::getloadavg(loadavg, nelem);
5390 }
5391
5392 void os::pause() {
5393 char filename[MAX_PATH];
5394 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5395 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5396 } else {
5397 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5398 }
5399
5400 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5401 if (fd != -1) {
5402 struct stat buf;
5403 ::close(fd);
5404 while (::stat(filename, &buf) == 0) {
5405 (void)::poll(NULL, 0, 100);
5406 }
5407 } else {
5408 jio_fprintf(stderr,
5409 "Could not open pause file '%s', continuing immediately.\n", filename);
5410 }
5411 }
5412
5413
5414 // Refer to the comments in os_solaris.cpp park-unpark.
5415 //
5416 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5417 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5418 // For specifics regarding the bug see GLIBC BUGID 261237 :
5419 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5420 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5421 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5422 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5423 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5424 // and monitorenter when we're using 1-0 locking. All those operations may result in
5425 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5426 // of libpthread avoids the problem, but isn't practical.
5427 //
5428 // Possible remedies:
5429 //
5430 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5431 // This is palliative and probabilistic, however. If the thread is preempted
5432 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5433 // than the minimum period may have passed, and the abstime may be stale (in the
5434 // past) resultin in a hang. Using this technique reduces the odds of a hang
5435 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5436 //
5437 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5438 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5439 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5440 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5441 // thread.
5442 //
5443 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5444 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5445 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5446 // This also works well. In fact it avoids kernel-level scalability impediments
5447 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5448 // timers in a graceful fashion.
5449 //
5450 // 4. When the abstime value is in the past it appears that control returns
5451 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5452 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5453 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5454 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5455 // It may be possible to avoid reinitialization by checking the return
5456 // value from pthread_cond_timedwait(). In addition to reinitializing the
5457 // condvar we must establish the invariant that cond_signal() is only called
5458 // within critical sections protected by the adjunct mutex. This prevents
5459 // cond_signal() from "seeing" a condvar that's in the midst of being
5460 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5461 // desirable signal-after-unlock optimization that avoids futile context switching.
5462 //
5463 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5464 // structure when a condvar is used or initialized. cond_destroy() would
5465 // release the helper structure. Our reinitialize-after-timedwait fix
5466 // put excessive stress on malloc/free and locks protecting the c-heap.
5467 //
5468 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5469 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5470 // and only enabling the work-around for vulnerable environments.
5471
5472 // utility to compute the abstime argument to timedwait:
5473 // millis is the relative timeout time
5474 // abstime will be the absolute timeout time
5475 // TODO: replace compute_abstime() with unpackTime()
5476
5477 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5478 if (millis < 0) millis = 0;
5479
5480 jlong seconds = millis / 1000;
5481 millis %= 1000;
5482 if (seconds > 50000000) { // see man cond_timedwait(3T)
5483 seconds = 50000000;
5484 }
5485
5486 if (os::supports_monotonic_clock()) {
5487 struct timespec now;
5488 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5489 assert_status(status == 0, status, "clock_gettime");
5490 abstime->tv_sec = now.tv_sec + seconds;
5491 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5492 if (nanos >= NANOSECS_PER_SEC) {
5493 abstime->tv_sec += 1;
5494 nanos -= NANOSECS_PER_SEC;
5495 }
5496 abstime->tv_nsec = nanos;
5497 } else {
5498 struct timeval now;
5499 int status = gettimeofday(&now, NULL);
5500 assert(status == 0, "gettimeofday");
5501 abstime->tv_sec = now.tv_sec + seconds;
5502 long usec = now.tv_usec + millis * 1000;
5503 if (usec >= 1000000) {
5504 abstime->tv_sec += 1;
5505 usec -= 1000000;
5506 }
5507 abstime->tv_nsec = usec * 1000;
5508 }
5509 return abstime;
5510 }
5511
5512 void os::PlatformEvent::park() { // AKA "down()"
5513 // Invariant: Only the thread associated with the Event/PlatformEvent
5514 // may call park().
5515 // TODO: assert that _Assoc != NULL or _Assoc == Self
5516 assert(_nParked == 0, "invariant");
5517
5518 int v;
5519 for (;;) {
5520 v = _Event;
5521 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5522 }
5523 guarantee(v >= 0, "invariant");
5524 if (v == 0) {
5525 // Do this the hard way by blocking ...
