1 /*
2 * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_share_linux.hpp"
40 #include "prims/jniFastGetField.hpp"
41 #include "prims/jvm.h"
42 #include "prims/jvm_misc.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/extendedPC.hpp"
45 #include "runtime/globals.hpp"
46 #include "runtime/interfaceSupport.hpp"
47 #include "runtime/init.hpp"
48 #include "runtime/java.hpp"
49 #include "runtime/javaCalls.hpp"
50 #include "runtime/mutexLocker.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/osThread.hpp"
53 #include "runtime/perfMemory.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/statSampler.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "runtime/thread.inline.hpp"
58 #include "runtime/threadCritical.hpp"
59 #include "runtime/timer.hpp"
60 #include "services/attachListener.hpp"
61 #include "services/memTracker.hpp"
62 #include "services/runtimeService.hpp"
63 #include "utilities/decoder.hpp"
64 #include "utilities/defaultStream.hpp"
65 #include "utilities/events.hpp"
66 #include "utilities/elfFile.hpp"
67 #include "utilities/growableArray.hpp"
68 #include "utilities/vmError.hpp"
69
70 // put OS-includes here
71 # include <sys/types.h>
72 # include <sys/mman.h>
73 # include <sys/stat.h>
74 # include <sys/select.h>
75 # include <pthread.h>
76 # include <signal.h>
77 # include <errno.h>
78 # include <dlfcn.h>
79 # include <stdio.h>
80 # include <unistd.h>
81 # include <sys/resource.h>
82 # include <pthread.h>
83 # include <sys/stat.h>
84 # include <sys/time.h>
85 # include <sys/times.h>
86 # include <sys/utsname.h>
87 # include <sys/socket.h>
88 # include <sys/wait.h>
89 # include <pwd.h>
90 # include <poll.h>
91 # include <semaphore.h>
92 # include <fcntl.h>
93 # include <string.h>
94 # include <syscall.h>
95 # include <sys/sysinfo.h>
96 # include <gnu/libc-version.h>
97 # include <sys/ipc.h>
98 # include <sys/shm.h>
99 # include <link.h>
100 # include <stdint.h>
101 # include <inttypes.h>
102 # include <sys/ioctl.h>
103
104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
105 // getrusage() is prepared to handle the associated failure.
106 #ifndef RUSAGE_THREAD
107 #define RUSAGE_THREAD (1) /* only the calling thread */
108 #endif
109
110 #define MAX_PATH (2 * K)
111
112 // for timer info max values which include all bits
113 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
114
115 #define LARGEPAGES_BIT (1 << 6)
116 ////////////////////////////////////////////////////////////////////////////////
117 // global variables
118 julong os::Linux::_physical_memory = 0;
119
120 address os::Linux::_initial_thread_stack_bottom = NULL;
121 uintptr_t os::Linux::_initial_thread_stack_size = 0;
122
123 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
124 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
125 Mutex* os::Linux::_createThread_lock = NULL;
126 pthread_t os::Linux::_main_thread;
127 int os::Linux::_page_size = -1;
128 const int os::Linux::_vm_default_page_size = (8 * K);
129 bool os::Linux::_is_floating_stack = false;
130 bool os::Linux::_is_NPTL = false;
131 bool os::Linux::_supports_fast_thread_cpu_time = false;
132 const char * os::Linux::_glibc_version = NULL;
133 const char * os::Linux::_libpthread_version = NULL;
134 pthread_condattr_t os::Linux::_condattr[1];
135
136 static jlong initial_time_count=0;
137
138 static int clock_tics_per_sec = 100;
139
140 // For diagnostics to print a message once. see run_periodic_checks
141 static sigset_t check_signal_done;
142 static bool check_signals = true;;
143
144 static pid_t _initial_pid = 0;
145
146 /* Signal number used to suspend/resume a thread */
147
148 /* do not use any signal number less than SIGSEGV, see 4355769 */
149 static int SR_signum = SIGUSR2;
150 sigset_t SR_sigset;
151
152 /* Used to protect dlsym() calls */
153 static pthread_mutex_t dl_mutex;
154
155 // Declarations
156 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
157
158 #ifdef JAVASE_EMBEDDED
159 class MemNotifyThread: public Thread {
160 friend class VMStructs;
161 public:
162 virtual void run();
163
164 private:
165 static MemNotifyThread* _memnotify_thread;
166 int _fd;
167
168 public:
169
170 // Constructor
171 MemNotifyThread(int fd);
172
173 // Tester
174 bool is_memnotify_thread() const { return true; }
175
176 // Printing
177 char* name() const { return (char*)"Linux MemNotify Thread"; }
178
179 // Returns the single instance of the MemNotifyThread
180 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
181
182 // Create and start the single instance of MemNotifyThread
183 static void start();
184 };
185 #endif // JAVASE_EMBEDDED
186
187 // utility functions
188
189 static int SR_initialize();
190
191 julong os::available_memory() {
192 return Linux::available_memory();
193 }
194
195 julong os::Linux::available_memory() {
196 // values in struct sysinfo are "unsigned long"
197 struct sysinfo si;
198 sysinfo(&si);
199
200 return (julong)si.freeram * si.mem_unit;
201 }
202
203 julong os::physical_memory() {
204 return Linux::physical_memory();
205 }
206
207 ////////////////////////////////////////////////////////////////////////////////
208 // environment support
209
210 bool os::getenv(const char* name, char* buf, int len) {
211 const char* val = ::getenv(name);
212 if (val != NULL && strlen(val) < (size_t)len) {
213 strcpy(buf, val);
214 return true;
215 }
216 if (len > 0) buf[0] = 0; // return a null string
217 return false;
218 }
219
220
221 // Return true if user is running as root.
