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 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
133 Mutex* os::Linux::_createThread_lock = NULL;
134 pthread_t os::Linux::_main_thread;
135 int os::Linux::_page_size = -1;
136 const int os::Linux::_vm_default_page_size = (8 * K);
137 bool os::Linux::_is_floating_stack = false;
138 bool os::Linux::_is_NPTL = false;
139 bool os::Linux::_supports_fast_thread_cpu_time = false;
140 const char * os::Linux::_glibc_version = NULL;
141 const char * os::Linux::_libpthread_version = NULL;
142 pthread_condattr_t os::Linux::_condattr[1];
143
144 static jlong initial_time_count=0;
145
146 static int clock_tics_per_sec = 100;
147
148 // For diagnostics to print a message once. see run_periodic_checks
149 static sigset_t check_signal_done;
150 static bool check_signals = true;
151
152 static pid_t _initial_pid = 0;
153
154 // Signal number used to suspend/resume a thread
155
156 // do not use any signal number less than SIGSEGV, see 4355769
157 static int SR_signum = SIGUSR2;
158 sigset_t SR_sigset;
159
160 // Used to protect dlsym() calls
161 static pthread_mutex_t dl_mutex;
162
163 // Declarations
164 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
165
166 #ifdef JAVASE_EMBEDDED
167 class MemNotifyThread: public Thread {
168 friend class VMStructs;
169 public:
170 virtual void run();
171
172 private:
173 static MemNotifyThread* _memnotify_thread;
174 int _fd;
175
176 public:
177
178 // Constructor
179 MemNotifyThread(int fd);
180
181 // Tester
182 bool is_memnotify_thread() const { return true; }
183
184 // Printing
185 char* name() const { return (char*)"Linux MemNotify Thread"; }
186
187 // Returns the single instance of the MemNotifyThread
188 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
189
190 // Create and start the single instance of MemNotifyThread
191 static void start();
192 };
193 #endif // JAVASE_EMBEDDED
194
195 // utility functions
196
197 static int SR_initialize();
198
199 julong os::available_memory() {
200 return Linux::available_memory();
201 }
202
203 julong os::Linux::available_memory() {
204 // values in struct sysinfo are "unsigned long"
205 struct sysinfo si;
206 sysinfo(&si);
207
208 return (julong)si.freeram * si.mem_unit;
209 }
210
211 julong os::physical_memory() {
212 return Linux::physical_memory();
213 }
214
215 ////////////////////////////////////////////////////////////////////////////////
216 // environment support
217
218 bool os::getenv(const char* name, char* buf, int len) {
219 const char* val = ::getenv(name);
220 if (val != NULL && strlen(val) < (size_t)len) {
221 strcpy(buf, val);
222 return true;
223 }
224 if (len > 0) buf[0] = 0; // return a null string
225 return false;
226 }
227
228
229 // Return true if user is running as root.
230
231 bool os::have_special_privileges() {
232 static bool init = false;
233 static bool privileges = false;
234 if (!init) {
235 privileges = (getuid() != geteuid()) || (getgid() != getegid());
236 init = true;
237 }
238 return privileges;
239 }
240
241
242 #ifndef SYS_gettid
243 // i386: 224, ia64: 1105, amd64: 186, sparc 143
244 #ifdef __ia64__
245 #define SYS_gettid 1105
246 #elif __i386__
247 #define SYS_gettid 224
248 #elif __amd64__
249 #define SYS_gettid 186
250 #elif __sparc__
251 #define SYS_gettid 143
252 #else
253 #error define gettid for the arch
254 #endif
255 #endif
256
257 // Cpu architecture string
258 #if defined(ZERO)
259 static char cpu_arch[] = ZERO_LIBARCH;
260 #elif defined(IA64)
261 static char cpu_arch[] = "ia64";
262 #elif defined(IA32)
263 static char cpu_arch[] = "i386";
264 #elif defined(AMD64)
265 static char cpu_arch[] = "amd64";
266 #elif defined(ARM)
267 static char cpu_arch[] = "arm";
268 #elif defined(PPC32)
269 static char cpu_arch[] = "ppc";
270 #elif defined(PPC64)
271 static char cpu_arch[] = "ppc64";
272 #elif defined(SPARC)
273 #ifdef _LP64
274 static char cpu_arch[] = "sparcv9";
275 #else
276 static char cpu_arch[] = "sparc";
277 #endif
278 #elif defined(AARCH64)
279 static char cpu_arch[] = "aarch64";
280 #else
281 #error Add appropriate cpu_arch setting
282 #endif
283
284
285 // pid_t gettid()
286 //
287 // Returns the kernel thread id of the currently running thread. Kernel
288 // thread id is used to access /proc.
289 //
290 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
291 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
292 //
293 pid_t os::Linux::gettid() {
294 int rslt = syscall(SYS_gettid);
295 if (rslt == -1) {
296 // old kernel, no NPTL support
297 return getpid();
298 } else {
299 return (pid_t)rslt;
300 }
301 }
302
303 // Most versions of linux have a bug where the number of processors are
304 // determined by looking at the /proc file system. In a chroot environment,
305 // the system call returns 1. This causes the VM to act as if it is
306 // a single processor and elide locking (see is_MP() call).
307 static bool unsafe_chroot_detected = false;
308 static const char *unstable_chroot_error = "/proc file system not found.\n"
309 "Java may be unstable running multithreaded in a chroot "
310 "environment on Linux when /proc filesystem is not mounted.";
311
312 void os::Linux::initialize_system_info() {
313 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
314 if (processor_count() == 1) {
315 pid_t pid = os::Linux::gettid();
316 char fname[32];
317 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
318 FILE *fp = fopen(fname, "r");
319 if (fp == NULL) {
320 unsafe_chroot_detected = true;
321 } else {
322 fclose(fp);
323 }
324 }
325 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
326 assert(processor_count() > 0, "linux error");
327 }
328
329 void os::init_system_properties_values() {
330 // The next steps are taken in the product version:
331 //
332 // Obtain the JAVA_HOME value from the location of libjvm.so.
333 // This library should be located at:
334 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
335 //
336 // If "/jre/lib/" appears at the right place in the path, then we
337 // assume libjvm.so is installed in a JDK and we use this path.
338 //
339 // Otherwise exit with message: "Could not create the Java virtual machine."
340 //
341 // The following extra steps are taken in the debugging version:
342 //
343 // If "/jre/lib/" does NOT appear at the right place in the path
344 // instead of exit check for $JAVA_HOME environment variable.
345 //
346 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
347 // then we append a fake suffix "hotspot/libjvm.so" to this path so
348 // it looks like libjvm.so is installed there
349 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
350 //
351 // Otherwise exit.
352 //
353 // Important note: if the location of libjvm.so changes this
354 // code needs to be changed accordingly.
355
356 // See ld(1):
357 // The linker uses the following search paths to locate required
358 // shared libraries:
359 // 1: ...
360 // ...
361 // 7: The default directories, normally /lib and /usr/lib.
362 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
363 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
364 #else
365 #define DEFAULT_LIBPATH "/lib:/usr/lib"
366 #endif
367
368 // Base path of extensions installed on the system.
369 #define SYS_EXT_DIR "/usr/java/packages"
370 #define EXTENSIONS_DIR "/lib/ext"
371 #define ENDORSED_DIR "/lib/endorsed"
372
373 // Buffer that fits several sprintfs.
374 // Note that the space for the colon and the trailing null are provided
375 // by the nulls included by the sizeof operator.
376 const size_t bufsize =
377 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
378 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
379 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
380 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
381
382 // sysclasspath, java_home, dll_dir
383 {
384 char *pslash;
385 os::jvm_path(buf, bufsize);
386
387 // Found the full path to libjvm.so.
388 // Now cut the path to <java_home>/jre if we can.
389 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
390 pslash = strrchr(buf, '/');
391 if (pslash != NULL) {
392 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
393 }
394 Arguments::set_dll_dir(buf);
395
396 if (pslash != NULL) {
397 pslash = strrchr(buf, '/');
398 if (pslash != NULL) {
399 *pslash = '\0'; // Get rid of /<arch>.
400 pslash = strrchr(buf, '/');
401 if (pslash != NULL) {
402 *pslash = '\0'; // Get rid of /lib.
403 }
404 }
405 }
406 Arguments::set_java_home(buf);
407 set_boot_path('/', ':');
408 }
409
410 // Where to look for native libraries.
411 //
412 // Note: Due to a legacy implementation, most of the library path
413 // is set in the launcher. This was to accomodate linking restrictions
414 // on legacy Linux implementations (which are no longer supported).
415 // Eventually, all the library path setting will be done here.
416 //
417 // However, to prevent the proliferation of improperly built native
418 // libraries, the new path component /usr/java/packages is added here.
419 // Eventually, all the library path setting will be done here.
420 {
421 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
422 // should always exist (until the legacy problem cited above is
423 // addressed).
424 const char *v = ::getenv("LD_LIBRARY_PATH");
425 const char *v_colon = ":";
426 if (v == NULL) { v = ""; v_colon = ""; }
427 // That's +1 for the colon and +1 for the trailing '\0'.
428 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
429 strlen(v) + 1 +
430 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
431 mtInternal);
432 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
433 Arguments::set_library_path(ld_library_path);
434 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
435 }
436
437 // Extensions directories.
438 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
439 Arguments::set_ext_dirs(buf);
440
441 // Endorsed standards default directory.
442 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
443 Arguments::set_endorsed_dirs(buf);
444
445 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
446
447 #undef DEFAULT_LIBPATH
448 #undef SYS_EXT_DIR
449 #undef EXTENSIONS_DIR
450 #undef ENDORSED_DIR
451 }
452
453 ////////////////////////////////////////////////////////////////////////////////
454 // breakpoint support
455
456 void os::breakpoint() {
457 BREAKPOINT;
458 }
459
460 extern "C" void breakpoint() {
461 // use debugger to set breakpoint here
462 }
463
464 ////////////////////////////////////////////////////////////////////////////////
465 // signal support
466
467 debug_only(static bool signal_sets_initialized = false);
468 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
469
470 bool os::Linux::is_sig_ignored(int sig) {
471 struct sigaction oact;
472 sigaction(sig, (struct sigaction*)NULL, &oact);
473 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
474 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
475 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
476 return true;
477 } else {
478 return false;
479 }
480 }
481
482 void os::Linux::signal_sets_init() {
483 // Should also have an assertion stating we are still single-threaded.
484 assert(!signal_sets_initialized, "Already initialized");
485 // Fill in signals that are necessarily unblocked for all threads in
486 // the VM. Currently, we unblock the following signals:
487 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
488 // by -Xrs (=ReduceSignalUsage));
489 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
490 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
491 // the dispositions or masks wrt these signals.
492 // Programs embedding the VM that want to use the above signals for their
493 // own purposes must, at this time, use the "-Xrs" option to prevent
494 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
495 // (See bug 4345157, and other related bugs).
496 // In reality, though, unblocking these signals is really a nop, since
497 // these signals are not blocked by default.
498 sigemptyset(&unblocked_sigs);
499 sigemptyset(&allowdebug_blocked_sigs);
500 sigaddset(&unblocked_sigs, SIGILL);
501 sigaddset(&unblocked_sigs, SIGSEGV);
502 sigaddset(&unblocked_sigs, SIGBUS);
503 sigaddset(&unblocked_sigs, SIGFPE);
504 #if defined(PPC64)
505 sigaddset(&unblocked_sigs, SIGTRAP);
506 #endif
507 sigaddset(&unblocked_sigs, SR_signum);
508
509 if (!ReduceSignalUsage) {
510 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
511 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
512 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
513 }
514 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
515 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
516 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
517 }
518 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
519 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
520 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
521 }
522 }
523 // Fill in signals that are blocked by all but the VM thread.
524 sigemptyset(&vm_sigs);
525 if (!ReduceSignalUsage) {
526 sigaddset(&vm_sigs, BREAK_SIGNAL);
527 }
528 debug_only(signal_sets_initialized = true);
529
530 }
531
532 // These are signals that are unblocked while a thread is running Java.
533 // (For some reason, they get blocked by default.)
534 sigset_t* os::Linux::unblocked_signals() {
535 assert(signal_sets_initialized, "Not initialized");
536 return &unblocked_sigs;
537 }
538
539 // These are the signals that are blocked while a (non-VM) thread is
540 // running Java. Only the VM thread handles these signals.
541 sigset_t* os::Linux::vm_signals() {
542 assert(signal_sets_initialized, "Not initialized");
543 return &vm_sigs;
544 }
545
546 // These are signals that are blocked during cond_wait to allow debugger in
547 sigset_t* os::Linux::allowdebug_blocked_signals() {
548 assert(signal_sets_initialized, "Not initialized");
549 return &allowdebug_blocked_sigs;
550 }
551
552 void os::Linux::hotspot_sigmask(Thread* thread) {
553
554 //Save caller's signal mask before setting VM signal mask
555 sigset_t caller_sigmask;
556 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
557
558 OSThread* osthread = thread->osthread();
559 osthread->set_caller_sigmask(caller_sigmask);
560
561 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
562
563 if (!ReduceSignalUsage) {
564 if (thread->is_VM_thread()) {
565 // Only the VM thread handles BREAK_SIGNAL ...
566 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
567 } else {
568 // ... all other threads block BREAK_SIGNAL
569 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
570 }
571 }
572 }
573
574 //////////////////////////////////////////////////////////////////////////////
575 // detecting pthread library
576
577 void os::Linux::libpthread_init() {
578 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
579 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
580 // generic name for earlier versions.
581 // Define macros here so we can build HotSpot on old systems.
582 #ifndef _CS_GNU_LIBC_VERSION
583 #define _CS_GNU_LIBC_VERSION 2
584 #endif
585 #ifndef _CS_GNU_LIBPTHREAD_VERSION
586 #define _CS_GNU_LIBPTHREAD_VERSION 3
587 #endif
588
589 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
590 if (n > 0) {
591 char *str = (char *)malloc(n, mtInternal);
592 confstr(_CS_GNU_LIBC_VERSION, str, n);
593 os::Linux::set_glibc_version(str);
594 } else {
595 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
596 static char _gnu_libc_version[32];
597 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
598 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
599 os::Linux::set_glibc_version(_gnu_libc_version);
600 }
601
602 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
603 if (n > 0) {
604 char *str = (char *)malloc(n, mtInternal);
605 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
606 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
607 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
608 // is the case. LinuxThreads has a hard limit on max number of threads.
609 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
610 // On the other hand, NPTL does not have such a limit, sysconf()
611 // will return -1 and errno is not changed. Check if it is really NPTL.
612 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
613 strstr(str, "NPTL") &&
614 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
615 free(str);
616 os::Linux::set_libpthread_version("linuxthreads");
617 } else {
618 os::Linux::set_libpthread_version(str);
619 }
620 } else {
621 // glibc before 2.3.2 only has LinuxThreads.
622 os::Linux::set_libpthread_version("linuxthreads");
623 }
624
625 if (strstr(libpthread_version(), "NPTL")) {
626 os::Linux::set_is_NPTL();
627 } else {
628 os::Linux::set_is_LinuxThreads();
629 }
630
631 // LinuxThreads have two flavors: floating-stack mode, which allows variable
632 // stack size; and fixed-stack mode. NPTL is always floating-stack.
633 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
634 os::Linux::set_is_floating_stack();
635 }
636 }
637
638 /////////////////////////////////////////////////////////////////////////////
639 // thread stack
640
641 // Force Linux kernel to expand current thread stack. If "bottom" is close
642 // to the stack guard, caller should block all signals.
643 //
644 // MAP_GROWSDOWN:
645 // A special mmap() flag that is used to implement thread stacks. It tells
646 // kernel that the memory region should extend downwards when needed. This
647 // allows early versions of LinuxThreads to only mmap the first few pages
648 // when creating a new thread. Linux kernel will automatically expand thread
649 // stack as needed (on page faults).
650 //
651 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
652 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
653 // region, it's hard to tell if the fault is due to a legitimate stack
654 // access or because of reading/writing non-exist memory (e.g. buffer
655 // overrun). As a rule, if the fault happens below current stack pointer,
656 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
657 // application (see Linux kernel fault.c).
658 //
659 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
660 // stack overflow detection.
661 //
662 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
663 // not use this flag. However, the stack of initial thread is not created
664 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
665 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
666 // and then attach the thread to JVM.
667 //
668 // To get around the problem and allow stack banging on Linux, we need to
669 // manually expand thread stack after receiving the SIGSEGV.
670 //
671 // There are two ways to expand thread stack to address "bottom", we used
672 // both of them in JVM before 1.5:
673 // 1. adjust stack pointer first so that it is below "bottom", and then
674 // touch "bottom"
675 // 2. mmap() the page in question
676 //
677 // Now alternate signal stack is gone, it's harder to use 2. For instance,
678 // if current sp is already near the lower end of page 101, and we need to
679 // call mmap() to map page 100, it is possible that part of the mmap() frame
680 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
681 // That will destroy the mmap() frame and cause VM to crash.
682 //
683 // The following code works by adjusting sp first, then accessing the "bottom"
684 // page to force a page fault. Linux kernel will then automatically expand the
685 // stack mapping.
686 //
687 // _expand_stack_to() assumes its frame size is less than page size, which
688 // should always be true if the function is not inlined.
689
690 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
691 #define NOINLINE
692 #else
693 #define NOINLINE __attribute__ ((noinline))
694 #endif
695
696 static void _expand_stack_to(address bottom) NOINLINE;
697
698 static void _expand_stack_to(address bottom) {
699 address sp;
700 size_t size;
701 volatile char *p;
702
703 // Adjust bottom to point to the largest address within the same page, it
704 // gives us a one-page buffer if alloca() allocates slightly more memory.
705 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
706 bottom += os::Linux::page_size() - 1;
707
708 // sp might be slightly above current stack pointer; if that's the case, we
709 // will alloca() a little more space than necessary, which is OK. Don't use
710 // os::current_stack_pointer(), as its result can be slightly below current
711 // stack pointer, causing us to not alloca enough to reach "bottom".
712 sp = (address)&sp;
713
714 if (sp > bottom) {
715 size = sp - bottom;
716 p = (volatile char *)alloca(size);
717 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
718 p[0] = '\0';
719 }
720 }
721
722 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
723 assert(t!=NULL, "just checking");
724 assert(t->osthread()->expanding_stack(), "expand should be set");
725 assert(t->stack_base() != NULL, "stack_base was not initialized");
726
727 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
728 sigset_t mask_all, old_sigset;
729 sigfillset(&mask_all);
730 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
731 _expand_stack_to(addr);
732 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
733 return true;
734 }
735 return false;
736 }
737
738 //////////////////////////////////////////////////////////////////////////////
739 // create new thread
740
741 static address highest_vm_reserved_address();
742
743 // check if it's safe to start a new thread
744 static bool _thread_safety_check(Thread* thread) {
745 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
746 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
747 // Heap is mmap'ed at lower end of memory space. Thread stacks are
748 // allocated (MAP_FIXED) from high address space. Every thread stack
749 // occupies a fixed size slot (usually 2Mbytes, but user can change
750 // it to other values if they rebuild LinuxThreads).
751 //
752 // Problem with MAP_FIXED is that mmap() can still succeed even part of
753 // the memory region has already been mmap'ed. That means if we have too
754 // many threads and/or very large heap, eventually thread stack will
755 // collide with heap.
756 //
757 // Here we try to prevent heap/stack collision by comparing current
758 // stack bottom with the highest address that has been mmap'ed by JVM
759 // plus a safety margin for memory maps created by native code.
760 //
761 // This feature can be disabled by setting ThreadSafetyMargin to 0
762 //
763 if (ThreadSafetyMargin > 0) {
764 address stack_bottom = os::current_stack_base() - os::current_stack_size();
765
766 // not safe if our stack extends below the safety margin
767 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
768 } else {
769 return true;
770 }
771 } else {
772 // Floating stack LinuxThreads or NPTL:
773 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
774 // there's not enough space left, pthread_create() will fail. If we come
775 // here, that means enough space has been reserved for stack.
776 return true;
777 }
778 }
779
780 // Thread start routine for all newly created threads
781 static void *java_start(Thread *thread) {
782 // Try to randomize the cache line index of hot stack frames.
783 // This helps when threads of the same stack traces evict each other's
784 // cache lines. The threads can be either from the same JVM instance, or
785 // from different JVM instances. The benefit is especially true for
786 // processors with hyperthreading technology.
787 static int counter = 0;
788 int pid = os::current_process_id();
789 alloca(((pid ^ counter++) & 7) * 128);
790
791 ThreadLocalStorage::set_thread(thread);
792
793 OSThread* osthread = thread->osthread();
794 Monitor* sync = osthread->startThread_lock();
795
796 // non floating stack LinuxThreads needs extra check, see above
797 if (!_thread_safety_check(thread)) {
798 // notify parent thread
799 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
800 osthread->set_state(ZOMBIE);
801 sync->notify_all();
802 return NULL;
803 }
804
805 // thread_id is kernel thread id (similar to Solaris LWP id)
806 osthread->set_thread_id(os::Linux::gettid());
807
808 if (UseNUMA) {
809 int lgrp_id = os::numa_get_group_id();
810 if (lgrp_id != -1) {
811 thread->set_lgrp_id(lgrp_id);
812 }
813 }
814 // initialize signal mask for this thread
815 os::Linux::hotspot_sigmask(thread);
816
817 // initialize floating point control register
818 os::Linux::init_thread_fpu_state();
819
820 // handshaking with parent thread
821 {
822 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
823
824 // notify parent thread
825 osthread->set_state(INITIALIZED);
826 sync->notify_all();
827
828 // wait until os::start_thread()
829 while (osthread->get_state() == INITIALIZED) {
830 sync->wait(Mutex::_no_safepoint_check_flag);
831 }
832 }
833
834 // call one more level start routine
835 thread->run();
836
837 return 0;
838 }
839
840 bool os::create_thread(Thread* thread, ThreadType thr_type,
841 size_t stack_size) {
842 assert(thread->osthread() == NULL, "caller responsible");
843
844 // Allocate the OSThread object
845 OSThread* osthread = new OSThread(NULL, NULL);
846 if (osthread == NULL) {
847 return false;
848 }
849
850 // set the correct thread state
851 osthread->set_thread_type(thr_type);
852
853 // Initial state is ALLOCATED but not INITIALIZED
854 osthread->set_state(ALLOCATED);
855
856 thread->set_osthread(osthread);
857
858 // init thread attributes
859 pthread_attr_t attr;
860 pthread_attr_init(&attr);
861 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
862
863 // stack size
864 if (os::Linux::supports_variable_stack_size()) {
865 // calculate stack size if it's not specified by caller
866 if (stack_size == 0) {
867 stack_size = os::Linux::default_stack_size(thr_type);
868
869 switch (thr_type) {
870 case os::java_thread:
871 // Java threads use ThreadStackSize which default value can be
872 // changed with the flag -Xss
873 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
874 stack_size = JavaThread::stack_size_at_create();
875 break;
876 case os::compiler_thread:
877 if (CompilerThreadStackSize > 0) {
878 stack_size = (size_t)(CompilerThreadStackSize * K);
879 break;
880 } // else fall through:
881 // use VMThreadStackSize if CompilerThreadStackSize is not defined
882 case os::vm_thread:
883 case os::pgc_thread:
884 case os::cgc_thread:
885 case os::watcher_thread:
886 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
887 break;
888 }
889 }
890
891 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
892 pthread_attr_setstacksize(&attr, stack_size);
893 } else {
894 // let pthread_create() pick the default value.