5526 int status = pthread_mutex_lock(_mutex);
5527 assert_status(status == 0, status, "mutex_lock");
5528 guarantee(_nParked == 0, "invariant");
5529 ++_nParked;
5530 while (_Event < 0) {
5531 status = pthread_cond_wait(_cond, _mutex);
5532 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5533 // Treat this the same as if the wait was interrupted
5534 if (status == ETIME) { status = EINTR; }
5535 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5536 }
5537 --_nParked;
5538
5539 _Event = 0;
5540 status = pthread_mutex_unlock(_mutex);
5541 assert_status(status == 0, status, "mutex_unlock");
5542 // Paranoia to ensure our locked and lock-free paths interact
5543 // correctly with each other.
5544 OrderAccess::fence();
5545 }
5546 guarantee(_Event >= 0, "invariant");
5547 }
5548
5549 int os::PlatformEvent::park(jlong millis) {
5550 guarantee(_nParked == 0, "invariant");
5551
5552 int v;
5553 for (;;) {
5554 v = _Event;
5555 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5556 }
5557 guarantee(v >= 0, "invariant");
5558 if (v != 0) return OS_OK;
5559
5560 // We do this the hard way, by blocking the thread.
5561 // Consider enforcing a minimum timeout value.
5562 struct timespec abst;
5563 compute_abstime(&abst, millis);
5564
5565 int ret = OS_TIMEOUT;
5566 int status = pthread_mutex_lock(_mutex);
5567 assert_status(status == 0, status, "mutex_lock");
5568 guarantee(_nParked == 0, "invariant");
5569 ++_nParked;
5570
5571 // Object.wait(timo) will return because of
5572 // (a) notification
5573 // (b) timeout
5574 // (c) thread.interrupt
5575 //
5576 // Thread.interrupt and object.notify{All} both call Event::set.
5577 // That is, we treat thread.interrupt as a special case of notification.
5578 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5579 // We assume all ETIME returns are valid.
5580 //
5581 // TODO: properly differentiate simultaneous notify+interrupt.
5582 // In that case, we should propagate the notify to another waiter.
5583
5584 while (_Event < 0) {
5585 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5586 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5587 pthread_cond_destroy(_cond);
5588 pthread_cond_init(_cond, os::Linux::condAttr());
5589 }
5590 assert_status(status == 0 || status == EINTR ||
5591 status == ETIME || status == ETIMEDOUT,
5592 status, "cond_timedwait");
5593 if (!FilterSpuriousWakeups) break; // previous semantics
5594 if (status == ETIME || status == ETIMEDOUT) break;
5595 // We consume and ignore EINTR and spurious wakeups.
5596 }
5597 --_nParked;
5598 if (_Event >= 0) {
5599 ret = OS_OK;
5600 }
5601 _Event = 0;
5602 status = pthread_mutex_unlock(_mutex);
5603 assert_status(status == 0, status, "mutex_unlock");
5604 assert(_nParked == 0, "invariant");
5605 // Paranoia to ensure our locked and lock-free paths interact
5606 // correctly with each other.
5607 OrderAccess::fence();
5608 return ret;
5609 }
5610
5611 void os::PlatformEvent::unpark() {
5612 // Transitions for _Event:
5613 // 0 :=> 1
5614 // 1 :=> 1
5615 // -1 :=> either 0 or 1; must signal target thread
5616 // That is, we can safely transition _Event from -1 to either
5617 // 0 or 1.
5618 // See also: "Semaphores in Plan 9" by Mullender & Cox
5619 //
5620 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5621 // that it will take two back-to-back park() calls for the owning
5622 // thread to block. This has the benefit of forcing a spurious return
5623 // from the first park() call after an unpark() call which will help
5624 // shake out uses of park() and unpark() without condition variables.
5625
5626 if (Atomic::xchg(1, &_Event) >= 0) return;
5627
5628 // Wait for the thread associated with the event to vacate
5629 int status = pthread_mutex_lock(_mutex);
5630 assert_status(status == 0, status, "mutex_lock");
5631 int AnyWaiters = _nParked;
5632 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5633 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5634 AnyWaiters = 0;
5635 pthread_cond_signal(_cond);
5636 }
5637 status = pthread_mutex_unlock(_mutex);
5638 assert_status(status == 0, status, "mutex_unlock");
5639 if (AnyWaiters != 0) {
5640 status = pthread_cond_signal(_cond);
5641 assert_status(status == 0, status, "cond_signal");
5642 }
5643
5644 // Note that we signal() _after dropping the lock for "immortal" Events.