222
223 bool os::have_special_privileges() {
224 static bool init = false;
225 static bool privileges = false;
226 if (!init) {
227 privileges = (getuid() != geteuid()) || (getgid() != getegid());
228 init = true;
229 }
230 return privileges;
231 }
232
233
234 #ifndef SYS_gettid
235 // i386: 224, ia64: 1105, amd64: 186, sparc 143
236 #ifdef __ia64__
237 #define SYS_gettid 1105
238 #elif __i386__
239 #define SYS_gettid 224
240 #elif __amd64__
241 #define SYS_gettid 186
242 #elif __sparc__
243 #define SYS_gettid 143
244 #else
245 #error define gettid for the arch
246 #endif
247 #endif
248
249 // Cpu architecture string
250 #if defined(ZERO)
251 static char cpu_arch[] = ZERO_LIBARCH;
252 #elif defined(IA64)
253 static char cpu_arch[] = "ia64";
254 #elif defined(IA32)
255 static char cpu_arch[] = "i386";
256 #elif defined(AMD64)
257 static char cpu_arch[] = "amd64";
258 #elif defined(ARM)
259 static char cpu_arch[] = "arm";
260 #elif defined(PPC)
261 static char cpu_arch[] = "ppc";
262 #elif defined(SPARC)
263 # ifdef _LP64
264 static char cpu_arch[] = "sparcv9";
265 # else
266 static char cpu_arch[] = "sparc";
267 # endif
268 #else
269 #error Add appropriate cpu_arch setting
270 #endif
271
272
273 // pid_t gettid()
274 //
275 // Returns the kernel thread id of the currently running thread. Kernel
276 // thread id is used to access /proc.
277 //
278 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
279 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
280 //
281 pid_t os::Linux::gettid() {
282 int rslt = syscall(SYS_gettid);
283 if (rslt == -1) {
284 // old kernel, no NPTL support
285 return getpid();
286 } else {
287 return (pid_t)rslt;
288 }
289 }
290
291 // Most versions of linux have a bug where the number of processors are
292 // determined by looking at the /proc file system. In a chroot environment,
293 // the system call returns 1. This causes the VM to act as if it is
294 // a single processor and elide locking (see is_MP() call).
295 static bool unsafe_chroot_detected = false;
296 static const char *unstable_chroot_error = "/proc file system not found.\n"
297 "Java may be unstable running multithreaded in a chroot "
298 "environment on Linux when /proc filesystem is not mounted.";
299
300 void os::Linux::initialize_system_info() {
301 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
302 if (processor_count() == 1) {
303 pid_t pid = os::Linux::gettid();
304 char fname[32];
305 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
306 FILE *fp = fopen(fname, "r");
307 if (fp == NULL) {
308 unsafe_chroot_detected = true;
309 } else {
310 fclose(fp);
311 }
312 }
313 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
314 assert(processor_count() > 0, "linux error");
315 }
316
317 void os::init_system_properties_values() {
318 // char arch[12];
319 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
320
321 // The next steps are taken in the product version:
322 //
323 // Obtain the JAVA_HOME value from the location of libjvm.so.
324 // This library should be located at:
325 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
326 //
327 // If "/jre/lib/" appears at the right place in the path, then we
328 // assume libjvm.so is installed in a JDK and we use this path.
329 //
330 // Otherwise exit with message: "Could not create the Java virtual machine."
331 //
332 // The following extra steps are taken in the debugging version:
333 //
334 // If "/jre/lib/" does NOT appear at the right place in the path
335 // instead of exit check for $JAVA_HOME environment variable.
336 //
337 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
338 // then we append a fake suffix "hotspot/libjvm.so" to this path so
339 // it looks like libjvm.so is installed there
340 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
341 //
342 // Otherwise exit.
343 //
344 // Important note: if the location of libjvm.so changes this
345 // code needs to be changed accordingly.
346
347 // The next few definitions allow the code to be verbatim:
348 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
349 #define getenv(n) ::getenv(n)
350
351 /*
352 * See ld(1):
353 * The linker uses the following search paths to locate required
354 * shared libraries:
355 * 1: ...
356 * ...
357 * 7: The default directories, normally /lib and /usr/lib.
358 */
359 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
360 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
361 #else
362 #define DEFAULT_LIBPATH "/lib:/usr/lib"
363 #endif
364
365 #define EXTENSIONS_DIR "/lib/ext"
366 #define ENDORSED_DIR "/lib/endorsed"
367 #define REG_DIR "/usr/java/packages"
368
369 {
370 /* sysclasspath, java_home, dll_dir */
371 {
372 char *home_path;
373 char *dll_path;
374 char *pslash;
375 char buf[MAXPATHLEN];
376 os::jvm_path(buf, sizeof(buf));
377
378 // Found the full path to libjvm.so.
379 // Now cut the path to <java_home>/jre if we can.
380 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
381 pslash = strrchr(buf, '/');
382 if (pslash != NULL)
383 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
384 dll_path = malloc(strlen(buf) + 1);
385 if (dll_path == NULL)
386 return;
387 strcpy(dll_path, buf);
388 Arguments::set_dll_dir(dll_path);
389
390 if (pslash != NULL) {
391 pslash = strrchr(buf, '/');
392 if (pslash != NULL) {
393 *pslash = '\0'; /* get rid of /<arch> */
394 pslash = strrchr(buf, '/');
395 if (pslash != NULL)
396 *pslash = '\0'; /* get rid of /lib */
397 }
398 }
399
400 home_path = malloc(strlen(buf) + 1);
401 if (home_path == NULL)
402 return;
403 strcpy(home_path, buf);
404 Arguments::set_java_home(home_path);
405
406 if (!set_boot_path('/', ':'))
407 return;
408 }
409
410 /*
411 * Where to look for native libraries
412 *
413 * Note: Due to a legacy implementation, most of the library path
414 * is set in the launcher. This was to accomodate linking restrictions
415 * on legacy Linux implementations (which are no longer supported).