895 }
896
897 // glibc guard page
898 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
899
900 ThreadState state;
901
902 {
903 // Serialize thread creation if we are running with fixed stack LinuxThreads
904 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
905 if (lock) {
906 os::Linux::createThread_lock()->lock_without_safepoint_check();
907 }
908
909 pthread_t tid;
910 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
911
912 pthread_attr_destroy(&attr);
913
914 if (ret != 0) {
915 if (PrintMiscellaneous && (Verbose || WizardMode)) {
916 perror("pthread_create()");
917 }
918 // Need to clean up stuff we've allocated so far
919 thread->set_osthread(NULL);
920 delete osthread;
921 if (lock) os::Linux::createThread_lock()->unlock();
922 return false;
923 }
924
925 // Store pthread info into the OSThread
926 osthread->set_pthread_id(tid);
927
928 // Wait until child thread is either initialized or aborted
929 {
930 Monitor* sync_with_child = osthread->startThread_lock();
931 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
932 while ((state = osthread->get_state()) == ALLOCATED) {
933 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
934 }
935 }
936
937 if (lock) {
938 os::Linux::createThread_lock()->unlock();
939 }
940 }
941
942 // Aborted due to thread limit being reached
943 if (state == ZOMBIE) {
944 thread->set_osthread(NULL);
945 delete osthread;
946 return false;
947 }
948
949 // The thread is returned suspended (in state INITIALIZED),
950 // and is started higher up in the call chain
951 assert(state == INITIALIZED, "race condition");
952 return true;
953 }
954
955 /////////////////////////////////////////////////////////////////////////////
956 // attach existing thread
957
958 // bootstrap the main thread
959 bool os::create_main_thread(JavaThread* thread) {
960 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
961 return create_attached_thread(thread);
962 }
963
964 bool os::create_attached_thread(JavaThread* thread) {
965 #ifdef ASSERT
966 thread->verify_not_published();
967 #endif
968
969 // Allocate the OSThread object
970 OSThread* osthread = new OSThread(NULL, NULL);
971
972 if (osthread == NULL) {
973 return false;
974 }
975
976 // Store pthread info into the OSThread
977 osthread->set_thread_id(os::Linux::gettid());
978 osthread->set_pthread_id(::pthread_self());
979
980 // initialize floating point control register
981 os::Linux::init_thread_fpu_state();
982
983 // Initial thread state is RUNNABLE
984 osthread->set_state(RUNNABLE);
985
986 thread->set_osthread(osthread);
987
988 if (UseNUMA) {
989 int lgrp_id = os::numa_get_group_id();
990 if (lgrp_id != -1) {
991 thread->set_lgrp_id(lgrp_id);
992 }
993 }
994
995 if (os::Linux::is_initial_thread()) {
996 // If current thread is initial thread, its stack is mapped on demand,
997 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
998 // the entire stack region to avoid SEGV in stack banging.
999 // It is also useful to get around the heap-stack-gap problem on SuSE
1000 // kernel (see 4821821 for details). We first expand stack to the top
1001 // of yellow zone, then enable stack yellow zone (order is significant,
1002 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1003 // is no gap between the last two virtual memory regions.
1004
1005 JavaThread *jt = (JavaThread *)thread;
1006 address addr = jt->stack_yellow_zone_base();
1007 assert(addr != NULL, "initialization problem?");
1008 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1009
1010 osthread->set_expanding_stack();
1011 os::Linux::manually_expand_stack(jt, addr);
1012 osthread->clear_expanding_stack();
1013 }
1014
1015 // initialize signal mask for this thread
1016 // and save the caller's signal mask
1017 os::Linux::hotspot_sigmask(thread);
1018
1019 return true;
1020 }
1021
1022 void os::pd_start_thread(Thread* thread) {
1023 OSThread * osthread = thread->osthread();
1024 assert(osthread->get_state() != INITIALIZED, "just checking");
1025 Monitor* sync_with_child = osthread->startThread_lock();
1026 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1027 sync_with_child->notify();
1028 }
1029
1030 // Free Linux resources related to the OSThread
1031 void os::free_thread(OSThread* osthread) {
1032 assert(osthread != NULL, "osthread not set");
1033
1034 if (Thread::current()->osthread() == osthread) {
1035 // Restore caller's signal mask
1036 sigset_t sigmask = osthread->caller_sigmask();
1037 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1038 }
1039
1040 delete osthread;
1041 }
1042
1043 //////////////////////////////////////////////////////////////////////////////
1044 // thread local storage
1045
1046 // Restore the thread pointer if the destructor is called. This is in case
1047 // someone from JNI code sets up a destructor with pthread_key_create to run
1048 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1049 // will hang or crash. When detachCurrentThread is called the key will be set
1050 // to null and we will not be called again. If detachCurrentThread is never
1051 // called we could loop forever depending on the pthread implementation.
1052 static void restore_thread_pointer(void* p) {
1053 Thread* thread = (Thread*) p;
1054 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1055 }
1056
1057 int os::allocate_thread_local_storage() {
1058 pthread_key_t key;
1059 int rslt = pthread_key_create(&key, restore_thread_pointer);
1060 assert(rslt == 0, "cannot allocate thread local storage");
1061 return (int)key;
1062 }
1063
1064 // Note: This is currently not used by VM, as we don't destroy TLS key
1065 // on VM exit.
1066 void os::free_thread_local_storage(int index) {
1067 int rslt = pthread_key_delete((pthread_key_t)index);
1068 assert(rslt == 0, "invalid index");
1069 }
1070
1071 void os::thread_local_storage_at_put(int index, void* value) {
1072 int rslt = pthread_setspecific((pthread_key_t)index, value);
1073 assert(rslt == 0, "pthread_setspecific failed");
1074 }
1075
1076 extern "C" Thread* get_thread() {
1077 return ThreadLocalStorage::thread();
1078 }
1079
1080 //////////////////////////////////////////////////////////////////////////////
1081 // initial thread
1082
1083 // Check if current thread is the initial thread, similar to Solaris thr_main.
1084 bool os::Linux::is_initial_thread(void) {
1085 char dummy;
1086 // If called before init complete, thread stack bottom will be null.
1087 // Can be called if fatal error occurs before initialization.
1088 if (initial_thread_stack_bottom() == NULL) return false;
1089 assert(initial_thread_stack_bottom() != NULL &&
1090 initial_thread_stack_size() != 0,
1091 "os::init did not locate initial thread's stack region");
1092 if ((address)&dummy >= initial_thread_stack_bottom() &&
1093 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1094 return true;
1095 } else {
1096 return false;
1097 }
1098 }
1099
1100 // Find the virtual memory area that contains addr
1101 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1102 FILE *fp = fopen("/proc/self/maps", "r");
1103 if (fp) {
1104 address low, high;
1105 while (!feof(fp)) {
1106 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1107 if (low <= addr && addr < high) {
1108 if (vma_low) *vma_low = low;
1109 if (vma_high) *vma_high = high;
1110 fclose(fp);
1111 return true;
1112 }
1113 }
1114 for (;;) {
1115 int ch = fgetc(fp);
1116 if (ch == EOF || ch == (int)'\n') break;
1117 }
1118 }
1119 fclose(fp);
1120 }
1121 return false;
1122 }
1123
1124 // Locate initial thread stack. This special handling of initial thread stack
1125 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1126 // bogus value for initial thread.
1127 void os::Linux::capture_initial_stack(size_t max_size) {
1128 // stack size is the easy part, get it from RLIMIT_STACK
1129 size_t stack_size;
1130 struct rlimit rlim;
1131 getrlimit(RLIMIT_STACK, &rlim);
1132 stack_size = rlim.rlim_cur;
1133
1134 // 6308388: a bug in ld.so will relocate its own .data section to the
1135 // lower end of primordial stack; reduce ulimit -s value a little bit
1136 // so we won't install guard page on ld.so's data section.
1137 stack_size -= 2 * page_size();
1138
1139 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1140 // 7.1, in both cases we will get 2G in return value.
1141 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1142 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1143 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1144 // in case other parts in glibc still assumes 2M max stack size.
1145 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1146 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1147 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1148 stack_size = 2 * K * K IA64_ONLY(*2);
1149 }
1150 // Try to figure out where the stack base (top) is. This is harder.
1151 //
1152 // When an application is started, glibc saves the initial stack pointer in
1153 // a global variable "__libc_stack_end", which is then used by system
1154 // libraries. __libc_stack_end should be pretty close to stack top. The
1155 // variable is available since the very early days. However, because it is
1156 // a private interface, it could disappear in the future.
1157 //
1158 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1159 // to __libc_stack_end, it is very close to stack top, but isn't the real
1160 // stack top. Note that /proc may not exist if VM is running as a chroot
1161 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1162 // /proc/<pid>/stat could change in the future (though unlikely).
1163 //
1164 // We try __libc_stack_end first. If that doesn't work, look for
1165 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1166 // as a hint, which should work well in most cases.
1167
1168 uintptr_t stack_start;
1169
1170 // try __libc_stack_end first
1171 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1172 if (p && *p) {
1173 stack_start = *p;
1174 } else {
1175 // see if we can get the start_stack field from /proc/self/stat
1176 FILE *fp;
1177 int pid;
1178 char state;
1179 int ppid;
1180 int pgrp;
1181 int session;
1182 int nr;
1183 int tpgrp;
1184 unsigned long flags;
1185 unsigned long minflt;
1186 unsigned long cminflt;
1187 unsigned long majflt;
1188 unsigned long cmajflt;
1189 unsigned long utime;
1190 unsigned long stime;
1191 long cutime;
1192 long cstime;
1193 long prio;
1194 long nice;
1195 long junk;
1196 long it_real;
1197 uintptr_t start;
1198 uintptr_t vsize;
1199 intptr_t rss;
1200 uintptr_t rsslim;
1201 uintptr_t scodes;
1202 uintptr_t ecode;
1203 int i;
1204
1205 // Figure what the primordial thread stack base is. Code is inspired
1206 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1207 // followed by command name surrounded by parentheses, state, etc.
1208 char stat[2048];
1209 int statlen;
1210
1211 fp = fopen("/proc/self/stat", "r");
1212 if (fp) {
1213 statlen = fread(stat, 1, 2047, fp);
1214 stat[statlen] = '\0';
1215 fclose(fp);
1216
1217 // Skip pid and the command string. Note that we could be dealing with
1218 // weird command names, e.g. user could decide to rename java launcher
1219 // to "java 1.4.2 :)", then the stat file would look like
1220 // 1234 (java 1.4.2 :)) R ... ...
1221 // We don't really need to know the command string, just find the last
1222 // occurrence of ")" and then start parsing from there. See bug 4726580.
1223 char * s = strrchr(stat, ')');
1224
1225 i = 0;
1226 if (s) {
1227 // Skip blank chars
1228 do s++; while (isspace(*s));
1229
1230 #define _UFM UINTX_FORMAT
1231 #define _DFM INTX_FORMAT
1232
1233 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1234 // 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8
1235 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1236 &state, // 3 %c
1237 &ppid, // 4 %d
1238 &pgrp, // 5 %d
1239 &session, // 6 %d
1240 &nr, // 7 %d
1241 &tpgrp, // 8 %d
1242 &flags, // 9 %lu
1243 &minflt, // 10 %lu
1244 &cminflt, // 11 %lu
1245 &majflt, // 12 %lu
1246 &cmajflt, // 13 %lu
1247 &utime, // 14 %lu
1248 &stime, // 15 %lu
1249 &cutime, // 16 %ld
1250 &cstime, // 17 %ld
1251 &prio, // 18 %ld
1252 &nice, // 19 %ld
1253 &junk, // 20 %ld
1254 &it_real, // 21 %ld
1255 &start, // 22 UINTX_FORMAT
1256 &vsize, // 23 UINTX_FORMAT
1257 &rss, // 24 INTX_FORMAT
1258 &rsslim, // 25 UINTX_FORMAT
1259 &scodes, // 26 UINTX_FORMAT
1260 &ecode, // 27 UINTX_FORMAT
1261 &stack_start); // 28 UINTX_FORMAT
1262 }
1263
1264 #undef _UFM
1265 #undef _DFM
1266
1267 if (i != 28 - 2) {
1268 assert(false, "Bad conversion from /proc/self/stat");
1269 // product mode - assume we are the initial thread, good luck in the
1270 // embedded case.
1271 warning("Can't detect initial thread stack location - bad conversion");
1272 stack_start = (uintptr_t) &rlim;
1273 }
1274 } else {
1275 // For some reason we can't open /proc/self/stat (for example, running on
1276 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1277 // most cases, so don't abort:
1278 warning("Can't detect initial thread stack location - no /proc/self/stat");
1279 stack_start = (uintptr_t) &rlim;
1280 }
1281 }
1282
1283 // Now we have a pointer (stack_start) very close to the stack top, the
1284 // next thing to do is to figure out the exact location of stack top. We
1285 // can find out the virtual memory area that contains stack_start by
1286 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1287 // and its upper limit is the real stack top. (again, this would fail if
1288 // running inside chroot, because /proc may not exist.)
1289
1290 uintptr_t stack_top;
1291 address low, high;
1292 if (find_vma((address)stack_start, &low, &high)) {
1293 // success, "high" is the true stack top. (ignore "low", because initial
1294 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1295 stack_top = (uintptr_t)high;
1296 } else {
1297 // failed, likely because /proc/self/maps does not exist
1298 warning("Can't detect initial thread stack location - find_vma failed");
1299 // best effort: stack_start is normally within a few pages below the real
1300 // stack top, use it as stack top, and reduce stack size so we won't put
1301 // guard page outside stack.
1302 stack_top = stack_start;
1303 stack_size -= 16 * page_size();
1304 }
1305
1306 // stack_top could be partially down the page so align it
1307 stack_top = align_size_up(stack_top, page_size());
1308
1309 if (max_size && stack_size > max_size) {
1310 _initial_thread_stack_size = max_size;
1311 } else {
1312 _initial_thread_stack_size = stack_size;
1313 }
1314
1315 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1316 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1317 }
1318
1319 ////////////////////////////////////////////////////////////////////////////////
1320 // time support
1321
1322 // Time since start-up in seconds to a fine granularity.
1323 // Used by VMSelfDestructTimer and the MemProfiler.
1324 double os::elapsedTime() {
1325
1326 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1327 }
1328
1329 jlong os::elapsed_counter() {
1330 return javaTimeNanos() - initial_time_count;
1331 }
1332
1333 jlong os::elapsed_frequency() {
1334 return NANOSECS_PER_SEC; // nanosecond resolution
1335 }
1336
1337 bool os::supports_vtime() { return true; }
1338 bool os::enable_vtime() { return false; }
1339 bool os::vtime_enabled() { return false; }
1340
1341 double os::elapsedVTime() {
1342 struct rusage usage;
1343 int retval = getrusage(RUSAGE_THREAD, &usage);
1344 if (retval == 0) {
1345 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);
1346 } else {
1347 // better than nothing, but not much
1348 return elapsedTime();
1349 }
1350 }
1351
1352 jlong os::javaTimeMillis() {
1353 timeval time;
1354 int status = gettimeofday(&time, NULL);
1355 assert(status != -1, "linux error");
1356 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1357 }
1358
1359 #ifndef CLOCK_MONOTONIC
1360 #define CLOCK_MONOTONIC (1)
1361 #endif
1362
1363 void os::Linux::clock_init() {
1364 // we do dlopen's in this particular order due to bug in linux
1365 // dynamical loader (see 6348968) leading to crash on exit
1366 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1367 if (handle == NULL) {
1368 handle = dlopen("librt.so", RTLD_LAZY);
1369 }
1370
1371 if (handle) {
1372 int (*clock_getres_func)(clockid_t, struct timespec*) =
1373 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1374 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1375 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1376 if (clock_getres_func && clock_gettime_func) {
1377 // See if monotonic clock is supported by the kernel. Note that some
1378 // early implementations simply return kernel jiffies (updated every
1379 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1380 // for nano time (though the monotonic property is still nice to have).
1381 // It's fixed in newer kernels, however clock_getres() still returns
1382 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1383 // resolution for now. Hopefully as people move to new kernels, this
1384 // won't be a problem.
1385 struct timespec res;
1386 struct timespec tp;
1387 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1388 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1389 // yes, monotonic clock is supported
1390 _clock_gettime = clock_gettime_func;
1391 return;
1392 } else {
1393 // close librt if there is no monotonic clock
1394 dlclose(handle);
1395 }
1396 }
1397 }
1398 warning("No monotonic clock was available - timed services may " \
1399 "be adversely affected if the time-of-day clock changes");
1400 }
1401
1402 #ifndef SYS_clock_getres
1403 #if defined(IA32) || defined(AMD64)
1404 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1405 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1406 #else
1407 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1408 #define sys_clock_getres(x,y) -1
1409 #endif
1410 #else
1411 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1412 #endif
1413
1414 void os::Linux::fast_thread_clock_init() {
1415 if (!UseLinuxPosixThreadCPUClocks) {
1416 return;
1417 }
1418 clockid_t clockid;
1419 struct timespec tp;
1420 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1421 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1422
1423 // Switch to using fast clocks for thread cpu time if
1424 // the sys_clock_getres() returns 0 error code.
1425 // Note, that some kernels may support the current thread
1426 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1427 // returned by the pthread_getcpuclockid().
1428 // If the fast Posix clocks are supported then the sys_clock_getres()
1429 // must return at least tp.tv_sec == 0 which means a resolution
1430 // better than 1 sec. This is extra check for reliability.
1431
1432 if (pthread_getcpuclockid_func &&
1433 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1434 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1435 _supports_fast_thread_cpu_time = true;
1436 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1437 }
1438 }
1439
1440 jlong os::javaTimeNanos() {
1441 if (os::supports_monotonic_clock()) {
1442 struct timespec tp;
1443 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1444 assert(status == 0, "gettime error");
1445 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1446 return result;
1447 } else {
1448 timeval time;
1449 int status = gettimeofday(&time, NULL);
1450 assert(status != -1, "linux error");
1451 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1452 return 1000 * usecs;
1453 }
1454 }
1455
1456 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1457 if (os::supports_monotonic_clock()) {
1458 info_ptr->max_value = ALL_64_BITS;
1459
1460 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1461 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1462 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1463 } else {
1464 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1465 info_ptr->max_value = ALL_64_BITS;
1466
1467 // gettimeofday is a real time clock so it skips
1468 info_ptr->may_skip_backward = true;
1469 info_ptr->may_skip_forward = true;
1470 }
1471
1472 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1473 }
1474
1475 // Return the real, user, and system times in seconds from an
1476 // arbitrary fixed point in the past.
1477 bool os::getTimesSecs(double* process_real_time,
1478 double* process_user_time,
1479 double* process_system_time) {
1480 struct tms ticks;
1481 clock_t real_ticks = times(&ticks);
1482
1483 if (real_ticks == (clock_t) (-1)) {
1484 return false;
1485 } else {
1486 double ticks_per_second = (double) clock_tics_per_sec;
1487 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1488 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1489 *process_real_time = ((double) real_ticks) / ticks_per_second;
1490
1491 return true;
1492 }
1493 }
1494
1495
1496 char * os::local_time_string(char *buf, size_t buflen) {
1497 struct tm t;
1498 time_t long_time;
1499 time(&long_time);
1500 localtime_r(&long_time, &t);
1501 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1502 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1503 t.tm_hour, t.tm_min, t.tm_sec);
1504 return buf;
1505 }
1506
1507 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1508 return localtime_r(clock, res);
1509 }
1510
1511 ////////////////////////////////////////////////////////////////////////////////
1512 // runtime exit support
1513
1514 // Note: os::shutdown() might be called very early during initialization, or
1515 // called from signal handler. Before adding something to os::shutdown(), make
1516 // sure it is async-safe and can handle partially initialized VM.
1517 void os::shutdown() {
1518
1519 // allow PerfMemory to attempt cleanup of any persistent resources
1520 perfMemory_exit();
1521
1522 // needs to remove object in file system
1523 AttachListener::abort();
1524
1525 // flush buffered output, finish log files
1526 ostream_abort();
1527
1528 // Check for abort hook
1529 abort_hook_t abort_hook = Arguments::abort_hook();
1530 if (abort_hook != NULL) {
1531 abort_hook();
1532 }
1533
1534 }
1535
1536 // Note: os::abort() might be called very early during initialization, or
1537 // called from signal handler. Before adding something to os::abort(), make
1538 // sure it is async-safe and can handle partially initialized VM.
1539 void os::abort(bool dump_core) {
1540 os::shutdown();
1541 if (dump_core) {
1542 #ifndef PRODUCT
1543 fdStream out(defaultStream::output_fd());
1544 out.print_raw("Current thread is ");
1545 char buf[16];
1546 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1547 out.print_raw_cr(buf);
1548 out.print_raw_cr("Dumping core ...");
1549 #endif
1550 ::abort(); // dump core
1551 }
1552
1553 ::exit(1);
1554 }
1555
1556 // Die immediately, no exit hook, no abort hook, no cleanup.
1557 void os::die() {
1558 // _exit() on LinuxThreads only kills current thread
1559 ::abort();
1560 }
1561
1562
1563 // This method is a copy of JDK's sysGetLastErrorString
1564 // from src/solaris/hpi/src/system_md.c
1565
1566 size_t os::lasterror(char *buf, size_t len) {
1567 if (errno == 0) return 0;
1568
1569 const char *s = ::strerror(errno);
1570 size_t n = ::strlen(s);
1571 if (n >= len) {
1572 n = len - 1;
1573 }
1574 ::strncpy(buf, s, n);
1575 buf[n] = '\0';
1576 return n;
1577 }
1578
1579 intx os::current_thread_id() { return (intx)pthread_self(); }
1580 int os::current_process_id() {
1581
1582 // Under the old linux thread library, linux gives each thread
1583 // its own process id. Because of this each thread will return
1584 // a different pid if this method were to return the result
1585 // of getpid(2). Linux provides no api that returns the pid
1586 // of the launcher thread for the vm. This implementation
1587 // returns a unique pid, the pid of the launcher thread
1588 // that starts the vm 'process'.
1589
1590 // Under the NPTL, getpid() returns the same pid as the
1591 // launcher thread rather than a unique pid per thread.
1592 // Use gettid() if you want the old pre NPTL behaviour.