5645 // This is safe and avoids a common class of futile wakeups. In rare
5646 // circumstances this can cause a thread to return prematurely from
5647 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5648 // simply re-test the condition and re-park itself.
5649 }
5650
5651
5652 // JSR166
5653 // -------------------------------------------------------
5654
5655 // The solaris and linux implementations of park/unpark are fairly
5656 // conservative for now, but can be improved. They currently use a
5657 // mutex/condvar pair, plus a a count.
5658 // Park decrements count if > 0, else does a condvar wait. Unpark
5659 // sets count to 1 and signals condvar. Only one thread ever waits
5660 // on the condvar. Contention seen when trying to park implies that someone
5661 // is unparking you, so don't wait. And spurious returns are fine, so there
5662 // is no need to track notifications.
5663
5664 // This code is common to linux and solaris and will be moved to a
5665 // common place in dolphin.
5666 //
5667 // The passed in time value is either a relative time in nanoseconds
5668 // or an absolute time in milliseconds. Either way it has to be unpacked
5669 // into suitable seconds and nanoseconds components and stored in the
5670 // given timespec structure.
5671 // Given time is a 64-bit value and the time_t used in the timespec is only
5672 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5673 // overflow if times way in the future are given. Further on Solaris versions
5674 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5675 // number of seconds, in abstime, is less than current_time + 100,000,000.
5676 // As it will be 28 years before "now + 100000000" will overflow we can
5677 // ignore overflow and just impose a hard-limit on seconds using the value
5678 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5679 // years from "now".
5680
5681 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5682 assert(time > 0, "convertTime");
5683 time_t max_secs = 0;
5684
5685 if (!os::supports_monotonic_clock() || isAbsolute) {
5686 struct timeval now;
5687 int status = gettimeofday(&now, NULL);
5688 assert(status == 0, "gettimeofday");
5689
5690 max_secs = now.tv_sec + MAX_SECS;
5691
5692 if (isAbsolute) {
5693 jlong secs = time / 1000;
5694 if (secs > max_secs) {
5695 absTime->tv_sec = max_secs;
5696 } else {
5697 absTime->tv_sec = secs;
5698 }
5699 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5700 } else {
5701 jlong secs = time / NANOSECS_PER_SEC;
5702 if (secs >= MAX_SECS) {
5703 absTime->tv_sec = max_secs;
5704 absTime->tv_nsec = 0;
5705 } else {
5706 absTime->tv_sec = now.tv_sec + secs;
5707 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5708 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5709 absTime->tv_nsec -= NANOSECS_PER_SEC;
5710 ++absTime->tv_sec; // note: this must be <= max_secs
5711 }
5712 }
5713 }
5714 } else {
5715 // must be relative using monotonic clock
5716 struct timespec now;
5717 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5718 assert_status(status == 0, status, "clock_gettime");
5719 max_secs = now.tv_sec + MAX_SECS;
5720 jlong secs = time / NANOSECS_PER_SEC;
5721 if (secs >= MAX_SECS) {
5722 absTime->tv_sec = max_secs;
5723 absTime->tv_nsec = 0;
5724 } else {
5725 absTime->tv_sec = now.tv_sec + secs;
5726 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5727 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5728 absTime->tv_nsec -= NANOSECS_PER_SEC;
5729 ++absTime->tv_sec; // note: this must be <= max_secs
5730 }
5731 }
5732 }
5733 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5734 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5735 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5736 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5737 }
5738
5739 void Parker::park(bool isAbsolute, jlong time) {
5740 // Ideally we'd do something useful while spinning, such
5741 // as calling unpackTime().
5742
5743 // Optional fast-path check:
5744 // Return immediately if a permit is available.
5745 // We depend on Atomic::xchg() having full barrier semantics
5746 // since we are doing a lock-free update to _counter.
5747 if (Atomic::xchg(0, &_counter) > 0) return;
5748
5749 Thread* thread = Thread::current();
5750 assert(thread->is_Java_thread(), "Must be JavaThread");
5751 JavaThread *jt = (JavaThread *)thread;
5752
5753 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5754 // Check interrupt before trying to wait
5755 if (Thread::is_interrupted(thread, false)) {
5756 return;
5757 }
5758
5759 // Next, demultiplex/decode time arguments
5760 timespec absTime;
5761 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5762 return;
5763 }
5764 if (time > 0) {
5765 unpackTime(&absTime, isAbsolute, time);
5766 }
5767
5768
5769 // Enter safepoint region
5770 // Beware of deadlocks such as 6317397.