416 * Eventually, all the library path setting will be done here.
417 *
418 * However, to prevent the proliferation of improperly built native
419 * libraries, the new path component /usr/java/packages is added here.
420 * Eventually, all the library path setting will be done here.
421 */
422 {
423 char *ld_library_path;
424
425 /*
426 * Construct the invariant part of ld_library_path. Note that the
427 * space for the colon and the trailing null are provided by the
428 * nulls included by the sizeof operator (so actually we allocate
429 * a byte more than necessary).
430 */
431 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
432 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
433 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
434
435 /*
436 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
437 * should always exist (until the legacy problem cited above is
438 * addressed).
439 */
440 char *v = getenv("LD_LIBRARY_PATH");
441 if (v != NULL) {
442 char *t = ld_library_path;
443 /* That's +1 for the colon and +1 for the trailing '\0' */
444 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
445 sprintf(ld_library_path, "%s:%s", v, t);
446 }
447 Arguments::set_library_path(ld_library_path);
448 }
449
450 /*
451 * Extensions directories.
452 *
453 * Note that the space for the colon and the trailing null are provided
454 * by the nulls included by the sizeof operator (so actually one byte more
455 * than necessary is allocated).
456 */
457 {
458 char *buf = malloc(strlen(Arguments::get_java_home()) +
459 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
460 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
461 Arguments::get_java_home());
462 Arguments::set_ext_dirs(buf);
463 }
464
465 /* Endorsed standards default directory. */
466 {
467 char * buf;
468 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
469 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
470 Arguments::set_endorsed_dirs(buf);
471 }
472 }
473
474 #undef malloc
475 #undef getenv
476 #undef EXTENSIONS_DIR
477 #undef ENDORSED_DIR
478
479 // Done
480 return;
481 }
482
483 ////////////////////////////////////////////////////////////////////////////////
484 // breakpoint support
485
486 void os::breakpoint() {
487 BREAKPOINT;
488 }
489
490 extern "C" void breakpoint() {
491 // use debugger to set breakpoint here
492 }
493
494 ////////////////////////////////////////////////////////////////////////////////
495 // signal support
496
497 debug_only(static bool signal_sets_initialized = false);
498 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
499
500 bool os::Linux::is_sig_ignored(int sig) {
501 struct sigaction oact;
502 sigaction(sig, (struct sigaction*)NULL, &oact);
503 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
504 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
505 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
506 return true;
507 else
508 return false;
509 }
510
511 void os::Linux::signal_sets_init() {
512 // Should also have an assertion stating we are still single-threaded.
513 assert(!signal_sets_initialized, "Already initialized");
514 // Fill in signals that are necessarily unblocked for all threads in
515 // the VM. Currently, we unblock the following signals:
516 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
517 // by -Xrs (=ReduceSignalUsage));
518 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
519 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
520 // the dispositions or masks wrt these signals.
521 // Programs embedding the VM that want to use the above signals for their
522 // own purposes must, at this time, use the "-Xrs" option to prevent
523 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
524 // (See bug 4345157, and other related bugs).
525 // In reality, though, unblocking these signals is really a nop, since
526 // these signals are not blocked by default.
527 sigemptyset(&unblocked_sigs);
528 sigemptyset(&allowdebug_blocked_sigs);
529 sigaddset(&unblocked_sigs, SIGILL);
530 sigaddset(&unblocked_sigs, SIGSEGV);
531 sigaddset(&unblocked_sigs, SIGBUS);
532 sigaddset(&unblocked_sigs, SIGFPE);
533 sigaddset(&unblocked_sigs, SR_signum);
534
535 if (!ReduceSignalUsage) {
536 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
537 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
538 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
539 }
540 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
541 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
542 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
543 }
544 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
545 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
546 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
547 }
548 }
549 // Fill in signals that are blocked by all but the VM thread.
550 sigemptyset(&vm_sigs);
551 if (!ReduceSignalUsage)
552 sigaddset(&vm_sigs, BREAK_SIGNAL);
553 debug_only(signal_sets_initialized = true);
554
555 }
556
557 // These are signals that are unblocked while a thread is running Java.
558 // (For some reason, they get blocked by default.)
559 sigset_t* os::Linux::unblocked_signals() {
560 assert(signal_sets_initialized, "Not initialized");
561 return &unblocked_sigs;
562 }
563
564 // These are the signals that are blocked while a (non-VM) thread is
565 // running Java. Only the VM thread handles these signals.
566 sigset_t* os::Linux::vm_signals() {
567 assert(signal_sets_initialized, "Not initialized");
568 return &vm_sigs;
569 }
570
571 // These are signals that are blocked during cond_wait to allow debugger in
572 sigset_t* os::Linux::allowdebug_blocked_signals() {
573 assert(signal_sets_initialized, "Not initialized");
574 return &allowdebug_blocked_sigs;
575 }
576
577 void os::Linux::hotspot_sigmask(Thread* thread) {
578
579 //Save caller's signal mask before setting VM signal mask
580 sigset_t caller_sigmask;
581 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
582
583 OSThread* osthread = thread->osthread();
584 osthread->set_caller_sigmask(caller_sigmask);
585
586 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
587
588 if (!ReduceSignalUsage) {
589 if (thread->is_VM_thread()) {
590 // Only the VM thread handles BREAK_SIGNAL ...
591 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
592 } else {
593 // ... all other threads block BREAK_SIGNAL
594 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
595 }
596 }
597 }
598
599 //////////////////////////////////////////////////////////////////////////////
600 // detecting pthread library
601
602 void os::Linux::libpthread_init() {
603 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
604 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
605 // generic name for earlier versions.
606 // Define macros here so we can build HotSpot on old systems.