1593
1594 // if you are looking for the result of a call to getpid() that
1595 // returns a unique pid for the calling thread, then look at the
1596 // OSThread::thread_id() method in osThread_linux.hpp file
1597
1598 return (int)(_initial_pid ? _initial_pid : getpid());
1599 }
1600
1601 // DLL functions
1602
1603 const char* os::dll_file_extension() { return ".so"; }
1604
1605 // This must be hard coded because it's the system's temporary
1606 // directory not the java application's temp directory, ala java.io.tmpdir.
1607 const char* os::get_temp_directory() { return "/tmp"; }
1608
1609 static bool file_exists(const char* filename) {
1610 struct stat statbuf;
1611 if (filename == NULL || strlen(filename) == 0) {
1612 return false;
1613 }
1614 return os::stat(filename, &statbuf) == 0;
1615 }
1616
1617 bool os::dll_build_name(char* buffer, size_t buflen,
1618 const char* pname, const char* fname) {
1619 bool retval = false;
1620 // Copied from libhpi
1621 const size_t pnamelen = pname ? strlen(pname) : 0;
1622
1623 // Return error on buffer overflow.
1624 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1625 return retval;
1626 }
1627
1628 if (pnamelen == 0) {
1629 snprintf(buffer, buflen, "lib%s.so", fname);
1630 retval = true;
1631 } else if (strchr(pname, *os::path_separator()) != NULL) {
1632 int n;
1633 char** pelements = split_path(pname, &n);
1634 if (pelements == NULL) {
1635 return false;
1636 }
1637 for (int i = 0; i < n; i++) {
1638 // Really shouldn't be NULL, but check can't hurt
1639 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1640 continue; // skip the empty path values
1641 }
1642 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1643 if (file_exists(buffer)) {
1644 retval = true;
1645 break;
1646 }
1647 }
1648 // release the storage
1649 for (int i = 0; i < n; i++) {
1650 if (pelements[i] != NULL) {
1651 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1652 }
1653 }
1654 if (pelements != NULL) {
1655 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1656 }
1657 } else {
1658 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1659 retval = true;
1660 }
1661 return retval;
1662 }
1663
1664 // check if addr is inside libjvm.so
1665 bool os::address_is_in_vm(address addr) {
1666 static address libjvm_base_addr;
1667 Dl_info dlinfo;
1668
1669 if (libjvm_base_addr == NULL) {
1670 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1671 libjvm_base_addr = (address)dlinfo.dli_fbase;
1672 }
1673 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1674 }
1675
1676 if (dladdr((void *)addr, &dlinfo) != 0) {
1677 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1678 }
1679
1680 return false;
1681 }
1682
1683 bool os::dll_address_to_function_name(address addr, char *buf,
1684 int buflen, int *offset) {
1685 // buf is not optional, but offset is optional
1686 assert(buf != NULL, "sanity check");
1687
1688 Dl_info dlinfo;
1689
1690 if (dladdr((void*)addr, &dlinfo) != 0) {
1691 // see if we have a matching symbol
1692 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1693 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1694 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1695 }
1696 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1697 return true;
1698 }
1699 // no matching symbol so try for just file info
1700 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1701 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1702 buf, buflen, offset, dlinfo.dli_fname)) {
1703 return true;
1704 }
1705 }
1706 }
1707
1708 buf[0] = '\0';
1709 if (offset != NULL) *offset = -1;
1710 return false;
1711 }
1712
1713 struct _address_to_library_name {
1714 address addr; // input : memory address
1715 size_t buflen; // size of fname
1716 char* fname; // output: library name
1717 address base; // library base addr
1718 };
1719
1720 static int address_to_library_name_callback(struct dl_phdr_info *info,
1721 size_t size, void *data) {
1722 int i;
1723 bool found = false;
1724 address libbase = NULL;
1725 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1726
1727 // iterate through all loadable segments
1728 for (i = 0; i < info->dlpi_phnum; i++) {
1729 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1730 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1731 // base address of a library is the lowest address of its loaded
1732 // segments.
1733 if (libbase == NULL || libbase > segbase) {
1734 libbase = segbase;
1735 }
1736 // see if 'addr' is within current segment
1737 if (segbase <= d->addr &&
1738 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1739 found = true;
1740 }
1741 }
1742 }
1743
1744 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1745 // so dll_address_to_library_name() can fall through to use dladdr() which
1746 // can figure out executable name from argv[0].
1747 if (found && info->dlpi_name && info->dlpi_name[0]) {
1748 d->base = libbase;
1749 if (d->fname) {
1750 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1751 }
1752 return 1;
1753 }
1754 return 0;
1755 }
1756
1757 bool os::dll_address_to_library_name(address addr, char* buf,
1758 int buflen, int* offset) {
1759 // buf is not optional, but offset is optional
1760 assert(buf != NULL, "sanity check");
1761
1762 Dl_info dlinfo;
1763 struct _address_to_library_name data;
1764
1765 // There is a bug in old glibc dladdr() implementation that it could resolve
1766 // to wrong library name if the .so file has a base address != NULL. Here
1767 // we iterate through the program headers of all loaded libraries to find
1768 // out which library 'addr' really belongs to. This workaround can be
1769 // removed once the minimum requirement for glibc is moved to 2.3.x.
1770 data.addr = addr;
1771 data.fname = buf;
1772 data.buflen = buflen;
1773 data.base = NULL;
1774 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1775
1776 if (rslt) {
1777 // buf already contains library name
1778 if (offset) *offset = addr - data.base;
1779 return true;
1780 }
1781 if (dladdr((void*)addr, &dlinfo) != 0) {
1782 if (dlinfo.dli_fname != NULL) {
1783 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1784 }
1785 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1786 *offset = addr - (address)dlinfo.dli_fbase;
1787 }
1788 return true;
1789 }
1790
1791 buf[0] = '\0';
1792 if (offset) *offset = -1;
1793 return false;
1794 }
1795
1796 // Loads .dll/.so and
1797 // in case of error it checks if .dll/.so was built for the
1798 // same architecture as Hotspot is running on
1799
1800
1801 // Remember the stack's state. The Linux dynamic linker will change
1802 // the stack to 'executable' at most once, so we must safepoint only once.
1803 bool os::Linux::_stack_is_executable = false;
1804
1805 // VM operation that loads a library. This is necessary if stack protection
1806 // of the Java stacks can be lost during loading the library. If we
1807 // do not stop the Java threads, they can stack overflow before the stacks
1808 // are protected again.
1809 class VM_LinuxDllLoad: public VM_Operation {
1810 private:
1811 const char *_filename;
1812 char *_ebuf;
1813 int _ebuflen;
1814 void *_lib;
1815 public:
1816 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1817 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1818 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1819 void doit() {
1820 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1821 os::Linux::_stack_is_executable = true;
1822 }
1823 void* loaded_library() { return _lib; }
1824 };
1825
1826 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1827 void * result = NULL;
1828 bool load_attempted = false;
1829
1830 // Check whether the library to load might change execution rights
1831 // of the stack. If they are changed, the protection of the stack
1832 // guard pages will be lost. We need a safepoint to fix this.
1833 //
1834 // See Linux man page execstack(8) for more info.
1835 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1836 ElfFile ef(filename);
1837 if (!ef.specifies_noexecstack()) {
1838 if (!is_init_completed()) {
1839 os::Linux::_stack_is_executable = true;
1840 // This is OK - No Java threads have been created yet, and hence no
1841 // stack guard pages to fix.
1842 //
1843 // This should happen only when you are building JDK7 using a very
1844 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1845 //
1846 // Dynamic loader will make all stacks executable after
1847 // this function returns, and will not do that again.
1848 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1849 } else {
1850 warning("You have loaded library %s which might have disabled stack guard. "
1851 "The VM will try to fix the stack guard now.\n"
1852 "It's highly recommended that you fix the library with "
1853 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1854 filename);
1855
1856 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1857 JavaThread *jt = JavaThread::current();
1858 if (jt->thread_state() != _thread_in_native) {
1859 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1860 // that requires ExecStack. Cannot enter safe point. Let's give up.
1861 warning("Unable to fix stack guard. Giving up.");
1862 } else {
1863 if (!LoadExecStackDllInVMThread) {
1864 // This is for the case where the DLL has an static
1865 // constructor function that executes JNI code. We cannot
1866 // load such DLLs in the VMThread.
1867 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1868 }
1869
1870 ThreadInVMfromNative tiv(jt);
1871 debug_only(VMNativeEntryWrapper vew;)
1872
1873 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1874 VMThread::execute(&op);
1875 if (LoadExecStackDllInVMThread) {
1876 result = op.loaded_library();
1877 }
1878 load_attempted = true;
1879 }
1880 }
1881 }
1882 }
1883
1884 if (!load_attempted) {
1885 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1886 }
1887
1888 if (result != NULL) {
1889 // Successful loading
1890 return result;
1891 }
1892
1893 Elf32_Ehdr elf_head;
1894 int diag_msg_max_length=ebuflen-strlen(ebuf);
1895 char* diag_msg_buf=ebuf+strlen(ebuf);
1896
1897 if (diag_msg_max_length==0) {
1898 // No more space in ebuf for additional diagnostics message
1899 return NULL;
1900 }
1901
1902
1903 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1904
1905 if (file_descriptor < 0) {
1906 // Can't open library, report dlerror() message
1907 return NULL;
1908 }
1909
1910 bool failed_to_read_elf_head=
1911 (sizeof(elf_head)!=
1912 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1913
1914 ::close(file_descriptor);
1915 if (failed_to_read_elf_head) {
1916 // file i/o error - report dlerror() msg
1917 return NULL;
1918 }
1919
1920 typedef struct {
1921 Elf32_Half code; // Actual value as defined in elf.h
1922 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1923 char elf_class; // 32 or 64 bit
1924 char endianess; // MSB or LSB
1925 char* name; // String representation
1926 } arch_t;
1927
1928 #ifndef EM_486
1929 #define EM_486 6 /* Intel 80486 */
1930 #endif
1931 #ifndef EM_AARCH64
1932 #define EM_AARCH64 183 /* ARM AARCH64 */
1933 #endif
1934
1935 static const arch_t arch_array[]={
1936 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1937 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1938 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1939 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1940 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1941 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1942 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1943 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1944 #if defined(VM_LITTLE_ENDIAN)
1945 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1946 #else
1947 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1948 #endif
1949 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1950 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1951 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1952 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1953 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1954 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1955 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1956 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1957 };
1958
1959 #if (defined IA32)
1960 static Elf32_Half running_arch_code=EM_386;
1961 #elif (defined AMD64)
1962 static Elf32_Half running_arch_code=EM_X86_64;
1963 #elif (defined IA64)
1964 static Elf32_Half running_arch_code=EM_IA_64;
1965 #elif (defined __sparc) && (defined _LP64)
1966 static Elf32_Half running_arch_code=EM_SPARCV9;
1967 #elif (defined __sparc) && (!defined _LP64)
1968 static Elf32_Half running_arch_code=EM_SPARC;
1969 #elif (defined __powerpc64__)
1970 static Elf32_Half running_arch_code=EM_PPC64;
1971 #elif (defined __powerpc__)
1972 static Elf32_Half running_arch_code=EM_PPC;
1973 #elif (defined ARM)
1974 static Elf32_Half running_arch_code=EM_ARM;
1975 #elif (defined S390)
1976 static Elf32_Half running_arch_code=EM_S390;
1977 #elif (defined ALPHA)
1978 static Elf32_Half running_arch_code=EM_ALPHA;
1979 #elif (defined MIPSEL)
1980 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1981 #elif (defined PARISC)
1982 static Elf32_Half running_arch_code=EM_PARISC;
1983 #elif (defined MIPS)
1984 static Elf32_Half running_arch_code=EM_MIPS;
1985 #elif (defined M68K)
1986 static Elf32_Half running_arch_code=EM_68K;
1987 #elif (defined AARCH64)
1988 static Elf32_Half running_arch_code=EM_AARCH64;
1989 #else
1990 #error Method os::dll_load requires that one of following is defined:\
1991 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1992 #endif
1993
1994 // Identify compatability class for VM's architecture and library's architecture
1995 // Obtain string descriptions for architectures
1996
1997 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1998 int running_arch_index=-1;
1999
2000 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
2001 if (running_arch_code == arch_array[i].code) {
2002 running_arch_index = i;
2003 }
2004 if (lib_arch.code == arch_array[i].code) {
2005 lib_arch.compat_class = arch_array[i].compat_class;
2006 lib_arch.name = arch_array[i].name;
2007 }
2008 }
2009
2010 assert(running_arch_index != -1,
2011 "Didn't find running architecture code (running_arch_code) in arch_array");
2012 if (running_arch_index == -1) {
2013 // Even though running architecture detection failed
2014 // we may still continue with reporting dlerror() message
2015 return NULL;
2016 }
2017
2018 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2019 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2020 return NULL;
2021 }
2022
2023 #ifndef S390
2024 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2025 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2026 return NULL;
2027 }
2028 #endif // !S390
2029
2030 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2031 if (lib_arch.name!=NULL) {
2032 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2033 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2034 lib_arch.name, arch_array[running_arch_index].name);
2035 } else {
2036 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2037 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2038 lib_arch.code,
2039 arch_array[running_arch_index].name);
2040 }
2041 }
2042
2043 return NULL;
2044 }
2045
2046 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
2047 int ebuflen) {
2048 void * result = ::dlopen(filename, RTLD_LAZY);
2049 if (result == NULL) {
2050 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2051 ebuf[ebuflen-1] = '\0';
2052 }
2053 return result;
2054 }
2055
2056 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2057 int ebuflen) {
2058 void * result = NULL;
2059 if (LoadExecStackDllInVMThread) {
2060 result = dlopen_helper(filename, ebuf, ebuflen);
2061 }
2062
2063 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2064 // library that requires an executable stack, or which does not have this
2065 // stack attribute set, dlopen changes the stack attribute to executable. The
2066 // read protection of the guard pages gets lost.
2067 //
2068 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2069 // may have been queued at the same time.
2070
2071 if (!_stack_is_executable) {
2072 JavaThread *jt = Threads::first();
2073
2074 while (jt) {
2075 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2076 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2077 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2078 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2079 warning("Attempt to reguard stack yellow zone failed.");
2080 }
2081 }
2082 jt = jt->next();
2083 }
2084 }
2085
2086 return result;
2087 }
2088
2089 // glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2090 // chances are you might want to run the generated bits against glibc-2.0
2091 // libdl.so, so always use locking for any version of glibc.
2092 //
2093 void* os::dll_lookup(void* handle, const char* name) {
2094 pthread_mutex_lock(&dl_mutex);
2095 void* res = dlsym(handle, name);
2096 pthread_mutex_unlock(&dl_mutex);
2097 return res;
2098 }
2099
2100 void* os::get_default_process_handle() {
2101 return (void*)::dlopen(NULL, RTLD_LAZY);
2102 }
2103
2104 static bool _print_ascii_file(const char* filename, outputStream* st) {
2105 int fd = ::open(filename, O_RDONLY);
2106 if (fd == -1) {
2107 return false;
2108 }
2109
2110 char buf[32];
2111 int bytes;
2112 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2113 st->print_raw(buf, bytes);
2114 }
2115
2116 ::close(fd);
2117
2118 return true;
2119 }
2120
2121 void os::print_dll_info(outputStream *st) {
2122 st->print_cr("Dynamic libraries:");
2123
2124 char fname[32];
2125 pid_t pid = os::Linux::gettid();
2126
2127 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2128
2129 if (!_print_ascii_file(fname, st)) {
2130 st->print("Can not get library information for pid = %d\n", pid);
2131 }
2132 }
2133
2134 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2135 FILE *procmapsFile = NULL;
2136
2137 // Open the procfs maps file for the current process
2138 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2139 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2140 char line[PATH_MAX + 100];
2141
2142 // Read line by line from 'file'
2143 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2144 u8 base, top, offset, inode;
2145 char permissions[5];
2146 char device[6];
2147 char name[PATH_MAX + 1];
2148
2149 // Parse fields from line
2150 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2151
2152 // Filter by device id '00:00' so that we only get file system mapped files.
2153 if (strcmp(device, "00:00") != 0) {
2154
2155 // Call callback with the fields of interest
2156 if(callback(name, (address)base, (address)top, param)) {
2157 // Oops abort, callback aborted
2158 fclose(procmapsFile);
2159 return 1;
2160 }
2161 }
2162 }
2163 fclose(procmapsFile);
2164 }
2165 return 0;
2166 }
2167
2168 void os::print_os_info_brief(outputStream* st) {
2169 os::Linux::print_distro_info(st);
2170
2171 os::Posix::print_uname_info(st);
2172
2173 os::Linux::print_libversion_info(st);
2174
2175 }
2176
2177 void os::print_os_info(outputStream* st) {
2178 st->print("OS:");
2179
2180 os::Linux::print_distro_info(st);
2181
2182 os::Posix::print_uname_info(st);
2183
2184 // Print warning if unsafe chroot environment detected
2185 if (unsafe_chroot_detected) {
2186 st->print("WARNING!! ");
2187 st->print_cr("%s", unstable_chroot_error);
2188 }
2189
2190 os::Linux::print_libversion_info(st);
2191
2192 os::Posix::print_rlimit_info(st);
2193
2194 os::Posix::print_load_average(st);
2195
2196 os::Linux::print_full_memory_info(st);
2197 }
2198
2199 // Try to identify popular distros.
2200 // Most Linux distributions have a /etc/XXX-release file, which contains
2201 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2202 // file that also contains the OS version string. Some have more than one
2203 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2204 // /etc/redhat-release.), so the order is important.
2205 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2206 // their own specific XXX-release file as well as a redhat-release file.
2207 // Because of this the XXX-release file needs to be searched for before the
2208 // redhat-release file.
2209 // Since Red Hat has a lsb-release file that is not very descriptive the
2210 // search for redhat-release needs to be before lsb-release.
2211 // Since the lsb-release file is the new standard it needs to be searched
2212 // before the older style release files.
2213 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2214 // next to last resort. The os-release file is a new standard that contains
2215 // distribution information and the system-release file seems to be an old
2216 // standard that has been replaced by the lsb-release and os-release files.
2217 // Searching for the debian_version file is the last resort. It contains
2218 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2219 // "Debian " is printed before the contents of the debian_version file.
2220 void os::Linux::print_distro_info(outputStream* st) {
2221 if (!_print_ascii_file("/etc/oracle-release", st) &&
2222 !_print_ascii_file("/etc/mandriva-release", st) &&
2223 !_print_ascii_file("/etc/mandrake-release", st) &&
2224 !_print_ascii_file("/etc/sun-release", st) &&
2225 !_print_ascii_file("/etc/redhat-release", st) &&
2226 !_print_ascii_file("/etc/lsb-release", st) &&
2227 !_print_ascii_file("/etc/SuSE-release", st) &&
2228 !_print_ascii_file("/etc/turbolinux-release", st) &&
2229 !_print_ascii_file("/etc/gentoo-release", st) &&
2230 !_print_ascii_file("/etc/ltib-release", st) &&
2231 !_print_ascii_file("/etc/angstrom-version", st) &&
2232 !_print_ascii_file("/etc/system-release", st) &&
2233 !_print_ascii_file("/etc/os-release", st)) {
2234
2235 if (file_exists("/etc/debian_version")) {
2236 st->print("Debian ");
2237 _print_ascii_file("/etc/debian_version", st);
2238 } else {
2239 st->print("Linux");
2240 }
2241 }
2242 st->cr();
2243 }
2244
2245 void os::Linux::print_libversion_info(outputStream* st) {
2246 // libc, pthread
2247 st->print("libc:");
2248 st->print("%s ", os::Linux::glibc_version());
2249 st->print("%s ", os::Linux::libpthread_version());
2250 if (os::Linux::is_LinuxThreads()) {
2251 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2252 }
2253 st->cr();
2254 }
2255
2256 void os::Linux::print_full_memory_info(outputStream* st) {
2257 st->print("\n/proc/meminfo:\n");
2258 _print_ascii_file("/proc/meminfo", st);
2259 st->cr();
2260 }
2261
2262 void os::print_memory_info(outputStream* st) {
2263
2264 st->print("Memory:");
2265 st->print(" %dk page", os::vm_page_size()>>10);
2266
2267 // values in struct sysinfo are "unsigned long"
2268 struct sysinfo si;
2269 sysinfo(&si);
2270
2271 st->print(", physical " UINT64_FORMAT "k",
2272 os::physical_memory() >> 10);
2273 st->print("(" UINT64_FORMAT "k free)",
2274 os::available_memory() >> 10);
2275 st->print(", swap " UINT64_FORMAT "k",
2276 ((jlong)si.totalswap * si.mem_unit) >> 10);
2277 st->print("(" UINT64_FORMAT "k free)",
2278 ((jlong)si.freeswap * si.mem_unit) >> 10);
2279 st->cr();
2280 }
2281
2282 void os::pd_print_cpu_info(outputStream* st) {
2283 st->print("\n/proc/cpuinfo:\n");
2284 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2285 st->print(" <Not Available>");
2286 }
2287 st->cr();
2288 }
2289
2290 void os::print_siginfo(outputStream* st, void* siginfo) {
2291 const siginfo_t* si = (const siginfo_t*)siginfo;
2292
2293 os::Posix::print_siginfo_brief(st, si);
2294 #if INCLUDE_CDS
2295 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2296 UseSharedSpaces) {
2297 FileMapInfo* mapinfo = FileMapInfo::current_info();
2298 if (mapinfo->is_in_shared_space(si->si_addr)) {
2299 st->print("\n\nError accessing class data sharing archive." \
2300 " Mapped file inaccessible during execution, " \
2301 " possible disk/network problem.");
2302 }
2303 }
2304 #endif
2305 st->cr();
2306 }
2307
2308
2309 static void print_signal_handler(outputStream* st, int sig,
2310 char* buf, size_t buflen);
2311
2312 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2313 st->print_cr("Signal Handlers:");
2314 print_signal_handler(st, SIGSEGV, buf, buflen);
2315 print_signal_handler(st, SIGBUS , buf, buflen);
2316 print_signal_handler(st, SIGFPE , buf, buflen);
2317 print_signal_handler(st, SIGPIPE, buf, buflen);
2318 print_signal_handler(st, SIGXFSZ, buf, buflen);
2319 print_signal_handler(st, SIGILL , buf, buflen);
2320 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2321 print_signal_handler(st, SR_signum, buf, buflen);
2322 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2323 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2324 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2325 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2326 #if defined(PPC64)
2327 print_signal_handler(st, SIGTRAP, buf, buflen);
2328 #endif
2329 }
2330
2331 static char saved_jvm_path[MAXPATHLEN] = {0};
2332
2333 // Find the full path to the current module, libjvm.so
2334 void os::jvm_path(char *buf, jint buflen) {
2335 // Error checking.
2336 if (buflen < MAXPATHLEN) {
2337 assert(false, "must use a large-enough buffer");
2338 buf[0] = '\0';
2339 return;
2340 }
2341 // Lazy resolve the path to current module.