5771 // The per-thread Parker:: mutex is a classic leaf-lock.
5772 // In particular a thread must never block on the Threads_lock while
5773 // holding the Parker:: mutex. If safepoints are pending both the
5774 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5775 ThreadBlockInVM tbivm(jt);
5776
5777 // Don't wait if cannot get lock since interference arises from
5778 // unblocking. Also. check interrupt before trying wait
5779 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5780 return;
5781 }
5782
5783 int status;
5784 if (_counter > 0) { // no wait needed
5785 _counter = 0;
5786 status = pthread_mutex_unlock(_mutex);
5787 assert(status == 0, "invariant");
5788 // Paranoia to ensure our locked and lock-free paths interact
5789 // correctly with each other and Java-level accesses.
5790 OrderAccess::fence();
5791 return;
5792 }
5793
5794 #ifdef ASSERT
5795 // Don't catch signals while blocked; let the running threads have the signals.
5796 // (This allows a debugger to break into the running thread.)
5797 sigset_t oldsigs;
5798 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5799 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5800 #endif
5801
5802 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5803 jt->set_suspend_equivalent();
5804 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5805
5806 assert(_cur_index == -1, "invariant");
5807 if (time == 0) {
5808 _cur_index = REL_INDEX; // arbitrary choice when not timed
5809 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5810 } else {
5811 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5812 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5813 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5814 pthread_cond_destroy(&_cond[_cur_index]);
5815 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5816 }
5817 }
5818 _cur_index = -1;
5819 assert_status(status == 0 || status == EINTR ||
5820 status == ETIME || status == ETIMEDOUT,
5821 status, "cond_timedwait");
5822
5823 #ifdef ASSERT
5824 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5825 #endif
5826
5827 _counter = 0;
5828 status = pthread_mutex_unlock(_mutex);
5829 assert_status(status == 0, status, "invariant");
5830 // Paranoia to ensure our locked and lock-free paths interact
5831 // correctly with each other and Java-level accesses.
5832 OrderAccess::fence();
5833
5834 // If externally suspended while waiting, re-suspend
5835 if (jt->handle_special_suspend_equivalent_condition()) {
5836 jt->java_suspend_self();
5837 }
5838 }
5839
5840 void Parker::unpark() {
5841 int status = pthread_mutex_lock(_mutex);
5842 assert(status == 0, "invariant");
5843 const int s = _counter;
5844 _counter = 1;
5845 if (s < 1) {
5846 // thread might be parked
5847 if (_cur_index != -1) {
5848 // thread is definitely parked
5849 if (WorkAroundNPTLTimedWaitHang) {
5850 status = pthread_cond_signal(&_cond[_cur_index]);
5851 assert(status == 0, "invariant");
5852 status = pthread_mutex_unlock(_mutex);
5853 assert(status == 0, "invariant");
5854 } else {
5855 status = pthread_mutex_unlock(_mutex);
5856 assert(status == 0, "invariant");
5857 status = pthread_cond_signal(&_cond[_cur_index]);
5858 assert(status == 0, "invariant");
5859 }
5860 } else {
5861 pthread_mutex_unlock(_mutex);
5862 assert(status == 0, "invariant");
5863 }
5864 } else {
5865 pthread_mutex_unlock(_mutex);
5866 assert(status == 0, "invariant");
5867 }
5868 }
5869
5870
5871 extern char** environ;
5872
5873 #ifndef __NR_fork
5874 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5875 #endif
5876
5877 #ifndef __NR_execve
5878 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5879 #endif
5880
5881 // Run the specified command in a separate process. Return its exit value,
5882 // or -1 on failure (e.g. can't fork a new process).
5883 // Unlike system(), this function can be called from signal handler. It
5884 // doesn't block SIGINT et al.
5885 int os::fork_and_exec(char* cmd) {
5886 const char * argv[4] = {"sh", "-c", cmd, NULL};
5887
5888 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5889 // pthread_atfork handlers and reset pthread library. All we need is a
5890 // separate process to execve. Make a direct syscall to fork process.
5891 // On IA64 there's no fork syscall, we have to use fork() and hope for
5892 // the best...