607 # ifndef _CS_GNU_LIBC_VERSION
608 # define _CS_GNU_LIBC_VERSION 2
609 # endif
610 # ifndef _CS_GNU_LIBPTHREAD_VERSION
611 # define _CS_GNU_LIBPTHREAD_VERSION 3
612 # endif
613
614 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
615 if (n > 0) {
616 char *str = (char *)malloc(n, mtInternal);
617 confstr(_CS_GNU_LIBC_VERSION, str, n);
618 os::Linux::set_glibc_version(str);
619 } else {
620 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
621 static char _gnu_libc_version[32];
622 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
623 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
624 os::Linux::set_glibc_version(_gnu_libc_version);
625 }
626
627 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
628 if (n > 0) {
629 char *str = (char *)malloc(n, mtInternal);
630 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
631 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
632 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
633 // is the case. LinuxThreads has a hard limit on max number of threads.
634 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
635 // On the other hand, NPTL does not have such a limit, sysconf()
636 // will return -1 and errno is not changed. Check if it is really NPTL.
637 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
638 strstr(str, "NPTL") &&
639 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
640 free(str);
641 os::Linux::set_libpthread_version("linuxthreads");
642 } else {
643 os::Linux::set_libpthread_version(str);
644 }
645 } else {
646 // glibc before 2.3.2 only has LinuxThreads.
647 os::Linux::set_libpthread_version("linuxthreads");
648 }
649
650 if (strstr(libpthread_version(), "NPTL")) {
651 os::Linux::set_is_NPTL();
652 } else {
653 os::Linux::set_is_LinuxThreads();
654 }
655
656 // LinuxThreads have two flavors: floating-stack mode, which allows variable
657 // stack size; and fixed-stack mode. NPTL is always floating-stack.
658 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
659 os::Linux::set_is_floating_stack();
660 }
661 }
662
663 /////////////////////////////////////////////////////////////////////////////
664 // thread stack
665
666 // Force Linux kernel to expand current thread stack. If "bottom" is close
667 // to the stack guard, caller should block all signals.
668 //
669 // MAP_GROWSDOWN:
670 // A special mmap() flag that is used to implement thread stacks. It tells
671 // kernel that the memory region should extend downwards when needed. This
672 // allows early versions of LinuxThreads to only mmap the first few pages
673 // when creating a new thread. Linux kernel will automatically expand thread
674 // stack as needed (on page faults).
675 //
676 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
677 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
678 // region, it's hard to tell if the fault is due to a legitimate stack
679 // access or because of reading/writing non-exist memory (e.g. buffer
680 // overrun). As a rule, if the fault happens below current stack pointer,
681 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
682 // application (see Linux kernel fault.c).
683 //
684 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
685 // stack overflow detection.
686 //
687 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
688 // not use this flag. However, the stack of initial thread is not created
689 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
690 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
691 // and then attach the thread to JVM.
692 //
693 // To get around the problem and allow stack banging on Linux, we need to
694 // manually expand thread stack after receiving the SIGSEGV.
695 //
696 // There are two ways to expand thread stack to address "bottom", we used
697 // both of them in JVM before 1.5:
698 // 1. adjust stack pointer first so that it is below "bottom", and then
699 // touch "bottom"
700 // 2. mmap() the page in question
701 //
702 // Now alternate signal stack is gone, it's harder to use 2. For instance,
703 // if current sp is already near the lower end of page 101, and we need to
704 // call mmap() to map page 100, it is possible that part of the mmap() frame
705 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
706 // That will destroy the mmap() frame and cause VM to crash.
707 //
708 // The following code works by adjusting sp first, then accessing the "bottom"
709 // page to force a page fault. Linux kernel will then automatically expand the
710 // stack mapping.
711 //
712 // _expand_stack_to() assumes its frame size is less than page size, which
713 // should always be true if the function is not inlined.
714
715 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
716 #define NOINLINE
717 #else
718 #define NOINLINE __attribute__ ((noinline))
719 #endif
720
721 static void _expand_stack_to(address bottom) NOINLINE;
722
723 static void _expand_stack_to(address bottom) {
724 address sp;
725 size_t size;
726 volatile char *p;
727
728 // Adjust bottom to point to the largest address within the same page, it
729 // gives us a one-page buffer if alloca() allocates slightly more memory.
730 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
731 bottom += os::Linux::page_size() - 1;
732
733 // sp might be slightly above current stack pointer; if that's the case, we
734 // will alloca() a little more space than necessary, which is OK. Don't use
735 // os::current_stack_pointer(), as its result can be slightly below current
736 // stack pointer, causing us to not alloca enough to reach "bottom".
737 sp = (address)&sp;
738
739 if (sp > bottom) {
740 size = sp - bottom;
741 p = (volatile char *)alloca(size);
742 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
743 p[0] = '\0';
744 }
745 }
746
747 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
748 assert(t!=NULL, "just checking");
749 assert(t->osthread()->expanding_stack(), "expand should be set");
750 assert(t->stack_base() != NULL, "stack_base was not initialized");
751
752 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
753 sigset_t mask_all, old_sigset;
754 sigfillset(&mask_all);
755 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
756 _expand_stack_to(addr);
757 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
758 return true;
759 }
760 return false;
761 }
762
763 //////////////////////////////////////////////////////////////////////////////
764 // create new thread
765
766 static address highest_vm_reserved_address();
767
768 // check if it's safe to start a new thread
769 static bool _thread_safety_check(Thread* thread) {
770 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
771 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
772 // Heap is mmap'ed at lower end of memory space. Thread stacks are
773 // allocated (MAP_FIXED) from high address space. Every thread stack
774 // occupies a fixed size slot (usually 2Mbytes, but user can change
775 // it to other values if they rebuild LinuxThreads).