2342 if (saved_jvm_path[0] != 0) {
2343 strcpy(buf, saved_jvm_path);
2344 return;
2345 }
2346
2347 char dli_fname[MAXPATHLEN];
2348 bool ret = dll_address_to_library_name(
2349 CAST_FROM_FN_PTR(address, os::jvm_path),
2350 dli_fname, sizeof(dli_fname), NULL);
2351 assert(ret, "cannot locate libjvm");
2352 char *rp = NULL;
2353 if (ret && dli_fname[0] != '\0') {
2354 rp = realpath(dli_fname, buf);
2355 }
2356 if (rp == NULL) {
2357 return;
2358 }
2359
2360 if (Arguments::sun_java_launcher_is_altjvm()) {
2361 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2362 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2363 // If "/jre/lib/" appears at the right place in the string, then
2364 // assume we are installed in a JDK and we're done. Otherwise, check
2365 // for a JAVA_HOME environment variable and fix up the path so it
2366 // looks like libjvm.so is installed there (append a fake suffix
2367 // hotspot/libjvm.so).
2368 const char *p = buf + strlen(buf) - 1;
2369 for (int count = 0; p > buf && count < 5; ++count) {
2370 for (--p; p > buf && *p != '/'; --p)
2371 /* empty */ ;
2372 }
2373
2374 if (strncmp(p, "/jre/lib/", 9) != 0) {
2375 // Look for JAVA_HOME in the environment.
2376 char* java_home_var = ::getenv("JAVA_HOME");
2377 if (java_home_var != NULL && java_home_var[0] != 0) {
2378 char* jrelib_p;
2379 int len;
2380
2381 // Check the current module name "libjvm.so".
2382 p = strrchr(buf, '/');
2383 assert(strstr(p, "/libjvm") == p, "invalid library name");
2384
2385 rp = realpath(java_home_var, buf);
2386 if (rp == NULL) {
2387 return;
2388 }
2389
2390 // determine if this is a legacy image or modules image
2391 // modules image doesn't have "jre" subdirectory
2392 len = strlen(buf);
2393 assert(len < buflen, "Ran out of buffer room");
2394 jrelib_p = buf + len;
2395 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2396 if (0 != access(buf, F_OK)) {
2397 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2398 }
2399
2400 if (0 == access(buf, F_OK)) {
2401 // Use current module name "libjvm.so"
2402 len = strlen(buf);
2403 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2404 } else {
2405 // Go back to path of .so
2406 rp = realpath(dli_fname, buf);
2407 if (rp == NULL) {
2408 return;
2409 }
2410 }
2411 }
2412 }
2413 }
2414
2415 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2416 }
2417
2418 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2419 // no prefix required, not even "_"
2420 }
2421
2422 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2423 // no suffix required
2424 }
2425
2426 ////////////////////////////////////////////////////////////////////////////////
2427 // sun.misc.Signal support
2428
2429 static volatile jint sigint_count = 0;
2430
2431 static void UserHandler(int sig, void *siginfo, void *context) {
2432 // 4511530 - sem_post is serialized and handled by the manager thread. When
2433 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2434 // don't want to flood the manager thread with sem_post requests.
2435 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2436 return;
2437 }
2438
2439 // Ctrl-C is pressed during error reporting, likely because the error
2440 // handler fails to abort. Let VM die immediately.
2441 if (sig == SIGINT && is_error_reported()) {
2442 os::die();
2443 }
2444
2445 os::signal_notify(sig);
2446 }
2447
2448 void* os::user_handler() {
2449 return CAST_FROM_FN_PTR(void*, UserHandler);
2450 }
2451
2452 class Semaphore : public StackObj {
2453 public:
2454 Semaphore();
2455 ~Semaphore();
2456 void signal();
2457 void wait();
2458 bool trywait();
2459 bool timedwait(unsigned int sec, int nsec);
2460 private:
2461 sem_t _semaphore;
2462 };
2463
2464 Semaphore::Semaphore() {
2465 sem_init(&_semaphore, 0, 0);
2466 }
2467
2468 Semaphore::~Semaphore() {
2469 sem_destroy(&_semaphore);
2470 }
2471
2472 void Semaphore::signal() {
2473 sem_post(&_semaphore);
2474 }
2475
2476 void Semaphore::wait() {
2477 sem_wait(&_semaphore);
2478 }
2479
2480 bool Semaphore::trywait() {
2481 return sem_trywait(&_semaphore) == 0;
2482 }
2483
2484 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2485
2486 struct timespec ts;
2487 // Semaphore's are always associated with CLOCK_REALTIME
2488 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2489 // see unpackTime for discussion on overflow checking
2490 if (sec >= MAX_SECS) {
2491 ts.tv_sec += MAX_SECS;
2492 ts.tv_nsec = 0;
2493 } else {
2494 ts.tv_sec += sec;
2495 ts.tv_nsec += nsec;
2496 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2497 ts.tv_nsec -= NANOSECS_PER_SEC;
2498 ++ts.tv_sec; // note: this must be <= max_secs
2499 }
2500 }
2501
2502 while (1) {
2503 int result = sem_timedwait(&_semaphore, &ts);
2504 if (result == 0) {
2505 return true;
2506 } else if (errno == EINTR) {
2507 continue;
2508 } else if (errno == ETIMEDOUT) {
2509 return false;
2510 } else {
2511 return false;
2512 }
2513 }
2514 }
2515
2516 extern "C" {
2517 typedef void (*sa_handler_t)(int);
2518 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2519 }
2520
2521 void* os::signal(int signal_number, void* handler) {
2522 struct sigaction sigAct, oldSigAct;
2523
2524 sigfillset(&(sigAct.sa_mask));
2525 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2526 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2527
2528 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2529 // -1 means registration failed
2530 return (void *)-1;
2531 }
2532
2533 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2534 }
2535
2536 void os::signal_raise(int signal_number) {
2537 ::raise(signal_number);
2538 }
2539
2540 // The following code is moved from os.cpp for making this
2541 // code platform specific, which it is by its very nature.
2542
2543 // Will be modified when max signal is changed to be dynamic
2544 int os::sigexitnum_pd() {
2545 return NSIG;
2546 }
2547
2548 // a counter for each possible signal value
2549 static volatile jint pending_signals[NSIG+1] = { 0 };
2550
2551 // Linux(POSIX) specific hand shaking semaphore.
2552 static sem_t sig_sem;
2553 static Semaphore sr_semaphore;
2554
2555 void os::signal_init_pd() {
2556 // Initialize signal structures
2557 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2558
2559 // Initialize signal semaphore
2560 ::sem_init(&sig_sem, 0, 0);
2561 }
2562
2563 void os::signal_notify(int sig) {
2564 Atomic::inc(&pending_signals[sig]);
2565 ::sem_post(&sig_sem);
2566 }
2567
2568 static int check_pending_signals(bool wait) {
2569 Atomic::store(0, &sigint_count);
2570 for (;;) {
2571 for (int i = 0; i < NSIG + 1; i++) {
2572 jint n = pending_signals[i];
2573 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2574 return i;
2575 }
2576 }
2577 if (!wait) {
2578 return -1;
2579 }
2580 JavaThread *thread = JavaThread::current();
2581 ThreadBlockInVM tbivm(thread);
2582
2583 bool threadIsSuspended;
2584 do {
2585 thread->set_suspend_equivalent();
2586 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2587 ::sem_wait(&sig_sem);
2588
2589 // were we externally suspended while we were waiting?
2590 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2591 if (threadIsSuspended) {
2592 // The semaphore has been incremented, but while we were waiting
2593 // another thread suspended us. We don't want to continue running
2594 // while suspended because that would surprise the thread that
2595 // suspended us.
2596 ::sem_post(&sig_sem);
2597
2598 thread->java_suspend_self();
2599 }
2600 } while (threadIsSuspended);
2601 }
2602 }
2603
2604 int os::signal_lookup() {
2605 return check_pending_signals(false);
2606 }
2607
2608 int os::signal_wait() {
2609 return check_pending_signals(true);
2610 }
2611
2612 ////////////////////////////////////////////////////////////////////////////////
2613 // Virtual Memory
2614
2615 int os::vm_page_size() {
2616 // Seems redundant as all get out
2617 assert(os::Linux::page_size() != -1, "must call os::init");
2618 return os::Linux::page_size();
2619 }
2620
2621 // Solaris allocates memory by pages.
2622 int os::vm_allocation_granularity() {
2623 assert(os::Linux::page_size() != -1, "must call os::init");
2624 return os::Linux::page_size();
2625 }
2626
2627 // Rationale behind this function:
2628 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2629 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2630 // samples for JITted code. Here we create private executable mapping over the code cache
2631 // and then we can use standard (well, almost, as mapping can change) way to provide
2632 // info for the reporting script by storing timestamp and location of symbol
2633 void linux_wrap_code(char* base, size_t size) {
2634 static volatile jint cnt = 0;
2635
2636 if (!UseOprofile) {
2637 return;
2638 }
2639
2640 char buf[PATH_MAX+1];
2641 int num = Atomic::add(1, &cnt);
2642
2643 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2644 os::get_temp_directory(), os::current_process_id(), num);
2645 unlink(buf);
2646
2647 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2648
2649 if (fd != -1) {
2650 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2651 if (rv != (off_t)-1) {
2652 if (::write(fd, "", 1) == 1) {
2653 mmap(base, size,
2654 PROT_READ|PROT_WRITE|PROT_EXEC,
2655 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2656 }
2657 }
2658 ::close(fd);
2659 unlink(buf);
2660 }
2661 }
2662
2663 static bool recoverable_mmap_error(int err) {
2664 // See if the error is one we can let the caller handle. This
2665 // list of errno values comes from JBS-6843484. I can't find a
2666 // Linux man page that documents this specific set of errno
2667 // values so while this list currently matches Solaris, it may
2668 // change as we gain experience with this failure mode.
2669 switch (err) {
2670 case EBADF:
2671 case EINVAL:
2672 case ENOTSUP:
2673 // let the caller deal with these errors
2674 return true;
2675
2676 default:
2677 // Any remaining errors on this OS can cause our reserved mapping
2678 // to be lost. That can cause confusion where different data
2679 // structures think they have the same memory mapped. The worst
2680 // scenario is if both the VM and a library think they have the
2681 // same memory mapped.
2682 return false;
2683 }
2684 }
2685
2686 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2687 int err) {
2688 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2689 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2690 strerror(err), err);
2691 }
2692
2693 static void warn_fail_commit_memory(char* addr, size_t size,
2694 size_t alignment_hint, bool exec,
2695 int err) {
2696 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2697 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2698 alignment_hint, exec, strerror(err), err);
2699 }
2700
2701 // NOTE: Linux kernel does not really reserve the pages for us.
2702 // All it does is to check if there are enough free pages
2703 // left at the time of mmap(). This could be a potential
2704 // problem.
2705 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2706 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2707 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2708 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2709 if (res != (uintptr_t) MAP_FAILED) {
2710 if (UseNUMAInterleaving) {
2711 numa_make_global(addr, size);
2712 }
2713 return 0;
2714 }
2715
2716 int err = errno; // save errno from mmap() call above
2717
2718 if (!recoverable_mmap_error(err)) {
2719 warn_fail_commit_memory(addr, size, exec, err);
2720 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2721 }
2722
2723 return err;
2724 }
2725
2726 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2727 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2728 }
2729
2730 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2731 const char* mesg) {
2732 assert(mesg != NULL, "mesg must be specified");
2733 int err = os::Linux::commit_memory_impl(addr, size, exec);
2734 if (err != 0) {
2735 // the caller wants all commit errors to exit with the specified mesg:
2736 warn_fail_commit_memory(addr, size, exec, err);
2737 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2738 }
2739 }
2740
2741 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2742 #ifndef MAP_HUGETLB
2743 #define MAP_HUGETLB 0x40000
2744 #endif
2745
2746 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2747 #ifndef MADV_HUGEPAGE
2748 #define MADV_HUGEPAGE 14
2749 #endif
2750
2751 int os::Linux::commit_memory_impl(char* addr, size_t size,
2752 size_t alignment_hint, bool exec) {
2753 int err = os::Linux::commit_memory_impl(addr, size, exec);
2754 if (err == 0) {
2755 realign_memory(addr, size, alignment_hint);
2756 }
2757 return err;
2758 }
2759
2760 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2761 bool exec) {
2762 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2763 }
2764
2765 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2766 size_t alignment_hint, bool exec,
2767 const char* mesg) {
2768 assert(mesg != NULL, "mesg must be specified");
2769 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2770 if (err != 0) {
2771 // the caller wants all commit errors to exit with the specified mesg:
2772 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2773 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2774 }
2775 }
2776
2777 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2778 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2779 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2780 // be supported or the memory may already be backed by huge pages.
2781 ::madvise(addr, bytes, MADV_HUGEPAGE);
2782 }
2783 }
2784
2785 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2786 // This method works by doing an mmap over an existing mmaping and effectively discarding
2787 // the existing pages. However it won't work for SHM-based large pages that cannot be
2788 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2789 // small pages on top of the SHM segment. This method always works for small pages, so we
2790 // allow that in any case.
2791 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2792 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2793 }
2794 }
2795
2796 void os::numa_make_global(char *addr, size_t bytes) {
2797 Linux::numa_interleave_memory(addr, bytes);
2798 }
2799
2800 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2801 // bind policy to MPOL_PREFERRED for the current thread.
2802 #define USE_MPOL_PREFERRED 0
2803
2804 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2805 // To make NUMA and large pages more robust when both enabled, we need to ease
2806 // the requirements on where the memory should be allocated. MPOL_BIND is the
2807 // default policy and it will force memory to be allocated on the specified
2808 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2809 // the specified node, but will not force it. Using this policy will prevent
2810 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2811 // free large pages.
2812 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2813 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2814 }
2815
2816 bool os::numa_topology_changed() { return false; }
2817
2818 size_t os::numa_get_groups_num() {
2819 int max_node = Linux::numa_max_node();
2820 return max_node > 0 ? max_node + 1 : 1;
2821 }
2822
2823 int os::numa_get_group_id() {
2824 int cpu_id = Linux::sched_getcpu();
2825 if (cpu_id != -1) {
2826 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2827 if (lgrp_id != -1) {
2828 return lgrp_id;
2829 }
2830 }
2831 return 0;
2832 }
2833
2834 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2835 for (size_t i = 0; i < size; i++) {
2836 ids[i] = i;
2837 }
2838 return size;
2839 }
2840
2841 bool os::get_page_info(char *start, page_info* info) {
2842 return false;
2843 }
2844
2845 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2846 page_info* page_found) {
2847 return end;
2848 }
2849
2850
2851 int os::Linux::sched_getcpu_syscall(void) {
2852 unsigned int cpu;
2853 int retval = -1;
2854
2855 #if defined(IA32)
2856 #ifndef SYS_getcpu
2857 #define SYS_getcpu 318
2858 #endif
2859 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2860 #elif defined(AMD64)
2861 // Unfortunately we have to bring all these macros here from vsyscall.h
2862 // to be able to compile on old linuxes.
2863 #define __NR_vgetcpu 2
2864 #define VSYSCALL_START (-10UL << 20)
2865 #define VSYSCALL_SIZE 1024
2866 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2867 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2868 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2869 retval = vgetcpu(&cpu, NULL, NULL);
2870 #endif
2871
2872 return (retval == -1) ? retval : cpu;
2873 }
2874
2875 // Something to do with the numa-aware allocator needs these symbols
2876 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2877 extern "C" JNIEXPORT void numa_error(char *where) { }
2878 extern "C" JNIEXPORT int fork1() { return fork(); }
2879
2880
2881 // If we are running with libnuma version > 2, then we should
2882 // be trying to use symbols with versions 1.1
2883 // If we are running with earlier version, which did not have symbol versions,
2884 // we should use the base version.
2885 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2886 void *f = dlvsym(handle, name, "libnuma_1.1");
2887 if (f == NULL) {
2888 f = dlsym(handle, name);
2889 }
2890 return f;
2891 }
2892
2893 bool os::Linux::libnuma_init() {
2894 // sched_getcpu() should be in libc.
2895 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2896 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2897
2898 // If it's not, try a direct syscall.
2899 if (sched_getcpu() == -1) {
2900 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2901 (void*)&sched_getcpu_syscall));
2902 }
2903
2904 if (sched_getcpu() != -1) { // Does it work?
2905 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2906 if (handle != NULL) {
2907 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2908 libnuma_dlsym(handle, "numa_node_to_cpus")));
2909 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2910 libnuma_dlsym(handle, "numa_max_node")));
2911 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2912 libnuma_dlsym(handle, "numa_available")));
2913 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2914 libnuma_dlsym(handle, "numa_tonode_memory")));
2915 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2916 libnuma_dlsym(handle, "numa_interleave_memory")));
2917 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2918 libnuma_dlsym(handle, "numa_set_bind_policy")));
2919
2920
2921 if (numa_available() != -1) {
2922 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2923 // Create a cpu -> node mapping
2924 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2925 rebuild_cpu_to_node_map();
2926 return true;
2927 }
2928 }
2929 }
2930 return false;
2931 }
2932
2933 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2934 // The table is later used in get_node_by_cpu().
2935 void os::Linux::rebuild_cpu_to_node_map() {
2936 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2937 // in libnuma (possible values are starting from 16,
2938 // and continuing up with every other power of 2, but less
2939 // than the maximum number of CPUs supported by kernel), and
2940 // is a subject to change (in libnuma version 2 the requirements
2941 // are more reasonable) we'll just hardcode the number they use
2942 // in the library.
2943 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2944
2945 size_t cpu_num = os::active_processor_count();
2946 size_t cpu_map_size = NCPUS / BitsPerCLong;
2947 size_t cpu_map_valid_size =
2948 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2949
2950 cpu_to_node()->clear();
2951 cpu_to_node()->at_grow(cpu_num - 1);
2952 size_t node_num = numa_get_groups_num();
2953
2954 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2955 for (size_t i = 0; i < node_num; i++) {
2956 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2957 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2958 if (cpu_map[j] != 0) {
2959 for (size_t k = 0; k < BitsPerCLong; k++) {
2960 if (cpu_map[j] & (1UL << k)) {
2961 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2962 }
2963 }
2964 }
2965 }
2966 }
2967 }
2968 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2969 }
2970
2971 int os::Linux::get_node_by_cpu(int cpu_id) {
2972 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2973 return cpu_to_node()->at(cpu_id);
2974 }
2975 return -1;
2976 }
2977
2978 GrowableArray<int>* os::Linux::_cpu_to_node;
2979 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2980 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2981 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2982 os::Linux::numa_available_func_t os::Linux::_numa_available;
2983 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2984 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2985 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2986 unsigned long* os::Linux::_numa_all_nodes;
2987
2988 bool os::pd_uncommit_memory(char* addr, size_t size) {
2989 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2990 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2991 return res != (uintptr_t) MAP_FAILED;
2992 }
2993
2994 static address get_stack_commited_bottom(address bottom, size_t size) {
2995 address nbot = bottom;
2996 address ntop = bottom + size;
2997
2998 size_t page_sz = os::vm_page_size();
2999 unsigned pages = size / page_sz;
3000
3001 unsigned char vec[1];
3002 unsigned imin = 1, imax = pages + 1, imid;
3003 int mincore_return_value = 0;
3004
3005 assert(imin <= imax, "Unexpected page size");
3006
3007 while (imin < imax) {
3008 imid = (imax + imin) / 2;
3009 nbot = ntop - (imid * page_sz);
3010
3011 // Use a trick with mincore to check whether the page is mapped or not.
3012 // mincore sets vec to 1 if page resides in memory and to 0 if page
3013 // is swapped output but if page we are asking for is unmapped
3014 // it returns -1,ENOMEM
3015 mincore_return_value = mincore(nbot, page_sz, vec);
3016
3017 if (mincore_return_value == -1) {
3018 // Page is not mapped go up
3019 // to find first mapped page
3020 if (errno != EAGAIN) {
3021 assert(errno == ENOMEM, "Unexpected mincore errno");
3022 imax = imid;
3023 }
3024 } else {
3025 // Page is mapped go down
3026 // to find first not mapped page
3027 imin = imid + 1;
3028 }
3029 }
3030
3031 nbot = nbot + page_sz;
3032
3033 // Adjust stack bottom one page up if last checked page is not mapped
3034 if (mincore_return_value == -1) {
3035 nbot = nbot + page_sz;
3036 }
3037
3038 return nbot;
3039 }
3040
3041
3042 // Linux uses a growable mapping for the stack, and if the mapping for
3043 // the stack guard pages is not removed when we detach a thread the
3044 // stack cannot grow beyond the pages where the stack guard was
3045 // mapped. If at some point later in the process the stack expands to
3046 // that point, the Linux kernel cannot expand the stack any further
3047 // because the guard pages are in the way, and a segfault occurs.
3048 //
3049 // However, it's essential not to split the stack region by unmapping
3050 // a region (leaving a hole) that's already part of the stack mapping,
3051 // so if the stack mapping has already grown beyond the guard pages at
3052 // the time we create them, we have to truncate the stack mapping.
3053 // So, we need to know the extent of the stack mapping when
3054 // create_stack_guard_pages() is called.
3055
3056 // We only need this for stacks that are growable: at the time of
3057 // writing thread stacks don't use growable mappings (i.e. those
3058 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3059 // only applies to the main thread.
3060
3061 // If the (growable) stack mapping already extends beyond the point
3062 // where we're going to put our guard pages, truncate the mapping at
3063 // that point by munmap()ping it. This ensures that when we later
3064 // munmap() the guard pages we don't leave a hole in the stack
3065 // mapping. This only affects the main/initial thread
3066
3067 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3068 if (os::Linux::is_initial_thread()) {
3069 // As we manually grow stack up to bottom inside create_attached_thread(),
3070 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3071 // we don't need to do anything special.
3072 // Check it first, before calling heavy function.
3073 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3074 unsigned char vec[1];
3075
3076 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3077 // Fallback to slow path on all errors, including EAGAIN
3078 stack_extent = (uintptr_t) get_stack_commited_bottom(
3079 os::Linux::initial_thread_stack_bottom(),
3080 (size_t)addr - stack_extent);
3081 }
3082
3083 if (stack_extent < (uintptr_t)addr) {
3084 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3085 }
3086 }
3087
3088 return os::commit_memory(addr, size, !ExecMem);
3089 }
3090
3091 // If this is a growable mapping, remove the guard pages entirely by
3092 // munmap()ping them. If not, just call uncommit_memory(). This only
3093 // affects the main/initial thread, but guard against future OS changes
3094 // It's safe to always unmap guard pages for initial thread because we
3095 // always place it right after end of the mapped region
3096
3097 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3098 uintptr_t stack_extent, stack_base;
3099
3100 if (os::Linux::is_initial_thread()) {
3101 return ::munmap(addr, size) == 0;
3102 }
3103
3104 return os::uncommit_memory(addr, size);
3105 }
3106
3107 static address _highest_vm_reserved_address = NULL;
3108
3109 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3110 // at 'requested_addr'. If there are existing memory mappings at the same
3111 // location, however, they will be overwritten. If 'fixed' is false,
3112 // 'requested_addr' is only treated as a hint, the return value may or
3113 // may not start from the requested address. Unlike Linux mmap(), this
3114 // function returns NULL to indicate failure.