5893 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5894 IA64_ONLY(fork();)
5895
5896 if (pid < 0) {
5897 // fork failed
5898 return -1;
5899
5900 } else if (pid == 0) {
5901 // child process
5902
5903 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5904 // first to kill every thread on the thread list. Because this list is
5905 // not reset by fork() (see notes above), execve() will instead kill
5906 // every thread in the parent process. We know this is the only thread
5907 // in the new process, so make a system call directly.
5908 // IA64 should use normal execve() from glibc to match the glibc fork()
5909 // above.
5910 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5911 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5912
5913 // execve failed
5914 _exit(-1);
5915
5916 } else {
5917 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5918 // care about the actual exit code, for now.
5919
5920 int status;
5921
5922 // Wait for the child process to exit. This returns immediately if
5923 // the child has already exited. */
5924 while (waitpid(pid, &status, 0) < 0) {
5925 switch (errno) {
5926 case ECHILD: return 0;
5927 case EINTR: break;
5928 default: return -1;
5929 }
5930 }
5931
5932 if (WIFEXITED(status)) {
5933 // The child exited normally; get its exit code.
5934 return WEXITSTATUS(status);
5935 } else if (WIFSIGNALED(status)) {
5936 // The child exited because of a signal
5937 // The best value to return is 0x80 + signal number,
5938 // because that is what all Unix shells do, and because
5939 // it allows callers to distinguish between process exit and
5940 // process death by signal.
5941 return 0x80 + WTERMSIG(status);
5942 } else {
5943 // Unknown exit code; pass it through
5944 return status;
5945 }
5946 }
5947 }
5948
5949 // is_headless_jre()
5950 //
5951 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5952 // in order to report if we are running in a headless jre
5953 //
5954 // Since JDK8 xawt/libmawt.so was moved into the same directory
5955 // as libawt.so, and renamed libawt_xawt.so
5956 //
5957 bool os::is_headless_jre() {
5958 struct stat statbuf;
5959 char buf[MAXPATHLEN];
5960 char libmawtpath[MAXPATHLEN];
5961 const char *xawtstr = "/xawt/libmawt.so";
5962 const char *new_xawtstr = "/libawt_xawt.so";
5963 char *p;
5964
5965 // Get path to libjvm.so
5966 os::jvm_path(buf, sizeof(buf));
5967
5968 // Get rid of libjvm.so
5969 p = strrchr(buf, '/');
5970 if (p == NULL) {
5971 return false;
5972 } else {
5973 *p = '\0';
5974 }
5975
5976 // Get rid of client or server
5977 p = strrchr(buf, '/');
5978 if (p == NULL) {
5979 return false;
5980 } else {
5981 *p = '\0';
5982 }
5983
5984 // check xawt/libmawt.so
5985 strcpy(libmawtpath, buf);
5986 strcat(libmawtpath, xawtstr);
5987 if (::stat(libmawtpath, &statbuf) == 0) return false;
5988
5989 // check libawt_xawt.so
5990 strcpy(libmawtpath, buf);
5991 strcat(libmawtpath, new_xawtstr);
5992 if (::stat(libmawtpath, &statbuf) == 0) return false;
5993
5994 return true;
5995 }
5996
5997 // Get the default path to the core file
5998 // Returns the length of the string
5999 int os::get_core_path(char* buffer, size_t bufferSize) {
6000 const char* p = get_current_directory(buffer, bufferSize);
6001
6002 if (p == NULL) {
6003 assert(p != NULL, "failed to get current directory");
6004 return 0;
6005 }
6006
6007 return strlen(buffer);
6008 }
6009
6010 #ifdef JAVASE_EMBEDDED
6011 //
6012 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6013 //
6014 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6015
6016 // ctor
6017 //
6018 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6019 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6020 _fd = fd;
6021
6022 if (os::create_thread(this, os::os_thread)) {
6023 _memnotify_thread = this;
6024 os::set_priority(this, NearMaxPriority);
6025 os::start_thread(this);
6026 }
6027 }
6028
6029 // Where all the work gets done
6030 //
6031 void MemNotifyThread::run() {
6032 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6033
6034 // Set up the select arguments
6035 fd_set rfds;
6036 if (_fd != -1) {
6037 FD_ZERO(&rfds);
6038 FD_SET(_fd, &rfds);
6039 }
6040
6041 // Now wait for the mem_notify device to wake up
6042 while (1) {
6043 // Wait for the mem_notify device to signal us..