776 //
777 // Problem with MAP_FIXED is that mmap() can still succeed even part of
778 // the memory region has already been mmap'ed. That means if we have too
779 // many threads and/or very large heap, eventually thread stack will
780 // collide with heap.
781 //
782 // Here we try to prevent heap/stack collision by comparing current
783 // stack bottom with the highest address that has been mmap'ed by JVM
784 // plus a safety margin for memory maps created by native code.
785 //
786 // This feature can be disabled by setting ThreadSafetyMargin to 0
787 //
788 if (ThreadSafetyMargin > 0) {
789 address stack_bottom = os::current_stack_base() - os::current_stack_size();
790
791 // not safe if our stack extends below the safety margin
792 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
793 } else {
794 return true;
795 }
796 } else {
797 // Floating stack LinuxThreads or NPTL:
798 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
799 // there's not enough space left, pthread_create() will fail. If we come
800 // here, that means enough space has been reserved for stack.
801 return true;
802 }
803 }
804
805 // Thread start routine for all newly created threads
806 static void *java_start(Thread *thread) {
807 // Try to randomize the cache line index of hot stack frames.
808 // This helps when threads of the same stack traces evict each other's
809 // cache lines. The threads can be either from the same JVM instance, or
810 // from different JVM instances. The benefit is especially true for
811 // processors with hyperthreading technology.
812 static int counter = 0;
813 int pid = os::current_process_id();
814 alloca(((pid ^ counter++) & 7) * 128);
815
816 ThreadLocalStorage::set_thread(thread);
817
818 OSThread* osthread = thread->osthread();
819 Monitor* sync = osthread->startThread_lock();
820
821 // non floating stack LinuxThreads needs extra check, see above
822 if (!_thread_safety_check(thread)) {
823 // notify parent thread
824 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
825 osthread->set_state(ZOMBIE);
826 sync->notify_all();
827 return NULL;
828 }
829
830 // thread_id is kernel thread id (similar to Solaris LWP id)
831 osthread->set_thread_id(os::Linux::gettid());
832
833 if (UseNUMA) {
834 int lgrp_id = os::numa_get_group_id();
835 if (lgrp_id != -1) {
836 thread->set_lgrp_id(lgrp_id);
837 }
838 }
839 // initialize signal mask for this thread
840 os::Linux::hotspot_sigmask(thread);
841
842 // initialize floating point control register
843 os::Linux::init_thread_fpu_state();
844
845 // handshaking with parent thread
846 {
847 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
848
849 // notify parent thread
850 osthread->set_state(INITIALIZED);
851 sync->notify_all();
852
853 // wait until os::start_thread()
854 while (osthread->get_state() == INITIALIZED) {
855 sync->wait(Mutex::_no_safepoint_check_flag);
856 }
857 }
858
859 // call one more level start routine
860 thread->run();
861
862 return 0;
863 }
864
865 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
866 assert(thread->osthread() == NULL, "caller responsible");
867
868 // Allocate the OSThread object
869 OSThread* osthread = new OSThread(NULL, NULL);
870 if (osthread == NULL) {
871 return false;
872 }
873
874 // set the correct thread state
875 osthread->set_thread_type(thr_type);
876
877 // Initial state is ALLOCATED but not INITIALIZED
878 osthread->set_state(ALLOCATED);
879
880 thread->set_osthread(osthread);
881
882 // init thread attributes
883 pthread_attr_t attr;
884 pthread_attr_init(&attr);
885 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
886
887 // stack size
888 if (os::Linux::supports_variable_stack_size()) {
889 // calculate stack size if it's not specified by caller
890 if (stack_size == 0) {
891 stack_size = os::Linux::default_stack_size(thr_type);
892
893 switch (thr_type) {
894 case os::java_thread:
895 // Java threads use ThreadStackSize which default value can be
896 // changed with the flag -Xss
897 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
898 stack_size = JavaThread::stack_size_at_create();
899 break;
900 case os::compiler_thread:
901 if (CompilerThreadStackSize > 0) {
902 stack_size = (size_t)(CompilerThreadStackSize * K);
903 break;
904 } // else fall through:
905 // use VMThreadStackSize if CompilerThreadStackSize is not defined
906 case os::vm_thread:
907 case os::pgc_thread:
908 case os::cgc_thread:
909 case os::watcher_thread:
910 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
911 break;
912 }
913 }
914
915 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
916 pthread_attr_setstacksize(&attr, stack_size);
917 } else {
918 // let pthread_create() pick the default value.