3115 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3116 char * addr;
3117 int flags;
3118
3119 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3120 if (fixed) {
3121 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3122 flags |= MAP_FIXED;
3123 }
3124
3125 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3126 // touch an uncommitted page. Otherwise, the read/write might
3127 // succeed if we have enough swap space to back the physical page.
3128 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3129 flags, -1, 0);
3130
3131 if (addr != MAP_FAILED) {
3132 // anon_mmap() should only get called during VM initialization,
3133 // don't need lock (actually we can skip locking even it can be called
3134 // from multiple threads, because _highest_vm_reserved_address is just a
3135 // hint about the upper limit of non-stack memory regions.)
3136 if ((address)addr + bytes > _highest_vm_reserved_address) {
3137 _highest_vm_reserved_address = (address)addr + bytes;
3138 }
3139 }
3140
3141 return addr == MAP_FAILED ? NULL : addr;
3142 }
3143
3144 // Don't update _highest_vm_reserved_address, because there might be memory
3145 // regions above addr + size. If so, releasing a memory region only creates
3146 // a hole in the address space, it doesn't help prevent heap-stack collision.
3147 //
3148 static int anon_munmap(char * addr, size_t size) {
3149 return ::munmap(addr, size) == 0;
3150 }
3151
3152 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3153 size_t alignment_hint) {
3154 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3155 }
3156
3157 bool os::pd_release_memory(char* addr, size_t size) {
3158 return anon_munmap(addr, size);
3159 }
3160
3161 static address highest_vm_reserved_address() {
3162 return _highest_vm_reserved_address;
3163 }
3164
3165 static bool linux_mprotect(char* addr, size_t size, int prot) {
3166 // Linux wants the mprotect address argument to be page aligned.
3167 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3168
3169 // According to SUSv3, mprotect() should only be used with mappings
3170 // established by mmap(), and mmap() always maps whole pages. Unaligned
3171 // 'addr' likely indicates problem in the VM (e.g. trying to change
3172 // protection of malloc'ed or statically allocated memory). Check the
3173 // caller if you hit this assert.
3174 assert(addr == bottom, "sanity check");
3175
3176 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3177 return ::mprotect(bottom, size, prot) == 0;
3178 }
3179
3180 // Set protections specified
3181 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3182 bool is_committed) {
3183 unsigned int p = 0;
3184 switch (prot) {
3185 case MEM_PROT_NONE: p = PROT_NONE; break;
3186 case MEM_PROT_READ: p = PROT_READ; break;
3187 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3188 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3189 default:
3190 ShouldNotReachHere();
3191 }
3192 // is_committed is unused.
3193 return linux_mprotect(addr, bytes, p);
3194 }
3195
3196 bool os::guard_memory(char* addr, size_t size) {
3197 return linux_mprotect(addr, size, PROT_NONE);
3198 }
3199
3200 bool os::unguard_memory(char* addr, size_t size) {
3201 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3202 }
3203
3204 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3205 size_t page_size) {
3206 bool result = false;
3207 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3208 MAP_ANONYMOUS|MAP_PRIVATE,
3209 -1, 0);
3210 if (p != MAP_FAILED) {
3211 void *aligned_p = align_ptr_up(p, page_size);
3212
3213 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3214
3215 munmap(p, page_size * 2);
3216 }
3217
3218 if (warn && !result) {
3219 warning("TransparentHugePages is not supported by the operating system.");
3220 }
3221
3222 return result;
3223 }
3224
3225 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3226 bool result = false;
3227 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3228 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3229 -1, 0);
3230
3231 if (p != MAP_FAILED) {
3232 // We don't know if this really is a huge page or not.
3233 FILE *fp = fopen("/proc/self/maps", "r");
3234 if (fp) {
3235 while (!feof(fp)) {
3236 char chars[257];
3237 long x = 0;
3238 if (fgets(chars, sizeof(chars), fp)) {
3239 if (sscanf(chars, "%lx-%*x", &x) == 1
3240 && x == (long)p) {
3241 if (strstr (chars, "hugepage")) {
3242 result = true;
3243 break;
3244 }
3245 }
3246 }
3247 }
3248 fclose(fp);
3249 }
3250 munmap(p, page_size);
3251 }
3252
3253 if (warn && !result) {
3254 warning("HugeTLBFS is not supported by the operating system.");
3255 }
3256
3257 return result;
3258 }
3259
3260 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3261 //
3262 // From the coredump_filter documentation:
3263 //
3264 // - (bit 0) anonymous private memory
3265 // - (bit 1) anonymous shared memory
3266 // - (bit 2) file-backed private memory
3267 // - (bit 3) file-backed shared memory
3268 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3269 // effective only if the bit 2 is cleared)
3270 // - (bit 5) hugetlb private memory
3271 // - (bit 6) hugetlb shared memory
3272 //
3273 static void set_coredump_filter(void) {
3274 FILE *f;
3275 long cdm;
3276
3277 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3278 return;
3279 }
3280
3281 if (fscanf(f, "%lx", &cdm) != 1) {
3282 fclose(f);
3283 return;
3284 }
3285
3286 rewind(f);
3287
3288 if ((cdm & LARGEPAGES_BIT) == 0) {
3289 cdm |= LARGEPAGES_BIT;
3290 fprintf(f, "%#lx", cdm);
3291 }
3292
3293 fclose(f);
3294 }
3295
3296 // Large page support
3297
3298 static size_t _large_page_size = 0;
3299
3300 size_t os::Linux::find_large_page_size() {
3301 size_t large_page_size = 0;
3302
3303 // large_page_size on Linux is used to round up heap size. x86 uses either
3304 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3305 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3306 // page as large as 256M.
3307 //
3308 // Here we try to figure out page size by parsing /proc/meminfo and looking
3309 // for a line with the following format:
3310 // Hugepagesize: 2048 kB
3311 //
3312 // If we can't determine the value (e.g. /proc is not mounted, or the text
3313 // format has been changed), we'll use the largest page size supported by
3314 // the processor.
3315
3316 #ifndef ZERO
3317 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3318 ARM_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3319 #endif // ZERO
3320
3321 FILE *fp = fopen("/proc/meminfo", "r");
3322 if (fp) {
3323 while (!feof(fp)) {
3324 int x = 0;
3325 char buf[16];
3326 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3327 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3328 large_page_size = x * K;
3329 break;
3330 }
3331 } else {
3332 // skip to next line
3333 for (;;) {
3334 int ch = fgetc(fp);
3335 if (ch == EOF || ch == (int)'\n') break;
3336 }
3337 }
3338 }
3339 fclose(fp);
3340 }
3341
3342 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3343 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3344 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3345 proper_unit_for_byte_size(large_page_size));
3346 }
3347
3348 return large_page_size;
3349 }
3350
3351 size_t os::Linux::setup_large_page_size() {
3352 _large_page_size = Linux::find_large_page_size();
3353 const size_t default_page_size = (size_t)Linux::page_size();
3354 if (_large_page_size > default_page_size) {
3355 _page_sizes[0] = _large_page_size;
3356 _page_sizes[1] = default_page_size;
3357 _page_sizes[2] = 0;
3358 }
3359
3360 return _large_page_size;
3361 }
3362
3363 bool os::Linux::setup_large_page_type(size_t page_size) {
3364 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3365 FLAG_IS_DEFAULT(UseSHM) &&
3366 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3367
3368 // The type of large pages has not been specified by the user.
3369
3370 // Try UseHugeTLBFS and then UseSHM.
3371 UseHugeTLBFS = UseSHM = true;
3372
3373 // Don't try UseTransparentHugePages since there are known
3374 // performance issues with it turned on. This might change in the future.
3375 UseTransparentHugePages = false;
3376 }
3377
3378 if (UseTransparentHugePages) {
3379 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3380 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3381 UseHugeTLBFS = false;
3382 UseSHM = false;
3383 return true;
3384 }
3385 UseTransparentHugePages = false;
3386 }
3387
3388 if (UseHugeTLBFS) {
3389 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3390 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3391 UseSHM = false;
3392 return true;
3393 }
3394 UseHugeTLBFS = false;
3395 }
3396
3397 return UseSHM;
3398 }
3399
3400 void os::large_page_init() {
3401 if (!UseLargePages &&
3402 !UseTransparentHugePages &&
3403 !UseHugeTLBFS &&
3404 !UseSHM) {
3405 // Not using large pages.
3406 return;
3407 }
3408
3409 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3410 // The user explicitly turned off large pages.
3411 // Ignore the rest of the large pages flags.
3412 UseTransparentHugePages = false;
3413 UseHugeTLBFS = false;
3414 UseSHM = false;
3415 return;
3416 }
3417
3418 size_t large_page_size = Linux::setup_large_page_size();
3419 UseLargePages = Linux::setup_large_page_type(large_page_size);
3420
3421 set_coredump_filter();
3422 }
3423
3424 #ifndef SHM_HUGETLB
3425 #define SHM_HUGETLB 04000
3426 #endif
3427
3428 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3429 char* req_addr, bool exec) {
3430 // "exec" is passed in but not used. Creating the shared image for
3431 // the code cache doesn't have an SHM_X executable permission to check.
3432 assert(UseLargePages && UseSHM, "only for SHM large pages");
3433 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3434
3435 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3436 return NULL; // Fallback to small pages.
3437 }
3438
3439 key_t key = IPC_PRIVATE;
3440 char *addr;
3441
3442 bool warn_on_failure = UseLargePages &&
3443 (!FLAG_IS_DEFAULT(UseLargePages) ||
3444 !FLAG_IS_DEFAULT(UseSHM) ||
3445 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3446 char msg[128];
3447
3448 // Create a large shared memory region to attach to based on size.
3449 // Currently, size is the total size of the heap
3450 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3451 if (shmid == -1) {
3452 // Possible reasons for shmget failure:
3453 // 1. shmmax is too small for Java heap.
3454 // > check shmmax value: cat /proc/sys/kernel/shmmax
3455 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3456 // 2. not enough large page memory.
3457 // > check available large pages: cat /proc/meminfo
3458 // > increase amount of large pages:
3459 // echo new_value > /proc/sys/vm/nr_hugepages
3460 // Note 1: different Linux may use different name for this property,
3461 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3462 // Note 2: it's possible there's enough physical memory available but
3463 // they are so fragmented after a long run that they can't
3464 // coalesce into large pages. Try to reserve large pages when
3465 // the system is still "fresh".
3466 if (warn_on_failure) {
3467 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3468 warning("%s", msg);
3469 }
3470 return NULL;
3471 }
3472
3473 // attach to the region
3474 addr = (char*)shmat(shmid, req_addr, 0);
3475 int err = errno;
3476
3477 // Remove shmid. If shmat() is successful, the actual shared memory segment
3478 // will be deleted when it's detached by shmdt() or when the process
3479 // terminates. If shmat() is not successful this will remove the shared
3480 // segment immediately.
3481 shmctl(shmid, IPC_RMID, NULL);
3482
3483 if ((intptr_t)addr == -1) {
3484 if (warn_on_failure) {
3485 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3486 warning("%s", msg);
3487 }
3488 return NULL;
3489 }
3490
3491 return addr;
3492 }
3493
3494 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3495 int error) {
3496 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3497
3498 bool warn_on_failure = UseLargePages &&
3499 (!FLAG_IS_DEFAULT(UseLargePages) ||
3500 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3501 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3502
3503 if (warn_on_failure) {
3504 char msg[128];
3505 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3506 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3507 warning("%s", msg);
3508 }
3509 }
3510
3511 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3512 char* req_addr,
3513 bool exec) {
3514 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3515 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3516 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3517
3518 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3519 char* addr = (char*)::mmap(req_addr, bytes, prot,
3520 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3521 -1, 0);
3522
3523 if (addr == MAP_FAILED) {
3524 warn_on_large_pages_failure(req_addr, bytes, errno);
3525 return NULL;
3526 }
3527
3528 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3529
3530 return addr;
3531 }
3532
3533 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3534 size_t alignment,
3535 char* req_addr,
3536 bool exec) {
3537 size_t large_page_size = os::large_page_size();
3538
3539 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3540
3541 // Allocate small pages.
3542
3543 char* start;
3544 if (req_addr != NULL) {
3545 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3546 assert(is_size_aligned(bytes, alignment), "Must be");
3547 start = os::reserve_memory(bytes, req_addr);
3548 assert(start == NULL || start == req_addr, "Must be");
3549 } else {
3550 start = os::reserve_memory_aligned(bytes, alignment);
3551 }
3552
3553 if (start == NULL) {
3554 return NULL;
3555 }
3556
3557 assert(is_ptr_aligned(start, alignment), "Must be");
3558
3559 if (MemTracker::tracking_level() > NMT_minimal) {
3560 // os::reserve_memory_special will record this memory area.
3561 // Need to release it here to prevent overlapping reservations.
3562 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3563 tkr.record((address)start, bytes);
3564 }
3565
3566 char* end = start + bytes;
3567
3568 // Find the regions of the allocated chunk that can be promoted to large pages.
3569 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3570 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3571
3572 size_t lp_bytes = lp_end - lp_start;
3573
3574 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3575
3576 if (lp_bytes == 0) {
3577 // The mapped region doesn't even span the start and the end of a large page.
3578 // Fall back to allocate a non-special area.
3579 ::munmap(start, end - start);
3580 return NULL;
3581 }
3582
3583 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3584
3585
3586 void* result;
3587
3588 if (start != lp_start) {
3589 result = ::mmap(start, lp_start - start, prot,
3590 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3591 -1, 0);
3592 if (result == MAP_FAILED) {
3593 ::munmap(lp_start, end - lp_start);
3594 return NULL;
3595 }
3596 }
3597
3598 result = ::mmap(lp_start, lp_bytes, prot,
3599 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3600 -1, 0);
3601 if (result == MAP_FAILED) {
3602 warn_on_large_pages_failure(req_addr, bytes, errno);
3603 // If the mmap above fails, the large pages region will be unmapped and we
3604 // have regions before and after with small pages. Release these regions.
3605 //
3606 // | mapped | unmapped | mapped |
3607 // ^ ^ ^ ^
3608 // start lp_start lp_end end
3609 //
3610 ::munmap(start, lp_start - start);
3611 ::munmap(lp_end, end - lp_end);
3612 return NULL;
3613 }
3614
3615 if (lp_end != end) {
3616 result = ::mmap(lp_end, end - lp_end, prot,
3617 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3618 -1, 0);
3619 if (result == MAP_FAILED) {
3620 ::munmap(start, lp_end - start);
3621 return NULL;
3622 }
3623 }
3624
3625 return start;
3626 }
3627
3628 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3629 size_t alignment,
3630 char* req_addr,
3631 bool exec) {
3632 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3633 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3634 assert(is_power_of_2(alignment), "Must be");
3635 assert(is_power_of_2(os::large_page_size()), "Must be");
3636 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3637
3638 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3639 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3640 } else {
3641 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3642 }
3643 }
3644
3645 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3646 char* req_addr, bool exec) {
3647 assert(UseLargePages, "only for large pages");
3648
3649 char* addr;
3650 if (UseSHM) {
3651 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3652 } else {
3653 assert(UseHugeTLBFS, "must be");
3654 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3655 }
3656
3657 if (addr != NULL) {
3658 if (UseNUMAInterleaving) {
3659 numa_make_global(addr, bytes);
3660 }
3661
3662 // The memory is committed
3663 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3664 }
3665
3666 return addr;
3667 }
3668
3669 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3670 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3671 return shmdt(base) == 0;
3672 }
3673
3674 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3675 return pd_release_memory(base, bytes);
3676 }
3677
3678 bool os::release_memory_special(char* base, size_t bytes) {
3679 bool res;
3680 if (MemTracker::tracking_level() > NMT_minimal) {
3681 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3682 res = os::Linux::release_memory_special_impl(base, bytes);
3683 if (res) {
3684 tkr.record((address)base, bytes);
3685 }
3686
3687 } else {
3688 res = os::Linux::release_memory_special_impl(base, bytes);
3689 }
3690 return res;
3691 }
3692
3693 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3694 assert(UseLargePages, "only for large pages");
3695 bool res;
3696
3697 if (UseSHM) {
3698 res = os::Linux::release_memory_special_shm(base, bytes);
3699 } else {
3700 assert(UseHugeTLBFS, "must be");
3701 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3702 }
3703 return res;
3704 }
3705
3706 size_t os::large_page_size() {
3707 return _large_page_size;
3708 }
3709
3710 // With SysV SHM the entire memory region must be allocated as shared
3711 // memory.
3712 // HugeTLBFS allows application to commit large page memory on demand.
3713 // However, when committing memory with HugeTLBFS fails, the region
3714 // that was supposed to be committed will lose the old reservation
3715 // and allow other threads to steal that memory region. Because of this
3716 // behavior we can't commit HugeTLBFS memory.
3717 bool os::can_commit_large_page_memory() {
3718 return UseTransparentHugePages;
3719 }
3720
3721 bool os::can_execute_large_page_memory() {
3722 return UseTransparentHugePages || UseHugeTLBFS;
3723 }
3724
3725 // Reserve memory at an arbitrary address, only if that area is
3726 // available (and not reserved for something else).
3727
3728 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3729 const int max_tries = 10;
3730 char* base[max_tries];
3731 size_t size[max_tries];
3732 const size_t gap = 0x000000;
3733
3734 // Assert only that the size is a multiple of the page size, since
3735 // that's all that mmap requires, and since that's all we really know
3736 // about at this low abstraction level. If we need higher alignment,
3737 // we can either pass an alignment to this method or verify alignment
3738 // in one of the methods further up the call chain. See bug 5044738.
3739 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3740
3741 // Repeatedly allocate blocks until the block is allocated at the
3742 // right spot. Give up after max_tries. Note that reserve_memory() will
3743 // automatically update _highest_vm_reserved_address if the call is
3744 // successful. The variable tracks the highest memory address every reserved
3745 // by JVM. It is used to detect heap-stack collision if running with
3746 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3747 // space than needed, it could confuse the collision detecting code. To
3748 // solve the problem, save current _highest_vm_reserved_address and
3749 // calculate the correct value before return.
3750 address old_highest = _highest_vm_reserved_address;
3751
3752 // Linux mmap allows caller to pass an address as hint; give it a try first,
3753 // if kernel honors the hint then we can return immediately.
3754 char * addr = anon_mmap(requested_addr, bytes, false);
3755 if (addr == requested_addr) {
3756 return requested_addr;
3757 }
3758
3759 if (addr != NULL) {
3760 // mmap() is successful but it fails to reserve at the requested address
3761 anon_munmap(addr, bytes);
3762 }
3763
3764 int i;
3765 for (i = 0; i < max_tries; ++i) {
3766 base[i] = reserve_memory(bytes);
3767
3768 if (base[i] != NULL) {
3769 // Is this the block we wanted?
3770 if (base[i] == requested_addr) {
3771 size[i] = bytes;
3772 break;
3773 }
3774
3775 // Does this overlap the block we wanted? Give back the overlapped
3776 // parts and try again.
3777
3778 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3779 if (top_overlap >= 0 && top_overlap < bytes) {
3780 unmap_memory(base[i], top_overlap);
3781 base[i] += top_overlap;
3782 size[i] = bytes - top_overlap;
3783 } else {
3784 size_t bottom_overlap = base[i] + bytes - requested_addr;
3785 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3786 unmap_memory(requested_addr, bottom_overlap);
3787 size[i] = bytes - bottom_overlap;
3788 } else {
3789 size[i] = bytes;
3790 }
3791 }
3792 }
3793 }
3794
3795 // Give back the unused reserved pieces.
3796
3797 for (int j = 0; j < i; ++j) {
3798 if (base[j] != NULL) {
3799 unmap_memory(base[j], size[j]);
3800 }
3801 }
3802
3803 if (i < max_tries) {
3804 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3805 return requested_addr;
3806 } else {
3807 _highest_vm_reserved_address = old_highest;
3808 return NULL;
3809 }
3810 }
3811
3812 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3813 return ::read(fd, buf, nBytes);
3814 }
3815
3816 // Short sleep, direct OS call.
3817 //
3818 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3819 // sched_yield(2) will actually give up the CPU:
3820 //
3821 // * Alone on this pariticular CPU, keeps running.
3822 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3823 // (pre 2.6.39).
3824 //
3825 // So calling this with 0 is an alternative.
3826 //
3827 void os::naked_short_sleep(jlong ms) {
3828 struct timespec req;
3829
3830 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3831 req.tv_sec = 0;
3832 if (ms > 0) {
3833 req.tv_nsec = (ms % 1000) * 1000000;
3834 } else {
3835 req.tv_nsec = 1;
3836 }
3837
3838 nanosleep(&req, NULL);
3839
3840 return;
3841 }
3842
3843 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3844 void os::infinite_sleep() {
3845 while (true) { // sleep forever ...
3846 ::sleep(100); // ... 100 seconds at a time
3847 }
3848 }
3849
3850 // Used to convert frequent JVM_Yield() to nops
3851 bool os::dont_yield() {
3852 return DontYieldALot;
3853 }
3854
3855 void os::naked_yield() {
3856 sched_yield();
3857 }
3858
3859 ////////////////////////////////////////////////////////////////////////////////
3860 // thread priority support
3861
3862 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3863 // only supports dynamic priority, static priority must be zero. For real-time
3864 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3865 // However, for large multi-threaded applications, SCHED_RR is not only slower
3866 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3867 // of 5 runs - Sep 2005).
3868 //
3869 // The following code actually changes the niceness of kernel-thread/LWP. It
3870 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3871 // not the entire user process, and user level threads are 1:1 mapped to kernel
3872 // threads. It has always been the case, but could change in the future. For
3873 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3874 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3875
3876 int os::java_to_os_priority[CriticalPriority + 1] = {
3877 19, // 0 Entry should never be used
3878
3879 4, // 1 MinPriority
3880 3, // 2
3881 2, // 3
3882
3883 1, // 4
3884 0, // 5 NormPriority
3885 -1, // 6
3886
3887 -2, // 7
3888 -3, // 8
3889 -4, // 9 NearMaxPriority
3890
3891 -5, // 10 MaxPriority
3892
3893 -5 // 11 CriticalPriority
3894 };
3895
3896 static int prio_init() {
3897 if (ThreadPriorityPolicy == 1) {
3898 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3899 // if effective uid is not root. Perhaps, a more elegant way of doing
3900 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3901 if (geteuid() != 0) {
3902 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3903 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3904 }
3905 ThreadPriorityPolicy = 0;
3906 }
3907 }
3908 if (UseCriticalJavaThreadPriority) {
3909 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3910 }
3911 return 0;
3912 }
3913
3914 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3915 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3916
3917 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3918 return (ret == 0) ? OS_OK : OS_ERR;
3919 }
3920
3921 OSReturn os::get_native_priority(const Thread* const thread,
3922 int *priority_ptr) {
3923 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3924 *priority_ptr = java_to_os_priority[NormPriority];
3925 return OS_OK;
3926 }
3927
3928 errno = 0;
3929 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3930 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3931 }
3932
3933 // Hint to the underlying OS that a task switch would not be good.