6044 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6045 if (rc == -1) {
6046 perror("select!\n");
6047 break;
6048 } else if (rc) {
6049 //ssize_t free_before = os::available_memory();
6050 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6051
6052 // The kernel is telling us there is not much memory left...
6053 // try to do something about that
6054
6055 // If we are not already in a GC, try one.
6056 if (!Universe::heap()->is_gc_active()) {
6057 Universe::heap()->collect(GCCause::_allocation_failure);
6058
6059 //ssize_t free_after = os::available_memory();
6060 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6061 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6062 }
6063 // We might want to do something like the following if we find the GC's are not helping...
6064 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6065 }
6066 }
6067 }
6068
6069 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6070 //
6071 void MemNotifyThread::start() {
6072 int fd;
6073 fd = open("/dev/mem_notify", O_RDONLY, 0);
6074 if (fd < 0) {
6075 return;
6076 }
6077
6078 if (memnotify_thread() == NULL) {
6079 new MemNotifyThread(fd);
6080 }
6081 }
6082
6083 #endif // JAVASE_EMBEDDED
6084
6085
6086 /////////////// Unit tests ///////////////
6087
6088 #ifndef PRODUCT
6089
6090 #define test_log(...) \
6091 do { \
6092 if (VerboseInternalVMTests) { \
6093 tty->print_cr(__VA_ARGS__); \
6094 tty->flush(); \
6095 } \
6096 } while (false)
6097
6098 class TestReserveMemorySpecial : AllStatic {
6099 public:
6100 static void small_page_write(void* addr, size_t size) {
6101 size_t page_size = os::vm_page_size();
6102
6103 char* end = (char*)addr + size;
6104 for (char* p = (char*)addr; p < end; p += page_size) {
6105 *p = 1;
6106 }
6107 }
6108
6109 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6110 if (!UseHugeTLBFS) {
6111 return;
6112 }
6113
6114 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6115
6116 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6117
6118 if (addr != NULL) {
6119 small_page_write(addr, size);
6120
6121 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6122 }
6123 }
6124
6125 static void test_reserve_memory_special_huge_tlbfs_only() {
6126 if (!UseHugeTLBFS) {
6127 return;
6128 }
6129
6130 size_t lp = os::large_page_size();
6131
6132 for (size_t size = lp; size <= lp * 10; size += lp) {
6133 test_reserve_memory_special_huge_tlbfs_only(size);
6134 }
6135 }
6136
6137 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6138 if (!UseHugeTLBFS) {
6139 return;
6140 }
6141
6142 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6143 size, alignment);
6144
6145 assert(size >= os::large_page_size(), "Incorrect input to test");
6146
6147 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6148
6149 if (addr != NULL) {
6150 small_page_write(addr, size);
6151
6152 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6153 }
6154 }
6155
6156 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6157 size_t lp = os::large_page_size();
6158 size_t ag = os::vm_allocation_granularity();
6159
6160 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6161 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6162 }
6163 }
6164
6165 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6166 size_t lp = os::large_page_size();
6167 size_t ag = os::vm_allocation_granularity();
6168
6169 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6170 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6171 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6172 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6173 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6174 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6175 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6176 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6177 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6178 }
6179
6180 static void test_reserve_memory_special_huge_tlbfs() {
6181 if (!UseHugeTLBFS) {
6182 return;
6183 }
6184
6185 test_reserve_memory_special_huge_tlbfs_only();
6186 test_reserve_memory_special_huge_tlbfs_mixed();
6187 }
6188
6189 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6190 if (!UseSHM) {
6191 return;
6192 }
6193
6194 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6195
6196 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6197
6198 if (addr != NULL) {
6199 assert(is_ptr_aligned(addr, alignment), "Check");
6200 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6201
6202 small_page_write(addr, size);
6203
6204 os::Linux::release_memory_special_shm(addr, size);
6205 }
6206 }
6207
6208 static void test_reserve_memory_special_shm() {
6209 size_t lp = os::large_page_size();
6210 size_t ag = os::vm_allocation_granularity();
6211
6212 for (size_t size = ag; size < lp * 3; size += ag) {
6213 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6214 test_reserve_memory_special_shm(size, alignment);
6215 }
6216 }
6217 }
6218
6219 static void test() {
6220 test_reserve_memory_special_huge_tlbfs();
6221 test_reserve_memory_special_shm();
6222 }
6223 };
6224
6225 void TestReserveMemorySpecial_test() {
6226 TestReserveMemorySpecial::test();
6227 }
6228
6229 #endif