919 }
920
921 // glibc guard page
922 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
923
924 ThreadState state;
925
926 {
927 // Serialize thread creation if we are running with fixed stack LinuxThreads
928 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
929 if (lock) {
930 os::Linux::createThread_lock()->lock_without_safepoint_check();
931 }
932
933 pthread_t tid;
934 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
935
936 pthread_attr_destroy(&attr);
937
938 if (ret != 0) {
939 if (PrintMiscellaneous && (Verbose || WizardMode)) {
940 perror("pthread_create()");
941 }
942 // Need to clean up stuff we've allocated so far
943 thread->set_osthread(NULL);
944 delete osthread;
945 if (lock) os::Linux::createThread_lock()->unlock();
946 return false;
947 }
948
949 // Store pthread info into the OSThread
950 osthread->set_pthread_id(tid);
951
952 // Wait until child thread is either initialized or aborted
953 {
954 Monitor* sync_with_child = osthread->startThread_lock();
955 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
956 while ((state = osthread->get_state()) == ALLOCATED) {
957 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
958 }
959 }
960
961 if (lock) {
962 os::Linux::createThread_lock()->unlock();
963 }
964 }
965
966 // Aborted due to thread limit being reached
967 if (state == ZOMBIE) {
968 thread->set_osthread(NULL);
969 delete osthread;
970 return false;
971 }
972
973 // The thread is returned suspended (in state INITIALIZED),
974 // and is started higher up in the call chain
975 assert(state == INITIALIZED, "race condition");
976 return true;
977 }
978
979 /////////////////////////////////////////////////////////////////////////////
980 // attach existing thread
981
982 // bootstrap the main thread
983 bool os::create_main_thread(JavaThread* thread) {
984 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
985 return create_attached_thread(thread);
986 }
987
988 bool os::create_attached_thread(JavaThread* thread) {
989 #ifdef ASSERT
990 thread->verify_not_published();
991 #endif
992
993 // Allocate the OSThread object
994 OSThread* osthread = new OSThread(NULL, NULL);
995
996 if (osthread == NULL) {
997 return false;
998 }
999
1000 // Store pthread info into the OSThread
1001 osthread->set_thread_id(os::Linux::gettid());
1002 osthread->set_pthread_id(::pthread_self());
1003
1004 // initialize floating point control register
1005 os::Linux::init_thread_fpu_state();
1006
1007 // Initial thread state is RUNNABLE
1008 osthread->set_state(RUNNABLE);
1009
1010 thread->set_osthread(osthread);
1011
1012 if (UseNUMA) {
1013 int lgrp_id = os::numa_get_group_id();
1014 if (lgrp_id != -1) {
1015 thread->set_lgrp_id(lgrp_id);
1016 }
1017 }
1018
1019 if (os::Linux::is_initial_thread()) {
1020 // If current thread is initial thread, its stack is mapped on demand,
1021 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1022 // the entire stack region to avoid SEGV in stack banging.
1023 // It is also useful to get around the heap-stack-gap problem on SuSE
1024 // kernel (see 4821821 for details). We first expand stack to the top
1025 // of yellow zone, then enable stack yellow zone (order is significant,
1026 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1027 // is no gap between the last two virtual memory regions.
1028
1029 JavaThread *jt = (JavaThread *)thread;
1030 address addr = jt->stack_yellow_zone_base();
1031 assert(addr != NULL, "initialization problem?");
1032 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1033
1034 osthread->set_expanding_stack();
1035 os::Linux::manually_expand_stack(jt, addr);
1036 osthread->clear_expanding_stack();
1037 }
1038
1039 // initialize signal mask for this thread
1040 // and save the caller's signal mask
1041 os::Linux::hotspot_sigmask(thread);
1042
1043 return true;
1044 }
1045
1046 void os::pd_start_thread(Thread* thread) {
1047 OSThread * osthread = thread->osthread();
1048 assert(osthread->get_state() != INITIALIZED, "just checking");
1049 Monitor* sync_with_child = osthread->startThread_lock();
1050 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1051 sync_with_child->notify();
1052 }
1053
1054 // Free Linux resources related to the OSThread
1055 void os::free_thread(OSThread* osthread) {
1056 assert(osthread != NULL, "osthread not set");
1057
1058 if (Thread::current()->osthread() == osthread) {
1059 // Restore caller's signal mask
1060 sigset_t sigmask = osthread->caller_sigmask();
1061 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1062 }
1063
1064 delete osthread;
1065 }
1066
1067 //////////////////////////////////////////////////////////////////////////////
1068 // thread local storage
1069
1070 int os::allocate_thread_local_storage() {
1071 pthread_key_t key;
1072 int rslt = pthread_key_create(&key, NULL);
1073 assert(rslt == 0, "cannot allocate thread local storage");
1074 return (int)key;
1075 }
1076
1077 // Note: This is currently not used by VM, as we don't destroy TLS key
1078 // on VM exit.
1079 void os::free_thread_local_storage(int index) {
1080 int rslt = pthread_key_delete((pthread_key_t)index);
1081 assert(rslt == 0, "invalid index");
1082 }
1083
1084 void os::thread_local_storage_at_put(int index, void* value) {
1085 int rslt = pthread_setspecific((pthread_key_t)index, value);
1086 assert(rslt == 0, "pthread_setspecific failed");
1087 }
1088
1089 extern "C" Thread* get_thread() {
1090 return ThreadLocalStorage::thread();
1091 }
1092
1093 //////////////////////////////////////////////////////////////////////////////
1094 // initial thread
1095
1096 // Check if current thread is the initial thread, similar to Solaris thr_main.
1097 bool os::Linux::is_initial_thread(void) {
1098 char dummy;
1099 // If called before init complete, thread stack bottom will be null.
1100 // Can be called if fatal error occurs before initialization.
1101 if (initial_thread_stack_bottom() == NULL) return false;
1102 assert(initial_thread_stack_bottom() != NULL &&
1103 initial_thread_stack_size() != 0,
1104 "os::init did not locate initial thread's stack region");
1105 if ((address)&dummy >= initial_thread_stack_bottom() &&
1106 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1107 return true;
1108 else return false;
1109 }
1110
1111 // Find the virtual memory area that contains addr
1112 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1113 FILE *fp = fopen("/proc/self/maps", "r");
1114 if (fp) {
1115 address low, high;
1116 while (!feof(fp)) {
1117 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1118 if (low <= addr && addr < high) {
1119 if (vma_low) *vma_low = low;
1120 if (vma_high) *vma_high = high;
1121 fclose (fp);
1122 return true;
1123 }
1124 }
1125 for (;;) {
1126 int ch = fgetc(fp);
1127 if (ch == EOF || ch == (int)'\n') break;
1128 }
1129 }
1130 fclose(fp);
1131 }
1132 return false;
1133 }
1134
1135 // Locate initial thread stack. This special handling of initial thread stack
1136 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1137 // bogus value for initial thread.
1138 void os::Linux::capture_initial_stack(size_t max_size) {
1139 // stack size is the easy part, get it from RLIMIT_STACK
1140 size_t stack_size;
1141 struct rlimit rlim;
1142 getrlimit(RLIMIT_STACK, &rlim);
1143 stack_size = rlim.rlim_cur;
1144
1145 // 6308388: a bug in ld.so will relocate its own .data section to the
1146 // lower end of primordial stack; reduce ulimit -s value a little bit
1147 // so we won't install guard page on ld.so's data section.