3934 // Void return because it's a hint and can fail.
3935 void os::hint_no_preempt() {}
3936
3937 ////////////////////////////////////////////////////////////////////////////////
3938 // suspend/resume support
3939
3940 // the low-level signal-based suspend/resume support is a remnant from the
3941 // old VM-suspension that used to be for java-suspension, safepoints etc,
3942 // within hotspot. Now there is a single use-case for this:
3943 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3944 // that runs in the watcher thread.
3945 // The remaining code is greatly simplified from the more general suspension
3946 // code that used to be used.
3947 //
3948 // The protocol is quite simple:
3949 // - suspend:
3950 // - sends a signal to the target thread
3951 // - polls the suspend state of the osthread using a yield loop
3952 // - target thread signal handler (SR_handler) sets suspend state
3953 // and blocks in sigsuspend until continued
3954 // - resume:
3955 // - sets target osthread state to continue
3956 // - sends signal to end the sigsuspend loop in the SR_handler
3957 //
3958 // Note that the SR_lock plays no role in this suspend/resume protocol.
3959
3960 static void resume_clear_context(OSThread *osthread) {
3961 osthread->set_ucontext(NULL);
3962 osthread->set_siginfo(NULL);
3963 }
3964
3965 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3966 ucontext_t* context) {
3967 osthread->set_ucontext(context);
3968 osthread->set_siginfo(siginfo);
3969 }
3970
3971 // Handler function invoked when a thread's execution is suspended or
3972 // resumed. We have to be careful that only async-safe functions are
3973 // called here (Note: most pthread functions are not async safe and
3974 // should be avoided.)
3975 //
3976 // Note: sigwait() is a more natural fit than sigsuspend() from an
3977 // interface point of view, but sigwait() prevents the signal hander
3978 // from being run. libpthread would get very confused by not having
3979 // its signal handlers run and prevents sigwait()'s use with the
3980 // mutex granting granting signal.
3981 //
3982 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3983 //
3984 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3985 // Save and restore errno to avoid confusing native code with EINTR
3986 // after sigsuspend.
3987 int old_errno = errno;
3988
3989 Thread* thread = Thread::current();
3990 OSThread* osthread = thread->osthread();
3991 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3992
3993 os::SuspendResume::State current = osthread->sr.state();
3994 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3995 suspend_save_context(osthread, siginfo, context);
3996
3997 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3998 os::SuspendResume::State state = osthread->sr.suspended();
3999 if (state == os::SuspendResume::SR_SUSPENDED) {
4000 sigset_t suspend_set; // signals for sigsuspend()
4001
4002 // get current set of blocked signals and unblock resume signal
4003 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4004 sigdelset(&suspend_set, SR_signum);
4005
4006 sr_semaphore.signal();
4007 // wait here until we are resumed
4008 while (1) {
4009 sigsuspend(&suspend_set);
4010
4011 os::SuspendResume::State result = osthread->sr.running();
4012 if (result == os::SuspendResume::SR_RUNNING) {
4013 sr_semaphore.signal();
4014 break;
4015 }
4016 }
4017
4018 } else if (state == os::SuspendResume::SR_RUNNING) {
4019 // request was cancelled, continue
4020 } else {
4021 ShouldNotReachHere();
4022 }
4023
4024 resume_clear_context(osthread);
4025 } else if (current == os::SuspendResume::SR_RUNNING) {
4026 // request was cancelled, continue
4027 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4028 // ignore
4029 } else {
4030 // ignore
4031 }
4032
4033 errno = old_errno;
4034 }
4035
4036
4037 static int SR_initialize() {
4038 struct sigaction act;
4039 char *s;
4040 // Get signal number to use for suspend/resume
4041 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4042 int sig = ::strtol(s, 0, 10);
4043 if (sig > 0 || sig < _NSIG) {
4044 SR_signum = sig;
4045 }
4046 }
4047
4048 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4049 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4050
4051 sigemptyset(&SR_sigset);
4052 sigaddset(&SR_sigset, SR_signum);
4053
4054 // Set up signal handler for suspend/resume
4055 act.sa_flags = SA_RESTART|SA_SIGINFO;
4056 act.sa_handler = (void (*)(int)) SR_handler;
4057
4058 // SR_signum is blocked by default.
4059 // 4528190 - We also need to block pthread restart signal (32 on all
4060 // supported Linux platforms). Note that LinuxThreads need to block
4061 // this signal for all threads to work properly. So we don't have
4062 // to use hard-coded signal number when setting up the mask.
4063 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4064
4065 if (sigaction(SR_signum, &act, 0) == -1) {
4066 return -1;
4067 }
4068
4069 // Save signal flag
4070 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4071 return 0;
4072 }
4073
4074 static int sr_notify(OSThread* osthread) {
4075 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4076 assert_status(status == 0, status, "pthread_kill");
4077 return status;
4078 }
4079
4080 // "Randomly" selected value for how long we want to spin
4081 // before bailing out on suspending a thread, also how often
4082 // we send a signal to a thread we want to resume
4083 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4084 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4085
4086 // returns true on success and false on error - really an error is fatal
4087 // but this seems the normal response to library errors
4088 static bool do_suspend(OSThread* osthread) {
4089 assert(osthread->sr.is_running(), "thread should be running");
4090 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4091
4092 // mark as suspended and send signal
4093 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4094 // failed to switch, state wasn't running?
4095 ShouldNotReachHere();
4096 return false;
4097 }
4098
4099 if (sr_notify(osthread) != 0) {
4100 ShouldNotReachHere();
4101 }
4102
4103 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4104 while (true) {
4105 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4106 break;
4107 } else {
4108 // timeout
4109 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4110 if (cancelled == os::SuspendResume::SR_RUNNING) {
4111 return false;
4112 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4113 // make sure that we consume the signal on the semaphore as well
4114 sr_semaphore.wait();
4115 break;
4116 } else {
4117 ShouldNotReachHere();
4118 return false;
4119 }
4120 }
4121 }
4122
4123 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4124 return true;
4125 }
4126
4127 static void do_resume(OSThread* osthread) {
4128 assert(osthread->sr.is_suspended(), "thread should be suspended");
4129 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4130
4131 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4132 // failed to switch to WAKEUP_REQUEST
4133 ShouldNotReachHere();
4134 return;
4135 }
4136
4137 while (true) {
4138 if (sr_notify(osthread) == 0) {
4139 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4140 if (osthread->sr.is_running()) {
4141 return;
4142 }
4143 }
4144 } else {
4145 ShouldNotReachHere();
4146 }
4147 }
4148
4149 guarantee(osthread->sr.is_running(), "Must be running!");
4150 }
4151
4152 ///////////////////////////////////////////////////////////////////////////////////
4153 // signal handling (except suspend/resume)
4154
4155 // This routine may be used by user applications as a "hook" to catch signals.
4156 // The user-defined signal handler must pass unrecognized signals to this
4157 // routine, and if it returns true (non-zero), then the signal handler must
4158 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4159 // routine will never retun false (zero), but instead will execute a VM panic
4160 // routine kill the process.
4161 //
4162 // If this routine returns false, it is OK to call it again. This allows
4163 // the user-defined signal handler to perform checks either before or after
4164 // the VM performs its own checks. Naturally, the user code would be making
4165 // a serious error if it tried to handle an exception (such as a null check
4166 // or breakpoint) that the VM was generating for its own correct operation.
4167 //
4168 // This routine may recognize any of the following kinds of signals:
4169 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4170 // It should be consulted by handlers for any of those signals.
4171 //
4172 // The caller of this routine must pass in the three arguments supplied
4173 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4174 // field of the structure passed to sigaction(). This routine assumes that
4175 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4176 //
4177 // Note that the VM will print warnings if it detects conflicting signal
4178 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4179 //
4180 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4181 siginfo_t* siginfo,
4182 void* ucontext,
4183 int abort_if_unrecognized);
4184
4185 void signalHandler(int sig, siginfo_t* info, void* uc) {
4186 assert(info != NULL && uc != NULL, "it must be old kernel");
4187 int orig_errno = errno; // Preserve errno value over signal handler.
4188 JVM_handle_linux_signal(sig, info, uc, true);
4189 errno = orig_errno;
4190 }
4191
4192
4193 // This boolean allows users to forward their own non-matching signals
4194 // to JVM_handle_linux_signal, harmlessly.
4195 bool os::Linux::signal_handlers_are_installed = false;
4196
4197 // For signal-chaining
4198 struct sigaction os::Linux::sigact[MAXSIGNUM];
4199 unsigned int os::Linux::sigs = 0;
4200 bool os::Linux::libjsig_is_loaded = false;
4201 typedef struct sigaction *(*get_signal_t)(int);
4202 get_signal_t os::Linux::get_signal_action = NULL;
4203
4204 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4205 struct sigaction *actp = NULL;
4206
4207 if (libjsig_is_loaded) {
4208 // Retrieve the old signal handler from libjsig
4209 actp = (*get_signal_action)(sig);
4210 }
4211 if (actp == NULL) {
4212 // Retrieve the preinstalled signal handler from jvm
4213 actp = get_preinstalled_handler(sig);
4214 }
4215
4216 return actp;
4217 }
4218
4219 static bool call_chained_handler(struct sigaction *actp, int sig,
4220 siginfo_t *siginfo, void *context) {
4221 // Call the old signal handler
4222 if (actp->sa_handler == SIG_DFL) {
4223 // It's more reasonable to let jvm treat it as an unexpected exception
4224 // instead of taking the default action.
4225 return false;
4226 } else if (actp->sa_handler != SIG_IGN) {
4227 if ((actp->sa_flags & SA_NODEFER) == 0) {
4228 // automaticlly block the signal
4229 sigaddset(&(actp->sa_mask), sig);
4230 }
4231
4232 sa_handler_t hand;
4233 sa_sigaction_t sa;
4234 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4235 // retrieve the chained handler
4236 if (siginfo_flag_set) {
4237 sa = actp->sa_sigaction;
4238 } else {
4239 hand = actp->sa_handler;
4240 }
4241
4242 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4243 actp->sa_handler = SIG_DFL;
4244 }
4245
4246 // try to honor the signal mask
4247 sigset_t oset;
4248 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4249
4250 // call into the chained handler
4251 if (siginfo_flag_set) {
4252 (*sa)(sig, siginfo, context);
4253 } else {
4254 (*hand)(sig);
4255 }
4256
4257 // restore the signal mask
4258 pthread_sigmask(SIG_SETMASK, &oset, 0);
4259 }
4260 // Tell jvm's signal handler the signal is taken care of.
4261 return true;
4262 }
4263
4264 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4265 bool chained = false;
4266 // signal-chaining
4267 if (UseSignalChaining) {
4268 struct sigaction *actp = get_chained_signal_action(sig);
4269 if (actp != NULL) {
4270 chained = call_chained_handler(actp, sig, siginfo, context);
4271 }
4272 }
4273 return chained;
4274 }
4275
4276 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4277 if ((((unsigned int)1 << sig) & sigs) != 0) {
4278 return &sigact[sig];
4279 }
4280 return NULL;
4281 }
4282
4283 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4284 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4285 sigact[sig] = oldAct;
4286 sigs |= (unsigned int)1 << sig;
4287 }
4288
4289 // for diagnostic
4290 int os::Linux::sigflags[MAXSIGNUM];
4291
4292 int os::Linux::get_our_sigflags(int sig) {
4293 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4294 return sigflags[sig];
4295 }
4296
4297 void os::Linux::set_our_sigflags(int sig, int flags) {
4298 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4299 sigflags[sig] = flags;
4300 }
4301
4302 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4303 // Check for overwrite.
4304 struct sigaction oldAct;
4305 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4306
4307 void* oldhand = oldAct.sa_sigaction
4308 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4309 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4310 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4311 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4312 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4313 if (AllowUserSignalHandlers || !set_installed) {
4314 // Do not overwrite; user takes responsibility to forward to us.
4315 return;
4316 } else if (UseSignalChaining) {
4317 // save the old handler in jvm
4318 save_preinstalled_handler(sig, oldAct);
4319 // libjsig also interposes the sigaction() call below and saves the
4320 // old sigaction on it own.
4321 } else {
4322 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4323 "%#lx for signal %d.", (long)oldhand, sig));
4324 }
4325 }
4326
4327 struct sigaction sigAct;
4328 sigfillset(&(sigAct.sa_mask));
4329 sigAct.sa_handler = SIG_DFL;
4330 if (!set_installed) {
4331 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4332 } else {
4333 sigAct.sa_sigaction = signalHandler;
4334 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4335 }
4336 // Save flags, which are set by ours
4337 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4338 sigflags[sig] = sigAct.sa_flags;
4339
4340 int ret = sigaction(sig, &sigAct, &oldAct);
4341 assert(ret == 0, "check");
4342
4343 void* oldhand2 = oldAct.sa_sigaction
4344 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4345 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4346 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4347 }
4348
4349 // install signal handlers for signals that HotSpot needs to
4350 // handle in order to support Java-level exception handling.
4351
4352 void os::Linux::install_signal_handlers() {
4353 if (!signal_handlers_are_installed) {
4354 signal_handlers_are_installed = true;
4355
4356 // signal-chaining
4357 typedef void (*signal_setting_t)();
4358 signal_setting_t begin_signal_setting = NULL;
4359 signal_setting_t end_signal_setting = NULL;
4360 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4361 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4362 if (begin_signal_setting != NULL) {
4363 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4364 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4365 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4366 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4367 libjsig_is_loaded = true;
4368 assert(UseSignalChaining, "should enable signal-chaining");
4369 }
4370 if (libjsig_is_loaded) {
4371 // Tell libjsig jvm is setting signal handlers
4372 (*begin_signal_setting)();
4373 }
4374
4375 set_signal_handler(SIGSEGV, true);
4376 set_signal_handler(SIGPIPE, true);
4377 set_signal_handler(SIGBUS, true);
4378 set_signal_handler(SIGILL, true);
4379 set_signal_handler(SIGFPE, true);
4380 #if defined(PPC64)
4381 set_signal_handler(SIGTRAP, true);
4382 #endif
4383 set_signal_handler(SIGXFSZ, true);
4384
4385 if (libjsig_is_loaded) {
4386 // Tell libjsig jvm finishes setting signal handlers
4387 (*end_signal_setting)();
4388 }
4389
4390 // We don't activate signal checker if libjsig is in place, we trust ourselves
4391 // and if UserSignalHandler is installed all bets are off.
4392 // Log that signal checking is off only if -verbose:jni is specified.
4393 if (CheckJNICalls) {
4394 if (libjsig_is_loaded) {
4395 if (PrintJNIResolving) {
4396 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4397 }
4398 check_signals = false;
4399 }
4400 if (AllowUserSignalHandlers) {
4401 if (PrintJNIResolving) {
4402 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4403 }
4404 check_signals = false;
4405 }
4406 }
4407 }
4408 }
4409
4410 // This is the fastest way to get thread cpu time on Linux.
4411 // Returns cpu time (user+sys) for any thread, not only for current.
4412 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4413 // It might work on 2.6.10+ with a special kernel/glibc patch.
4414 // For reference, please, see IEEE Std 1003.1-2004:
4415 // http://www.unix.org/single_unix_specification
4416
4417 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4418 struct timespec tp;
4419 int rc = os::Linux::clock_gettime(clockid, &tp);
4420 assert(rc == 0, "clock_gettime is expected to return 0 code");
4421
4422 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4423 }
4424
4425 /////
4426 // glibc on Linux platform uses non-documented flag
4427 // to indicate, that some special sort of signal
4428 // trampoline is used.
4429 // We will never set this flag, and we should
4430 // ignore this flag in our diagnostic
4431 #ifdef SIGNIFICANT_SIGNAL_MASK
4432 #undef SIGNIFICANT_SIGNAL_MASK
4433 #endif
4434 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4435
4436 static const char* get_signal_handler_name(address handler,
4437 char* buf, int buflen) {
4438 int offset;
4439 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4440 if (found) {
4441 // skip directory names
4442 const char *p1, *p2;
4443 p1 = buf;
4444 size_t len = strlen(os::file_separator());
4445 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4446 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4447 } else {
4448 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4449 }
4450 return buf;
4451 }
4452
4453 static void print_signal_handler(outputStream* st, int sig,
4454 char* buf, size_t buflen) {
4455 struct sigaction sa;
4456
4457 sigaction(sig, NULL, &sa);
4458
4459 // See comment for SIGNIFICANT_SIGNAL_MASK define
4460 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4461
4462 st->print("%s: ", os::exception_name(sig, buf, buflen));
4463
4464 address handler = (sa.sa_flags & SA_SIGINFO)
4465 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4466 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4467
4468 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4469 st->print("SIG_DFL");
4470 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4471 st->print("SIG_IGN");
4472 } else {
4473 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4474 }
4475
4476 st->print(", sa_mask[0]=");
4477 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4478
4479 address rh = VMError::get_resetted_sighandler(sig);
4480 // May be, handler was resetted by VMError?
4481 if (rh != NULL) {
4482 handler = rh;
4483 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4484 }
4485
4486 st->print(", sa_flags=");
4487 os::Posix::print_sa_flags(st, sa.sa_flags);
4488
4489 // Check: is it our handler?
4490 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4491 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4492 // It is our signal handler
4493 // check for flags, reset system-used one!
4494 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4495 st->print(
4496 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4497 os::Linux::get_our_sigflags(sig));
4498 }
4499 }
4500 st->cr();
4501 }
4502
4503
4504 #define DO_SIGNAL_CHECK(sig) \
4505 do { \
4506 if (!sigismember(&check_signal_done, sig)) { \
4507 os::Linux::check_signal_handler(sig); \
4508 } \
4509 } while (0)
4510
4511 // This method is a periodic task to check for misbehaving JNI applications
4512 // under CheckJNI, we can add any periodic checks here
4513
4514 void os::run_periodic_checks() {
4515 if (check_signals == false) return;
4516
4517 // SEGV and BUS if overridden could potentially prevent
4518 // generation of hs*.log in the event of a crash, debugging
4519 // such a case can be very challenging, so we absolutely
4520 // check the following for a good measure:
4521 DO_SIGNAL_CHECK(SIGSEGV);
4522 DO_SIGNAL_CHECK(SIGILL);
4523 DO_SIGNAL_CHECK(SIGFPE);
4524 DO_SIGNAL_CHECK(SIGBUS);
4525 DO_SIGNAL_CHECK(SIGPIPE);
4526 DO_SIGNAL_CHECK(SIGXFSZ);
4527 #if defined(PPC64)
4528 DO_SIGNAL_CHECK(SIGTRAP);
4529 #endif
4530
4531 // ReduceSignalUsage allows the user to override these handlers
4532 // see comments at the very top and jvm_solaris.h
4533 if (!ReduceSignalUsage) {
4534 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4535 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4536 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4537 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4538 }
4539
4540 DO_SIGNAL_CHECK(SR_signum);
4541 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4542 }
4543
4544 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4545
4546 static os_sigaction_t os_sigaction = NULL;
4547
4548 void os::Linux::check_signal_handler(int sig) {
4549 char buf[O_BUFLEN];
4550 address jvmHandler = NULL;
4551
4552
4553 struct sigaction act;
4554 if (os_sigaction == NULL) {
4555 // only trust the default sigaction, in case it has been interposed
4556 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4557 if (os_sigaction == NULL) return;
4558 }
4559
4560 os_sigaction(sig, (struct sigaction*)NULL, &act);
4561
4562
4563 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4564
4565 address thisHandler = (act.sa_flags & SA_SIGINFO)
4566 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4567 : CAST_FROM_FN_PTR(address, act.sa_handler);
4568
4569
4570 switch (sig) {
4571 case SIGSEGV:
4572 case SIGBUS:
4573 case SIGFPE:
4574 case SIGPIPE:
4575 case SIGILL:
4576 case SIGXFSZ:
4577 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4578 break;
4579
4580 case SHUTDOWN1_SIGNAL:
4581 case SHUTDOWN2_SIGNAL:
4582 case SHUTDOWN3_SIGNAL:
4583 case BREAK_SIGNAL:
4584 jvmHandler = (address)user_handler();
4585 break;
4586
4587 case INTERRUPT_SIGNAL:
4588 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4589 break;
4590
4591 default:
4592 if (sig == SR_signum) {
4593 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4594 } else {
4595 return;
4596 }
4597 break;
4598 }
4599
4600 if (thisHandler != jvmHandler) {
4601 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4602 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4603 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4604 // No need to check this sig any longer
4605 sigaddset(&check_signal_done, sig);
4606 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4607 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4608 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4609 exception_name(sig, buf, O_BUFLEN));
4610 }
4611 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4612 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4613 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4614 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4615 // No need to check this sig any longer
4616 sigaddset(&check_signal_done, sig);
4617 }
4618
4619 // Dump all the signal
4620 if (sigismember(&check_signal_done, sig)) {
4621 print_signal_handlers(tty, buf, O_BUFLEN);
4622 }
4623 }
4624
4625 extern void report_error(char* file_name, int line_no, char* title,
4626 char* format, ...);
4627
4628 extern bool signal_name(int signo, char* buf, size_t len);
4629
4630 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4631 if (0 < exception_code && exception_code <= SIGRTMAX) {
4632 // signal
4633 if (!signal_name(exception_code, buf, size)) {
4634 jio_snprintf(buf, size, "SIG%d", exception_code);
4635 }
4636 return buf;
4637 } else {
4638 return NULL;
4639 }
4640 }
4641
4642 // this is called _before_ the most of global arguments have been parsed
4643 void os::init(void) {
4644 char dummy; // used to get a guess on initial stack address
4645 // first_hrtime = gethrtime();
4646
4647 // With LinuxThreads the JavaMain thread pid (primordial thread)
4648 // is different than the pid of the java launcher thread.
4649 // So, on Linux, the launcher thread pid is passed to the VM
4650 // via the sun.java.launcher.pid property.
4651 // Use this property instead of getpid() if it was correctly passed.
4652 // See bug 6351349.
4653 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4654
4655 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4656
4657 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4658
4659 init_random(1234567);
4660
4661 ThreadCritical::initialize();
4662
4663 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4664 if (Linux::page_size() == -1) {
4665 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4666 strerror(errno)));
4667 }
4668 init_page_sizes((size_t) Linux::page_size());
4669
4670 Linux::initialize_system_info();
4671
4672 // main_thread points to the aboriginal thread
4673 Linux::_main_thread = pthread_self();
4674
4675 Linux::clock_init();
4676 initial_time_count = javaTimeNanos();
4677
4678 // pthread_condattr initialization for monotonic clock
4679 int status;
4680 pthread_condattr_t* _condattr = os::Linux::condAttr();
4681 if ((status = pthread_condattr_init(_condattr)) != 0) {
4682 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4683 }
4684 // Only set the clock if CLOCK_MONOTONIC is available
4685 if (os::supports_monotonic_clock()) {
4686 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4687 if (status == EINVAL) {
4688 warning("Unable to use monotonic clock with relative timed-waits" \
4689 " - changes to the time-of-day clock may have adverse affects");
4690 } else {
4691 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4692 }
4693 }
4694 }
4695 // else it defaults to CLOCK_REALTIME
4696
4697 pthread_mutex_init(&dl_mutex, NULL);
4698
4699 // If the pagesize of the VM is greater than 8K determine the appropriate
4700 // number of initial guard pages. The user can change this with the
4701 // command line arguments, if needed.