1148 stack_size -= 2 * page_size();
1149
1150 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1151 // 7.1, in both cases we will get 2G in return value.
1152 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1153 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1154 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1155 // in case other parts in glibc still assumes 2M max stack size.
1156 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1157 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1158 if (stack_size > 2 * K * K IA64_ONLY(*2))
1159 stack_size = 2 * K * K IA64_ONLY(*2);
1160 // Try to figure out where the stack base (top) is. This is harder.
1161 //
1162 // When an application is started, glibc saves the initial stack pointer in
1163 // a global variable "__libc_stack_end", which is then used by system
1164 // libraries. __libc_stack_end should be pretty close to stack top. The
1165 // variable is available since the very early days. However, because it is
1166 // a private interface, it could disappear in the future.
1167 //
1168 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1169 // to __libc_stack_end, it is very close to stack top, but isn't the real
1170 // stack top. Note that /proc may not exist if VM is running as a chroot
1171 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1172 // /proc/<pid>/stat could change in the future (though unlikely).
1173 //
1174 // We try __libc_stack_end first. If that doesn't work, look for
1175 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1176 // as a hint, which should work well in most cases.
1177
1178 uintptr_t stack_start;
1179
1180 // try __libc_stack_end first
1181 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1182 if (p && *p) {
1183 stack_start = *p;
1184 } else {
1185 // see if we can get the start_stack field from /proc/self/stat
1186 FILE *fp;
1187 int pid;
1188 char state;
1189 int ppid;
1190 int pgrp;
1191 int session;
1192 int nr;
1193 int tpgrp;
1194 unsigned long flags;
1195 unsigned long minflt;
1196 unsigned long cminflt;
1197 unsigned long majflt;
1198 unsigned long cmajflt;
1199 unsigned long utime;
1200 unsigned long stime;
1201 long cutime;
1202 long cstime;
1203 long prio;
1204 long nice;
1205 long junk;
1206 long it_real;
1207 uintptr_t start;
1208 uintptr_t vsize;
1209 intptr_t rss;
1210 uintptr_t rsslim;
1211 uintptr_t scodes;
1212 uintptr_t ecode;
1213 int i;
1214
1215 // Figure what the primordial thread stack base is. Code is inspired
1216 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1217 // followed by command name surrounded by parentheses, state, etc.
1218 char stat[2048];
1219 int statlen;
1220
1221 fp = fopen("/proc/self/stat", "r");
1222 if (fp) {
1223 statlen = fread(stat, 1, 2047, fp);
1224 stat[statlen] = '\0';
1225 fclose(fp);
1226
1227 // Skip pid and the command string. Note that we could be dealing with
1228 // weird command names, e.g. user could decide to rename java launcher
1229 // to "java 1.4.2 :)", then the stat file would look like
1230 // 1234 (java 1.4.2 :)) R ... ...
1231 // We don't really need to know the command string, just find the last
1232 // occurrence of ")" and then start parsing from there. See bug 4726580.
1233 char * s = strrchr(stat, ')');
1234
1235 i = 0;
1236 if (s) {
1237 // Skip blank chars
1238 do s++; while (isspace(*s));
1239
1240 #define _UFM UINTX_FORMAT
1241 #define _DFM INTX_FORMAT
1242
1243 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1244 /* 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 */
1245 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,
1246 &state, /* 3 %c */
1247 &ppid, /* 4 %d */
1248 &pgrp, /* 5 %d */
1249 &session, /* 6 %d */
1250 &nr, /* 7 %d */
1251 &tpgrp, /* 8 %d */
1252 &flags, /* 9 %lu */
1253 &minflt, /* 10 %lu */
1254 &cminflt, /* 11 %lu */
1255 &majflt, /* 12 %lu */
1256 &cmajflt, /* 13 %lu */
1257 &utime, /* 14 %lu */
1258 &stime, /* 15 %lu */
1259 &cutime, /* 16 %ld */
1260 &cstime, /* 17 %ld */
1261 &prio, /* 18 %ld */
1262 &nice, /* 19 %ld */
1263 &junk, /* 20 %ld */
1264 &it_real, /* 21 %ld */
1265 &start, /* 22 UINTX_FORMAT */
1266 &vsize, /* 23 UINTX_FORMAT */
1267 &rss, /* 24 INTX_FORMAT */
1268 &rsslim, /* 25 UINTX_FORMAT */
1269 &scodes, /* 26 UINTX_FORMAT */
1270 &ecode, /* 27 UINTX_FORMAT */
1271 &stack_start); /* 28 UINTX_FORMAT */
1272 }
1273
1274 #undef _UFM
1275 #undef _DFM
1276
1277 if (i != 28 - 2) {
1278 assert(false, "Bad conversion from /proc/self/stat");
1279 // product mode - assume we are the initial thread, good luck in the
1280 // embedded case.
1281 warning("Can't detect initial thread stack location - bad conversion");
1282 stack_start = (uintptr_t) &rlim;
1283 }
1284 } else {
1285 // For some reason we can't open /proc/self/stat (for example, running on
1286 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1287 // most cases, so don't abort:
1288 warning("Can't detect initial thread stack location - no /proc/self/stat");
1289 stack_start = (uintptr_t) &rlim;
1290 }
1291 }
1292
1293 // Now we have a pointer (stack_start) very close to the stack top, the
1294 // next thing to do is to figure out the exact location of stack top. We
1295 // can find out the virtual memory area that contains stack_start by
1296 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1297 // and its upper limit is the real stack top. (again, this would fail if
1298 // running inside chroot, because /proc may not exist.)