4702 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4703 StackYellowPages = 1;
4704 StackRedPages = 1;
4705 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4706 }
4707
4708 // retrieve entry point for pthread_setname_np
4709 Linux::_pthread_setname_np =
4710 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4711
4712 }
4713
4714 // To install functions for atexit system call
4715 extern "C" {
4716 static void perfMemory_exit_helper() {
4717 perfMemory_exit();
4718 }
4719 }
4720
4721 // this is called _after_ the global arguments have been parsed
4722 jint os::init_2(void) {
4723 Linux::fast_thread_clock_init();
4724
4725 // Allocate a single page and mark it as readable for safepoint polling
4726 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4727 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4728
4729 os::set_polling_page(polling_page);
4730
4731 #ifndef PRODUCT
4732 if (Verbose && PrintMiscellaneous) {
4733 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4734 (intptr_t)polling_page);
4735 }
4736 #endif
4737
4738 if (!UseMembar) {
4739 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4740 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4741 os::set_memory_serialize_page(mem_serialize_page);
4742
4743 #ifndef PRODUCT
4744 if (Verbose && PrintMiscellaneous) {
4745 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4746 (intptr_t)mem_serialize_page);
4747 }
4748 #endif
4749 }
4750
4751 // initialize suspend/resume support - must do this before signal_sets_init()
4752 if (SR_initialize() != 0) {
4753 perror("SR_initialize failed");
4754 return JNI_ERR;
4755 }
4756
4757 Linux::signal_sets_init();
4758 Linux::install_signal_handlers();
4759
4760 // Check minimum allowable stack size for thread creation and to initialize
4761 // the java system classes, including StackOverflowError - depends on page
4762 // size. Add a page for compiler2 recursion in main thread.
4763 // Add in 2*BytesPerWord times page size to account for VM stack during
4764 // class initialization depending on 32 or 64 bit VM.
4765 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4766 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4767 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4768
4769 size_t threadStackSizeInBytes = ThreadStackSize * K;
4770 if (threadStackSizeInBytes != 0 &&
4771 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4772 tty->print_cr("\nThe stack size specified is too small, "
4773 "Specify at least %dk",
4774 os::Linux::min_stack_allowed/ K);
4775 return JNI_ERR;
4776 }
4777
4778 // Make the stack size a multiple of the page size so that
4779 // the yellow/red zones can be guarded.
4780 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4781 vm_page_size()));
4782
4783 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4784
4785 #if defined(IA32)
4786 workaround_expand_exec_shield_cs_limit();
4787 #endif
4788
4789 Linux::libpthread_init();
4790 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4791 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4792 Linux::glibc_version(), Linux::libpthread_version(),
4793 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4794 }
4795
4796 if (UseNUMA) {
4797 if (!Linux::libnuma_init()) {
4798 UseNUMA = false;
4799 } else {
4800 if ((Linux::numa_max_node() < 1)) {
4801 // There's only one node(they start from 0), disable NUMA.
4802 UseNUMA = false;
4803 }
4804 }
4805 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4806 // we can make the adaptive lgrp chunk resizing work. If the user specified
4807 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4808 // disable adaptive resizing.
4809 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4810 if (FLAG_IS_DEFAULT(UseNUMA)) {
4811 UseNUMA = false;
4812 } else {
4813 if (FLAG_IS_DEFAULT(UseLargePages) &&
4814 FLAG_IS_DEFAULT(UseSHM) &&
4815 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4816 UseLargePages = false;
4817 } else {
4818 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4819 UseAdaptiveSizePolicy = false;
4820 UseAdaptiveNUMAChunkSizing = false;
4821 }
4822 }
4823 }
4824 if (!UseNUMA && ForceNUMA) {
4825 UseNUMA = true;
4826 }
4827 }
4828
4829 if (MaxFDLimit) {
4830 // set the number of file descriptors to max. print out error
4831 // if getrlimit/setrlimit fails but continue regardless.
4832 struct rlimit nbr_files;
4833 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4834 if (status != 0) {
4835 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4836 perror("os::init_2 getrlimit failed");
4837 }
4838 } else {
4839 nbr_files.rlim_cur = nbr_files.rlim_max;
4840 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4841 if (status != 0) {
4842 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4843 perror("os::init_2 setrlimit failed");
4844 }
4845 }
4846 }
4847 }
4848
4849 // Initialize lock used to serialize thread creation (see os::create_thread)
4850 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4851
4852 // at-exit methods are called in the reverse order of their registration.
4853 // atexit functions are called on return from main or as a result of a
4854 // call to exit(3C). There can be only 32 of these functions registered
4855 // and atexit() does not set errno.
4856
4857 if (PerfAllowAtExitRegistration) {
4858 // only register atexit functions if PerfAllowAtExitRegistration is set.
4859 // atexit functions can be delayed until process exit time, which
4860 // can be problematic for embedded VM situations. Embedded VMs should
4861 // call DestroyJavaVM() to assure that VM resources are released.
4862
4863 // note: perfMemory_exit_helper atexit function may be removed in
4864 // the future if the appropriate cleanup code can be added to the
4865 // VM_Exit VMOperation's doit method.
4866 if (atexit(perfMemory_exit_helper) != 0) {
4867 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4868 }
4869 }
4870
4871 // initialize thread priority policy
4872 prio_init();
4873
4874 return JNI_OK;
4875 }
4876
4877 // this is called at the end of vm_initialization
4878 void os::init_3(void) {
4879 #ifdef JAVASE_EMBEDDED
4880 // Start the MemNotifyThread
4881 if (LowMemoryProtection) {
4882 MemNotifyThread::start();
4883 }
4884 return;
4885 #endif
4886 }
4887
4888 // Mark the polling page as unreadable
4889 void os::make_polling_page_unreadable(void) {
4890 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4891 fatal("Could not disable polling page");
4892 }
4893 }
4894
4895 // Mark the polling page as readable
4896 void os::make_polling_page_readable(void) {
4897 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4898 fatal("Could not enable polling page");
4899 }
4900 }
4901
4902 int os::active_processor_count() {
4903 // Linux doesn't yet have a (official) notion of processor sets,
4904 // so just return the number of online processors.
4905 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4906 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4907 return online_cpus;
4908 }
4909
4910 void os::set_native_thread_name(const char *name) {
4911 if (Linux::_pthread_setname_np) {
4912 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4913 snprintf(buf, sizeof(buf), "%s", name);
4914 buf[sizeof(buf) - 1] = '\0';
4915 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4916 // ERANGE should not happen; all other errors should just be ignored.
4917 assert(rc != ERANGE, "pthread_setname_np failed");
4918 }
4919 }
4920
4921 bool os::distribute_processes(uint length, uint* distribution) {
4922 // Not yet implemented.
4923 return false;
4924 }
4925
4926 bool os::bind_to_processor(uint processor_id) {
4927 // Not yet implemented.
4928 return false;
4929 }
4930
4931 ///
4932
4933 void os::SuspendedThreadTask::internal_do_task() {
4934 if (do_suspend(_thread->osthread())) {
4935 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4936 do_task(context);
4937 do_resume(_thread->osthread());
4938 }
4939 }
4940
4941 class PcFetcher : public os::SuspendedThreadTask {
4942 public:
4943 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4944 ExtendedPC result();
4945 protected:
4946 void do_task(const os::SuspendedThreadTaskContext& context);
4947 private:
4948 ExtendedPC _epc;
4949 };
4950
4951 ExtendedPC PcFetcher::result() {
4952 guarantee(is_done(), "task is not done yet.");
4953 return _epc;
4954 }
4955
4956 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4957 Thread* thread = context.thread();
4958 OSThread* osthread = thread->osthread();
4959 if (osthread->ucontext() != NULL) {
4960 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4961 } else {
4962 // NULL context is unexpected, double-check this is the VMThread
4963 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4964 }
4965 }
4966
4967 // Suspends the target using the signal mechanism and then grabs the PC before
4968 // resuming the target. Used by the flat-profiler only
4969 ExtendedPC os::get_thread_pc(Thread* thread) {
4970 // Make sure that it is called by the watcher for the VMThread
4971 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4972 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4973
4974 PcFetcher fetcher(thread);
4975 fetcher.run();
4976 return fetcher.result();
4977 }
4978
4979 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4980 pthread_mutex_t *_mutex,
4981 const struct timespec *_abstime) {
4982 if (is_NPTL()) {
4983 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4984 } else {
4985 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4986 // word back to default 64bit precision if condvar is signaled. Java
4987 // wants 53bit precision. Save and restore current value.
4988 int fpu = get_fpu_control_word();
4989 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4990 set_fpu_control_word(fpu);
4991 return status;
4992 }
4993 }
4994
4995 ////////////////////////////////////////////////////////////////////////////////
4996 // debug support
4997
4998 bool os::find(address addr, outputStream* st) {
4999 Dl_info dlinfo;
5000 memset(&dlinfo, 0, sizeof(dlinfo));
5001 if (dladdr(addr, &dlinfo) != 0) {
5002 st->print(PTR_FORMAT ": ", addr);
5003 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5004 st->print("%s+%#x", dlinfo.dli_sname,
5005 addr - (intptr_t)dlinfo.dli_saddr);
5006 } else if (dlinfo.dli_fbase != NULL) {
5007 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5008 } else {
5009 st->print("<absolute address>");
5010 }
5011 if (dlinfo.dli_fname != NULL) {
5012 st->print(" in %s", dlinfo.dli_fname);
5013 }
5014 if (dlinfo.dli_fbase != NULL) {
5015 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5016 }
5017 st->cr();
5018
5019 if (Verbose) {
5020 // decode some bytes around the PC
5021 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5022 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5023 address lowest = (address) dlinfo.dli_sname;
5024 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5025 if (begin < lowest) begin = lowest;
5026 Dl_info dlinfo2;
5027 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5028 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5029 end = (address) dlinfo2.dli_saddr;
5030 }
5031 Disassembler::decode(begin, end, st);
5032 }
5033 return true;
5034 }
5035 return false;
5036 }
5037
5038 ////////////////////////////////////////////////////////////////////////////////
5039 // misc
5040
5041 // This does not do anything on Linux. This is basically a hook for being
5042 // able to use structured exception handling (thread-local exception filters)
5043 // on, e.g., Win32.
5044 void
5045 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5046 JavaCallArguments* args, Thread* thread) {
5047 f(value, method, args, thread);
5048 }
5049
5050 void os::print_statistics() {
5051 }
5052
5053 int os::message_box(const char* title, const char* message) {
5054 int i;
5055 fdStream err(defaultStream::error_fd());
5056 for (i = 0; i < 78; i++) err.print_raw("=");
5057 err.cr();
5058 err.print_raw_cr(title);
5059 for (i = 0; i < 78; i++) err.print_raw("-");
5060 err.cr();
5061 err.print_raw_cr(message);
5062 for (i = 0; i < 78; i++) err.print_raw("=");
5063 err.cr();
5064
5065 char buf[16];
5066 // Prevent process from exiting upon "read error" without consuming all CPU
5067 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5068
5069 return buf[0] == 'y' || buf[0] == 'Y';
5070 }
5071
5072 int os::stat(const char *path, struct stat *sbuf) {
5073 char pathbuf[MAX_PATH];
5074 if (strlen(path) > MAX_PATH - 1) {
5075 errno = ENAMETOOLONG;
5076 return -1;
5077 }
5078 os::native_path(strcpy(pathbuf, path));
5079 return ::stat(pathbuf, sbuf);
5080 }
5081
5082 bool os::check_heap(bool force) {
5083 return true;
5084 }
5085
5086 // Is a (classpath) directory empty?
5087 bool os::dir_is_empty(const char* path) {
5088 DIR *dir = NULL;
5089 struct dirent *ptr;
5090
5091 dir = opendir(path);
5092 if (dir == NULL) return true;
5093
5094 // Scan the directory
5095 bool result = true;
5096 char buf[sizeof(struct dirent) + MAX_PATH];
5097 while (result && (ptr = ::readdir(dir)) != NULL) {
5098 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5099 result = false;
5100 }
5101 }
5102 closedir(dir);
5103 return result;
5104 }
5105
5106 // This code originates from JDK's sysOpen and open64_w
5107 // from src/solaris/hpi/src/system_md.c
5108
5109 int os::open(const char *path, int oflag, int mode) {
5110 if (strlen(path) > MAX_PATH - 1) {
5111 errno = ENAMETOOLONG;
5112 return -1;
5113 }
5114 int fd;
5115
5116 fd = ::open64(path, oflag, mode);
5117 if (fd == -1) return -1;
5118
5119 //If the open succeeded, the file might still be a directory
5120 {
5121 struct stat64 buf64;
5122 int ret = ::fstat64(fd, &buf64);
5123 int st_mode = buf64.st_mode;
5124
5125 if (ret != -1) {
5126 if ((st_mode & S_IFMT) == S_IFDIR) {
5127 errno = EISDIR;
5128 ::close(fd);
5129 return -1;
5130 }
5131 } else {
5132 ::close(fd);
5133 return -1;
5134 }
5135 }
5136
5137 // All file descriptors that are opened in the JVM and not
5138 // specifically destined for a subprocess should have the
5139 // close-on-exec flag set. If we don't set it, then careless 3rd
5140 // party native code might fork and exec without closing all
5141 // appropriate file descriptors (e.g. as we do in closeDescriptors in
5142 // UNIXProcess.c), and this in turn might:
5143 //
5144 // - cause end-of-file to fail to be detected on some file
5145 // descriptors, resulting in mysterious hangs, or
5146 //
5147 // - might cause an fopen in the subprocess to fail on a system
5148 // suffering from bug 1085341.
5149 //
5150 // (Yes, the default setting of the close-on-exec flag is a Unix
5151 // design flaw)
5152 //
5153 // See:
5154 // 1085341: 32-bit stdio routines should support file descriptors >255
5155 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5156 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5157 //
5158 #ifdef FD_CLOEXEC
5159 {
5160 int flags = ::fcntl(fd, F_GETFD);
5161 if (flags != -1) {
5162 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5163 }
5164 }
5165 #endif
5166
5167 return fd;
5168 }
5169
5170
5171 // create binary file, rewriting existing file if required
5172 int os::create_binary_file(const char* path, bool rewrite_existing) {
5173 int oflags = O_WRONLY | O_CREAT;
5174 if (!rewrite_existing) {
5175 oflags |= O_EXCL;
5176 }
5177 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5178 }
5179
5180 // return current position of file pointer
5181 jlong os::current_file_offset(int fd) {
5182 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5183 }
5184
5185 // move file pointer to the specified offset
5186 jlong os::seek_to_file_offset(int fd, jlong offset) {
5187 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5188 }
5189
5190 // This code originates from JDK's sysAvailable
5191 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5192
5193 int os::available(int fd, jlong *bytes) {
5194 jlong cur, end;
5195 int mode;
5196 struct stat64 buf64;
5197
5198 if (::fstat64(fd, &buf64) >= 0) {
5199 mode = buf64.st_mode;
5200 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5201 // XXX: is the following call interruptible? If so, this might
5202 // need to go through the INTERRUPT_IO() wrapper as for other
5203 // blocking, interruptible calls in this file.
5204 int n;
5205 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5206 *bytes = n;
5207 return 1;
5208 }
5209 }
5210 }
5211 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5212 return 0;
5213 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5214 return 0;
5215 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5216 return 0;
5217 }
5218 *bytes = end - cur;
5219 return 1;
5220 }
5221
5222 int os::socket_available(int fd, jint *pbytes) {
5223 // Linux doc says EINTR not returned, unlike Solaris
5224 int ret = ::ioctl(fd, FIONREAD, pbytes);
5225
5226 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5227 // is expected to return 0 on failure and 1 on success to the jdk.
5228 return (ret < 0) ? 0 : 1;
5229 }
5230
5231 // Map a block of memory.
5232 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5233 char *addr, size_t bytes, bool read_only,
5234 bool allow_exec) {
5235 int prot;
5236 int flags = MAP_PRIVATE;
5237
5238 if (read_only) {
5239 prot = PROT_READ;
5240 } else {
5241 prot = PROT_READ | PROT_WRITE;
5242 }
5243
5244 if (allow_exec) {
5245 prot |= PROT_EXEC;
5246 }
5247
5248 if (addr != NULL) {
5249 flags |= MAP_FIXED;
5250 }
5251
5252 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5253 fd, file_offset);
5254 if (mapped_address == MAP_FAILED) {
5255 return NULL;
5256 }
5257 return mapped_address;
5258 }
5259
5260
5261 // Remap a block of memory.
5262 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5263 char *addr, size_t bytes, bool read_only,
5264 bool allow_exec) {
5265 // same as map_memory() on this OS
5266 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5267 allow_exec);
5268 }
5269
5270
5271 // Unmap a block of memory.
5272 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5273 return munmap(addr, bytes) == 0;
5274 }
5275
5276 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5277
5278 static clockid_t thread_cpu_clockid(Thread* thread) {
5279 pthread_t tid = thread->osthread()->pthread_id();
5280 clockid_t clockid;
5281
5282 // Get thread clockid
5283 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5284 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5285 return clockid;
5286 }
5287
5288 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5289 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5290 // of a thread.
5291 //
5292 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5293 // the fast estimate available on the platform.
5294
5295 jlong os::current_thread_cpu_time() {
5296 if (os::Linux::supports_fast_thread_cpu_time()) {
5297 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5298 } else {
5299 // return user + sys since the cost is the same
5300 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5301 }
5302 }
5303
5304 jlong os::thread_cpu_time(Thread* thread) {
5305 // consistent with what current_thread_cpu_time() returns
5306 if (os::Linux::supports_fast_thread_cpu_time()) {
5307 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5308 } else {
5309 return slow_thread_cpu_time(thread, true /* user + sys */);
5310 }
5311 }
5312
5313 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5314 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5315 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5316 } else {
5317 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5318 }
5319 }
5320
5321 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5322 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5323 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5324 } else {
5325 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5326 }
5327 }
5328
5329 // -1 on error.
5330 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5331 pid_t tid = thread->osthread()->thread_id();
5332 char *s;
5333 char stat[2048];
5334 int statlen;
5335 char proc_name[64];
5336 int count;
5337 long sys_time, user_time;
5338 char cdummy;
5339 int idummy;
5340 long ldummy;
5341 FILE *fp;
5342
5343 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5344 fp = fopen(proc_name, "r");
5345 if (fp == NULL) return -1;
5346 statlen = fread(stat, 1, 2047, fp);
5347 stat[statlen] = '\0';
5348 fclose(fp);
5349
5350 // Skip pid and the command string. Note that we could be dealing with
5351 // weird command names, e.g. user could decide to rename java launcher
5352 // to "java 1.4.2 :)", then the stat file would look like
5353 // 1234 (java 1.4.2 :)) R ... ...
5354 // We don't really need to know the command string, just find the last
5355 // occurrence of ")" and then start parsing from there. See bug 4726580.
5356 s = strrchr(stat, ')');
5357 if (s == NULL) return -1;
5358
5359 // Skip blank chars
5360 do s++; while (isspace(*s));
5361
5362 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5363 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5364 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5365 &user_time, &sys_time);
5366 if (count != 13) return -1;
5367 if (user_sys_cpu_time) {
5368 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5369 } else {
5370 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5371 }
5372 }
5373
5374 void os::current_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 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5382 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5383 info_ptr->may_skip_backward = false; // elapsed time not wall time
5384 info_ptr->may_skip_forward = false; // elapsed time not wall time
5385 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5386 }
5387
5388 bool os::is_thread_cpu_time_supported() {
5389 return true;
5390 }
5391
5392 // System loadavg support. Returns -1 if load average cannot be obtained.
5393 // Linux doesn't yet have a (official) notion of processor sets,
5394 // so just return the system wide load average.
5395 int os::loadavg(double loadavg[], int nelem) {
5396 return ::getloadavg(loadavg, nelem);
5397 }
5398
5399 void os::pause() {
5400 char filename[MAX_PATH];
5401 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5402 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5403 } else {
5404 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5405 }
5406
5407 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5408 if (fd != -1) {
5409 struct stat buf;
5410 ::close(fd);
5411 while (::stat(filename, &buf) == 0) {
5412 (void)::poll(NULL, 0, 100);
5413 }
5414 } else {
5415 jio_fprintf(stderr,
5416 "Could not open pause file '%s', continuing immediately.\n", filename);
5417 }
5418 }
5419
5420
5421 // Refer to the comments in os_solaris.cpp park-unpark.
5422 //
5423 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5424 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5425 // For specifics regarding the bug see GLIBC BUGID 261237 :
5426 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5427 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5428 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5429 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5430 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5431 // and monitorenter when we're using 1-0 locking. All those operations may result in
5432 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5433 // of libpthread avoids the problem, but isn't practical.
5434 //
5435 // Possible remedies:
5436 //
5437 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5438 // This is palliative and probabilistic, however. If the thread is preempted
5439 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5440 // than the minimum period may have passed, and the abstime may be stale (in the
5441 // past) resultin in a hang. Using this technique reduces the odds of a hang
5442 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5443 //
5444 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5445 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5446 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5447 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5448 // thread.
5449 //
5450 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5451 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5452 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5453 // This also works well. In fact it avoids kernel-level scalability impediments
5454 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5455 // timers in a graceful fashion.
5456 //
5457 // 4. When the abstime value is in the past it appears that control returns
5458 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5459 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5460 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5461 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5462 // It may be possible to avoid reinitialization by checking the return
5463 // value from pthread_cond_timedwait(). In addition to reinitializing the
5464 // condvar we must establish the invariant that cond_signal() is only called
5465 // within critical sections protected by the adjunct mutex. This prevents
5466 // cond_signal() from "seeing" a condvar that's in the midst of being
5467 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5468 // desirable signal-after-unlock optimization that avoids futile context switching.
5469 //
5470 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5471 // structure when a condvar is used or initialized. cond_destroy() would
5472 // release the helper structure. Our reinitialize-after-timedwait fix
5473 // put excessive stress on malloc/free and locks protecting the c-heap.
5474 //
5475 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5476 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5477 // and only enabling the work-around for vulnerable environments.
5478
5479 // utility to compute the abstime argument to timedwait:
5480 // millis is the relative timeout time
5481 // abstime will be the absolute timeout time
5482 // TODO: replace compute_abstime() with unpackTime()
5483
5484 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5485 if (millis < 0) millis = 0;
5486
5487 jlong seconds = millis / 1000;
5488 millis %= 1000;
5489 if (seconds > 50000000) { // see man cond_timedwait(3T)
5490 seconds = 50000000;
5491 }
5492
5493 if (os::supports_monotonic_clock()) {
5494 struct timespec now;
5495 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5496 assert_status(status == 0, status, "clock_gettime");
5497 abstime->tv_sec = now.tv_sec + seconds;
5498 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5499 if (nanos >= NANOSECS_PER_SEC) {
5500 abstime->tv_sec += 1;
5501 nanos -= NANOSECS_PER_SEC;
5502 }
5503 abstime->tv_nsec = nanos;
5504 } else {
5505 struct timeval now;
5506 int status = gettimeofday(&now, NULL);
5507 assert(status == 0, "gettimeofday");
5508 abstime->tv_sec = now.tv_sec + seconds;
5509 long usec = now.tv_usec + millis * 1000;
5510 if (usec >= 1000000) {
5511 abstime->tv_sec += 1;
5512 usec -= 1000000;
5513 }
5514 abstime->tv_nsec = usec * 1000;
5515 }
5516 return abstime;
5517 }
5518
5519 void os::PlatformEvent::park() { // AKA "down()"
5520 // Invariant: Only the thread associated with the Event/PlatformEvent
5521 // may call park().
5522 // TODO: assert that _Assoc != NULL or _Assoc == Self
5523 assert(_nParked == 0, "invariant");
5524
5525 int v;
5526 for (;;) {
5527 v = _Event;
5528 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5529 }
5530 guarantee(v >= 0, "invariant");
5531 if (v == 0) {
5532 // Do this the hard way by blocking ...
5533 int status = pthread_mutex_lock(_mutex);
5534 assert_status(status == 0, status, "mutex_lock");
5535 guarantee(_nParked == 0, "invariant");
5536 ++_nParked;
5537 while (_Event < 0) {
5538 status = pthread_cond_wait(_cond, _mutex);
5539 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5540 // Treat this the same as if the wait was interrupted
5541 if (status == ETIME) { status = EINTR; }
5542 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5543 }
5544 --_nParked;
5545
5546 _Event = 0;
5547 status = pthread_mutex_unlock(_mutex);
5548 assert_status(status == 0, status, "mutex_unlock");
5549 // Paranoia to ensure our locked and lock-free paths interact
5550 // correctly with each other.
5551 OrderAccess::fence();
5552 }
5553 guarantee(_Event >= 0, "invariant");
5554 }
5555
5556 int os::PlatformEvent::park(jlong millis) {
5557 guarantee(_nParked == 0, "invariant");
5558
5559 int v;
5560 for (;;) {
5561 v = _Event;
5562 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5563 }
5564 guarantee(v >= 0, "invariant");
5565 if (v != 0) return OS_OK;
5566
5567 // We do this the hard way, by blocking the thread.
5568 // Consider enforcing a minimum timeout value.
5569 struct timespec abst;
5570 compute_abstime(&abst, millis);
5571
5572 int ret = OS_TIMEOUT;
5573 int status = pthread_mutex_lock(_mutex);
5574 assert_status(status == 0, status, "mutex_lock");
5575 guarantee(_nParked == 0, "invariant");
5576 ++_nParked;
5577
5578 // Object.wait(timo) will return because of
5579 // (a) notification
5580 // (b) timeout
5581 // (c) thread.interrupt
5582 //
5583 // Thread.interrupt and object.notify{All} both call Event::set.
5584 // That is, we treat thread.interrupt as a special case of notification.
5585 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5586 // We assume all ETIME returns are valid.
5587 //
5588 // TODO: properly differentiate simultaneous notify+interrupt.
5589 // In that case, we should propagate the notify to another waiter.
5590
5591 while (_Event < 0) {
5592 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5593 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5594 pthread_cond_destroy(_cond);
5595 pthread_cond_init(_cond, os::Linux::condAttr());
5596 }
5597 assert_status(status == 0 || status == EINTR ||
5598 status == ETIME || status == ETIMEDOUT,
5599 status, "cond_timedwait");
5600 if (!FilterSpuriousWakeups) break; // previous semantics
5601 if (status == ETIME || status == ETIMEDOUT) break;
5602 // We consume and ignore EINTR and spurious wakeups.
5603 }
5604 --_nParked;
5605 if (_Event >= 0) {
5606 ret = OS_OK;
5607 }
5608 _Event = 0;
5609 status = pthread_mutex_unlock(_mutex);
5610 assert_status(status == 0, status, "mutex_unlock");
5611 assert(_nParked == 0, "invariant");
5612 // Paranoia to ensure our locked and lock-free paths interact
5613 // correctly with each other.
5614 OrderAccess::fence();
5615 return ret;
5616 }
5617
5618 void os::PlatformEvent::unpark() {
5619 // Transitions for _Event:
5620 // 0 :=> 1
5621 // 1 :=> 1
5622 // -1 :=> either 0 or 1; must signal target thread
5623 // That is, we can safely transition _Event from -1 to either
5624 // 0 or 1.
5625 // See also: "Semaphores in Plan 9" by Mullender & Cox
5626 //
5627 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5628 // that it will take two back-to-back park() calls for the owning
5629 // thread to block. This has the benefit of forcing a spurious return
5630 // from the first park() call after an unpark() call which will help
5631 // shake out uses of park() and unpark() without condition variables.
5632
5633 if (Atomic::xchg(1, &_Event) >= 0) return;
5634
5635 // Wait for the thread associated with the event to vacate
5636 int status = pthread_mutex_lock(_mutex);
5637 assert_status(status == 0, status, "mutex_lock");
5638 int AnyWaiters = _nParked;
5639 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5640 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5641 AnyWaiters = 0;
5642 pthread_cond_signal(_cond);
5643 }
5644 status = pthread_mutex_unlock(_mutex);
5645 assert_status(status == 0, status, "mutex_unlock");
5646 if (AnyWaiters != 0) {
5647 status = pthread_cond_signal(_cond);
5648 assert_status(status == 0, status, "cond_signal");
5649 }
5650
5651 // Note that we signal() _after dropping the lock for "immortal" Events.
5652 // This is safe and avoids a common class of futile wakeups. In rare
5653 // circumstances this can cause a thread to return prematurely from
5654 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5655 // simply re-test the condition and re-park itself.
5656 }
5657
5658
5659 // JSR166
5660 // -------------------------------------------------------
5661
5662 // The solaris and linux implementations of park/unpark are fairly
5663 // conservative for now, but can be improved. They currently use a
5664 // mutex/condvar pair, plus a a count.
5665 // Park decrements count if > 0, else does a condvar wait. Unpark
5666 // sets count to 1 and signals condvar. Only one thread ever waits
5667 // on the condvar. Contention seen when trying to park implies that someone
5668 // is unparking you, so don't wait. And spurious returns are fine, so there
5669 // is no need to track notifications.
5670
5671 // This code is common to linux and solaris and will be moved to a
5672 // common place in dolphin.
5673 //
5674 // The passed in time value is either a relative time in nanoseconds
5675 // or an absolute time in milliseconds. Either way it has to be unpacked
5676 // into suitable seconds and nanoseconds components and stored in the
5677 // given timespec structure.
5678 // Given time is a 64-bit value and the time_t used in the timespec is only
5679 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5680 // overflow if times way in the future are given. Further on Solaris versions
5681 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5682 // number of seconds, in abstime, is less than current_time + 100,000,000.
5683 // As it will be 28 years before "now + 100000000" will overflow we can
5684 // ignore overflow and just impose a hard-limit on seconds using the value
5685 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5686 // years from "now".
5687
5688 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5689 assert(time > 0, "convertTime");
5690 time_t max_secs = 0;
5691
5692 if (!os::supports_monotonic_clock() || isAbsolute) {
5693 struct timeval now;
5694 int status = gettimeofday(&now, NULL);
5695 assert(status == 0, "gettimeofday");
5696
5697 max_secs = now.tv_sec + MAX_SECS;
5698
5699 if (isAbsolute) {
5700 jlong secs = time / 1000;
5701 if (secs > max_secs) {
5702 absTime->tv_sec = max_secs;
5703 } else {
5704 absTime->tv_sec = secs;
5705 }
5706 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5707 } else {
5708 jlong secs = time / NANOSECS_PER_SEC;
5709 if (secs >= MAX_SECS) {
5710 absTime->tv_sec = max_secs;
5711 absTime->tv_nsec = 0;
5712 } else {
5713 absTime->tv_sec = now.tv_sec + secs;
5714 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5715 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5716 absTime->tv_nsec -= NANOSECS_PER_SEC;
5717 ++absTime->tv_sec; // note: this must be <= max_secs
5718 }
5719 }
5720 }
5721 } else {
5722 // must be relative using monotonic clock
5723 struct timespec now;
5724 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5725 assert_status(status == 0, status, "clock_gettime");
5726 max_secs = now.tv_sec + MAX_SECS;
5727 jlong secs = time / NANOSECS_PER_SEC;
5728 if (secs >= MAX_SECS) {
5729 absTime->tv_sec = max_secs;
5730 absTime->tv_nsec = 0;
5731 } else {
5732 absTime->tv_sec = now.tv_sec + secs;
5733 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5734 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5735 absTime->tv_nsec -= NANOSECS_PER_SEC;
5736 ++absTime->tv_sec; // note: this must be <= max_secs
5737 }
5738 }
5739 }
5740 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5741 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5742 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5743 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5744 }
5745
5746 void Parker::park(bool isAbsolute, jlong time) {
5747 // Ideally we'd do something useful while spinning, such
5748 // as calling unpackTime().
5749
5750 // Optional fast-path check:
5751 // Return immediately if a permit is available.
5752 // We depend on Atomic::xchg() having full barrier semantics
5753 // since we are doing a lock-free update to _counter.
5754 if (Atomic::xchg(0, &_counter) > 0) return;
5755
5756 Thread* thread = Thread::current();
5757 assert(thread->is_Java_thread(), "Must be JavaThread");
5758 JavaThread *jt = (JavaThread *)thread;
5759
5760 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5761 // Check interrupt before trying to wait
5762 if (Thread::is_interrupted(thread, false)) {
5763 return;
5764 }
5765
5766 // Next, demultiplex/decode time arguments
5767 timespec absTime;
5768 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5769 return;
5770 }
5771 if (time > 0) {
5772 unpackTime(&absTime, isAbsolute, time);
5773 }
5774
5775
5776 // Enter safepoint region
5777 // Beware of deadlocks such as 6317397.
5778 // The per-thread Parker:: mutex is a classic leaf-lock.
5779 // In particular a thread must never block on the Threads_lock while
5780 // holding the Parker:: mutex. If safepoints are pending both the
5781 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5782 ThreadBlockInVM tbivm(jt);
5783
5784 // Don't wait if cannot get lock since interference arises from
5785 // unblocking. Also. check interrupt before trying wait
5786 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5787 return;
5788 }
5789
5790 int status;
5791 if (_counter > 0) { // no wait needed
5792 _counter = 0;
5793 status = pthread_mutex_unlock(_mutex);
5794 assert(status == 0, "invariant");
5795 // Paranoia to ensure our locked and lock-free paths interact
5796 // correctly with each other and Java-level accesses.
5797 OrderAccess::fence();
5798 return;
5799 }
5800
5801 #ifdef ASSERT
5802 // Don't catch signals while blocked; let the running threads have the signals.
5803 // (This allows a debugger to break into the running thread.)
5804 sigset_t oldsigs;
5805 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5806 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5807 #endif
5808
5809 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5810 jt->set_suspend_equivalent();
5811 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5812
5813 assert(_cur_index == -1, "invariant");
5814 if (time == 0) {
5815 _cur_index = REL_INDEX; // arbitrary choice when not timed
5816 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5817 } else {
5818 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5819 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5820 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5821 pthread_cond_destroy(&_cond[_cur_index]);
5822 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5823 }
5824 }
5825 _cur_index = -1;
5826 assert_status(status == 0 || status == EINTR ||
5827 status == ETIME || status == ETIMEDOUT,
5828 status, "cond_timedwait");
5829
5830 #ifdef ASSERT
5831 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5832 #endif
5833
5834 _counter = 0;
5835 status = pthread_mutex_unlock(_mutex);
5836 assert_status(status == 0, status, "invariant");
5837 // Paranoia to ensure our locked and lock-free paths interact
5838 // correctly with each other and Java-level accesses.
5839 OrderAccess::fence();
5840
5841 // If externally suspended while waiting, re-suspend
5842 if (jt->handle_special_suspend_equivalent_condition()) {
5843 jt->java_suspend_self();
5844 }
5845 }
5846
5847 void Parker::unpark() {
5848 int status = pthread_mutex_lock(_mutex);
5849 assert(status == 0, "invariant");
5850 const int s = _counter;
5851 _counter = 1;
5852 if (s < 1) {
5853 // thread might be parked
5854 if (_cur_index != -1) {
5855 // thread is definitely parked
5856 if (WorkAroundNPTLTimedWaitHang) {
5857 status = pthread_cond_signal(&_cond[_cur_index]);
5858 assert(status == 0, "invariant");
5859 status = pthread_mutex_unlock(_mutex);
5860 assert(status == 0, "invariant");
5861 } else {
5862 status = pthread_mutex_unlock(_mutex);
5863 assert(status == 0, "invariant");
5864 status = pthread_cond_signal(&_cond[_cur_index]);
5865 assert(status == 0, "invariant");
5866 }
5867 } else {
5868 pthread_mutex_unlock(_mutex);
5869 assert(status == 0, "invariant");
5870 }
5871 } else {
5872 pthread_mutex_unlock(_mutex);
5873 assert(status == 0, "invariant");
5874 }
5875 }
5876
5877
5878 extern char** environ;
5879
5880 #ifndef __NR_fork
5881 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079)
5882 #endif
5883
5884 #ifndef __NR_execve
5885 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221)
5886 #endif
5887
5888 // Run the specified command in a separate process. Return its exit value,
5889 // or -1 on failure (e.g. can't fork a new process).
5890 // Unlike system(), this function can be called from signal handler. It
5891 // doesn't block SIGINT et al.
5892 int os::fork_and_exec(char* cmd) {
5893 const char * argv[4] = {"sh", "-c", cmd, NULL};
5894
5895 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5896 // pthread_atfork handlers and reset pthread library. All we need is a
5897 // separate process to execve. Make a direct syscall to fork process.
5898 // On IA64 there's no fork syscall, we have to use fork() and hope for
5899 // the best...
5900 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5901 IA64_ONLY(fork();)
5902
5903 if (pid < 0) {
5904 // fork failed
5905 return -1;
5906
5907 } else if (pid == 0) {
5908 // child process
5909
5910 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5911 // first to kill every thread on the thread list. Because this list is
5912 // not reset by fork() (see notes above), execve() will instead kill
5913 // every thread in the parent process. We know this is the only thread
5914 // in the new process, so make a system call directly.
5915 // IA64 should use normal execve() from glibc to match the glibc fork()
5916 // above.
5917 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5918 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5919
5920 // execve failed
5921 _exit(-1);
5922
5923 } else {
5924 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5925 // care about the actual exit code, for now.
5926
5927 int status;
5928
5929 // Wait for the child process to exit. This returns immediately if
5930 // the child has already exited. */
5931 while (waitpid(pid, &status, 0) < 0) {
5932 switch (errno) {
5933 case ECHILD: return 0;
5934 case EINTR: break;
5935 default: return -1;
5936 }
5937 }
5938
5939 if (WIFEXITED(status)) {
5940 // The child exited normally; get its exit code.
5941 return WEXITSTATUS(status);
5942 } else if (WIFSIGNALED(status)) {
5943 // The child exited because of a signal
5944 // The best value to return is 0x80 + signal number,
5945 // because that is what all Unix shells do, and because
5946 // it allows callers to distinguish between process exit and
5947 // process death by signal.
5948 return 0x80 + WTERMSIG(status);
5949 } else {
5950 // Unknown exit code; pass it through
5951 return status;
5952 }
5953 }
5954 }
5955
5956 // is_headless_jre()
5957 //
5958 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5959 // in order to report if we are running in a headless jre
5960 //
5961 // Since JDK8 xawt/libmawt.so was moved into the same directory
5962 // as libawt.so, and renamed libawt_xawt.so
5963 //
5964 bool os::is_headless_jre() {
5965 struct stat statbuf;
5966 char buf[MAXPATHLEN];
5967 char libmawtpath[MAXPATHLEN];
5968 const char *xawtstr = "/xawt/libmawt.so";
5969 const char *new_xawtstr = "/libawt_xawt.so";
5970 char *p;
5971
5972 // Get path to libjvm.so
5973 os::jvm_path(buf, sizeof(buf));
5974
5975 // Get rid of libjvm.so
5976 p = strrchr(buf, '/');
5977 if (p == NULL) {
5978 return false;
5979 } else {
5980 *p = '\0';
5981 }
5982
5983 // Get rid of client or server
5984 p = strrchr(buf, '/');
5985 if (p == NULL) {
5986 return false;
5987 } else {
5988 *p = '\0';
5989 }
5990
5991 // check xawt/libmawt.so
5992 strcpy(libmawtpath, buf);
5993 strcat(libmawtpath, xawtstr);
5994 if (::stat(libmawtpath, &statbuf) == 0) return false;
5995
5996 // check libawt_xawt.so
5997 strcpy(libmawtpath, buf);
5998 strcat(libmawtpath, new_xawtstr);
5999 if (::stat(libmawtpath, &statbuf) == 0) return false;
6000
6001 return true;
6002 }
6003
6004 // Get the default path to the core file
6005 // Returns the length of the string
6006 int os::get_core_path(char* buffer, size_t bufferSize) {
6007 const char* p = get_current_directory(buffer, bufferSize);
6008
6009 if (p == NULL) {
6010 assert(p != NULL, "failed to get current directory");
6011 return 0;
6012 }
6013
6014 return strlen(buffer);
6015 }
6016
6017 #ifdef JAVASE_EMBEDDED
6018 //
6019 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6020 //
6021 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6022
6023 // ctor
6024 //
6025 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6026 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6027 _fd = fd;
6028
6029 if (os::create_thread(this, os::os_thread)) {
6030 _memnotify_thread = this;
6031 os::set_priority(this, NearMaxPriority);
6032 os::start_thread(this);
6033 }
6034 }
6035
6036 // Where all the work gets done
6037 //
6038 void MemNotifyThread::run() {
6039 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6040
6041 // Set up the select arguments
6042 fd_set rfds;
6043 if (_fd != -1) {
6044 FD_ZERO(&rfds);
6045 FD_SET(_fd, &rfds);
6046 }
6047
6048 // Now wait for the mem_notify device to wake up
6049 while (1) {
6050 // Wait for the mem_notify device to signal us..
6051 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6052 if (rc == -1) {
6053 perror("select!\n");
6054 break;
6055 } else if (rc) {
6056 //ssize_t free_before = os::available_memory();
6057 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6058
6059 // The kernel is telling us there is not much memory left...
6060 // try to do something about that
6061
6062 // If we are not already in a GC, try one.
6063 if (!Universe::heap()->is_gc_active()) {
6064 Universe::heap()->collect(GCCause::_allocation_failure);
6065
6066 //ssize_t free_after = os::available_memory();
6067 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6068 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6069 }
6070 // We might want to do something like the following if we find the GC's are not helping...
6071 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6072 }
6073 }
6074 }
6075
6076 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6077 //
6078 void MemNotifyThread::start() {
6079 int fd;
6080 fd = open("/dev/mem_notify", O_RDONLY, 0);
6081 if (fd < 0) {
6082 return;
6083 }
6084
6085 if (memnotify_thread() == NULL) {
6086 new MemNotifyThread(fd);
6087 }
6088 }
6089
6090 #endif // JAVASE_EMBEDDED
6091
6092
6093 /////////////// Unit tests ///////////////
6094
6095 #ifndef PRODUCT
6096
6097 #define test_log(...) \
6098 do { \
6099 if (VerboseInternalVMTests) { \
6100 tty->print_cr(__VA_ARGS__); \
6101 tty->flush(); \
6102 } \
6103 } while (false)
6104
6105 class TestReserveMemorySpecial : AllStatic {
6106 public:
6107 static void small_page_write(void* addr, size_t size) {
6108 size_t page_size = os::vm_page_size();
6109
6110 char* end = (char*)addr + size;
6111 for (char* p = (char*)addr; p < end; p += page_size) {
6112 *p = 1;
6113 }
6114 }
6115
6116 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6117 if (!UseHugeTLBFS) {
6118 return;
6119 }
6120
6121 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6122
6123 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6124
6125 if (addr != NULL) {
6126 small_page_write(addr, size);
6127
6128 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6129 }
6130 }
6131
6132 static void test_reserve_memory_special_huge_tlbfs_only() {
6133 if (!UseHugeTLBFS) {
6134 return;
6135 }
6136
6137 size_t lp = os::large_page_size();
6138
6139 for (size_t size = lp; size <= lp * 10; size += lp) {
6140 test_reserve_memory_special_huge_tlbfs_only(size);
6141 }
6142 }
6143
6144 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6145 if (!UseHugeTLBFS) {
6146 return;
6147 }
6148
6149 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6150 size, alignment);
6151
6152 assert(size >= os::large_page_size(), "Incorrect input to test");
6153
6154 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6155
6156 if (addr != NULL) {
6157 small_page_write(addr, size);
6158
6159 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6160 }
6161 }
6162
6163 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6164 size_t lp = os::large_page_size();
6165 size_t ag = os::vm_allocation_granularity();
6166
6167 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6168 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6169 }
6170 }
6171
6172 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6173 size_t lp = os::large_page_size();
6174 size_t ag = os::vm_allocation_granularity();
6175
6176 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6177 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6178 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6179 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6180 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6181 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6182 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6183 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6184 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6185 }
6186
6187 static void test_reserve_memory_special_huge_tlbfs() {
6188 if (!UseHugeTLBFS) {
6189 return;
6190 }
6191
6192 test_reserve_memory_special_huge_tlbfs_only();
6193 test_reserve_memory_special_huge_tlbfs_mixed();
6194 }
6195
6196 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6197 if (!UseSHM) {
6198 return;
6199 }
6200
6201 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6202
6203 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6204
6205 if (addr != NULL) {
6206 assert(is_ptr_aligned(addr, alignment), "Check");
6207 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6208
6209 small_page_write(addr, size);
6210
6211 os::Linux::release_memory_special_shm(addr, size);
6212 }
6213 }
6214
6215 static void test_reserve_memory_special_shm() {
6216 size_t lp = os::large_page_size();
6217 size_t ag = os::vm_allocation_granularity();
6218
6219 for (size_t size = ag; size < lp * 3; size += ag) {
6220 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6221 test_reserve_memory_special_shm(size, alignment);
6222 }
6223 }
6224 }
6225
6226 static void test() {
6227 test_reserve_memory_special_huge_tlbfs();
6228 test_reserve_memory_special_shm();
6229 }
6230 };
6231
6232 void TestReserveMemorySpecial_test() {
6233 TestReserveMemorySpecial::test();
6234 }
6235
6236 #endif