1299
1300 uintptr_t stack_top;
1301 address low, high;
1302 if (find_vma((address)stack_start, &low, &high)) {
1303 // success, "high" is the true stack top. (ignore "low", because initial
1304 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1305 stack_top = (uintptr_t)high;
1306 } else {
1307 // failed, likely because /proc/self/maps does not exist
1308 warning("Can't detect initial thread stack location - find_vma failed");
1309 // best effort: stack_start is normally within a few pages below the real
1310 // stack top, use it as stack top, and reduce stack size so we won't put
1311 // guard page outside stack.
1312 stack_top = stack_start;
1313 stack_size -= 16 * page_size();
1314 }
1315
1316 // stack_top could be partially down the page so align it
1317 stack_top = align_size_up(stack_top, page_size());
1318
1319 if (max_size && stack_size > max_size) {
1320 _initial_thread_stack_size = max_size;
1321 } else {
1322 _initial_thread_stack_size = stack_size;
1323 }
1324
1325 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1326 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1327 }
1328
1329 ////////////////////////////////////////////////////////////////////////////////
1330 // time support
1331
1332 // Time since start-up in seconds to a fine granularity.
1333 // Used by VMSelfDestructTimer and the MemProfiler.
1334 double os::elapsedTime() {
1335
1336 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1337 }
1338
1339 jlong os::elapsed_counter() {
1340 return javaTimeNanos() - initial_time_count;
1341 }
1342
1343 jlong os::elapsed_frequency() {
1344 return NANOSECS_PER_SEC; // nanosecond resolution
1345 }
1346
1347 bool os::supports_vtime() { return true; }
1348 bool os::enable_vtime() { return false; }
1349 bool os::vtime_enabled() { return false; }
1350
1351 double os::elapsedVTime() {
1352 struct rusage usage;
1353 int retval = getrusage(RUSAGE_THREAD, &usage);
1354 if (retval == 0) {
1355 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);
1356 } else {
1357 // better than nothing, but not much
1358 return elapsedTime();
1359 }
1360 }
1361
1362 jlong os::javaTimeMillis() {
1363 timeval time;
1364 int status = gettimeofday(&time, NULL);
1365 assert(status != -1, "linux error");
1366 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1367 }
1368
1369 #ifndef CLOCK_MONOTONIC
1370 #define CLOCK_MONOTONIC (1)
1371 #endif
1372
1373 void os::Linux::clock_init() {
1374 // we do dlopen's in this particular order due to bug in linux
1375 // dynamical loader (see 6348968) leading to crash on exit
1376 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1377 if (handle == NULL) {
1378 handle = dlopen("librt.so", RTLD_LAZY);
1379 }
1380
1381 if (handle) {
1382 int (*clock_getres_func)(clockid_t, struct timespec*) =
1383 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1384 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1385 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1386 if (clock_getres_func && clock_gettime_func) {
1387 // See if monotonic clock is supported by the kernel. Note that some
1388 // early implementations simply return kernel jiffies (updated every
1389 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1390 // for nano time (though the monotonic property is still nice to have).
1391 // It's fixed in newer kernels, however clock_getres() still returns
1392 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1393 // resolution for now. Hopefully as people move to new kernels, this
1394 // won't be a problem.
1395 struct timespec res;
1396 struct timespec tp;
1397 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1398 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1399 // yes, monotonic clock is supported
1400 _clock_gettime = clock_gettime_func;
1401 return;
1402 } else {
1403 // close librt if there is no monotonic clock
1404 dlclose(handle);
1405 }
1406 }
1407 }
1408 warning("No monotonic clock was available - timed services may " \
1409 "be adversely affected if the time-of-day clock changes");
1410 }
1411
1412 #ifndef SYS_clock_getres
1413
1414 #if defined(IA32) || defined(AMD64)
1415 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1416 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1417 #else
1418 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1419 #define sys_clock_getres(x,y) -1
1420 #endif
1421
1422 #else
1423 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1424 #endif
1425
1426 void os::Linux::fast_thread_clock_init() {
1427 if (!UseLinuxPosixThreadCPUClocks) {
1428 return;
1429 }
1430 clockid_t clockid;
1431 struct timespec tp;
1432 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1433 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1434
1435 // Switch to using fast clocks for thread cpu time if
1436 // the sys_clock_getres() returns 0 error code.
1437 // Note, that some kernels may support the current thread
1438 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1439 // returned by the pthread_getcpuclockid().
1440 // If the fast Posix clocks are supported then the sys_clock_getres()
1441 // must return at least tp.tv_sec == 0 which means a resolution
1442 // better than 1 sec. This is extra check for reliability.
1443
1444 if(pthread_getcpuclockid_func &&
1445 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1446 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1447
1448 _supports_fast_thread_cpu_time = true;
1449 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1450 }
1451 }
1452
1453 jlong os::javaTimeNanos() {
1454 if (Linux::supports_monotonic_clock()) {
1455 struct timespec tp;
1456 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1457 assert(status == 0, "gettime error");
1458 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1459 return result;
1460 } else {
1461 timeval time;
1462 int status = gettimeofday(&time, NULL);
1463 assert(status != -1, "linux error");
1464 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1465 return 1000 * usecs;
1466 }
1467 }
1468
1469 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1470 if (Linux::supports_monotonic_clock()) {
1471 info_ptr->max_value = ALL_64_BITS;
1472
1473 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1474 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1475 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1476 } else {
1477 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1478 info_ptr->max_value = ALL_64_BITS;
1479
1480 // gettimeofday is a real time clock so it skips
1481 info_ptr->may_skip_backward = true;
1482 info_ptr->may_skip_forward = true;
1483 }
1484
1485 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1486 }
1487
1488 // Return the real, user, and system times in seconds from an
1489 // arbitrary fixed point in the past.
1490 bool os::getTimesSecs(double* process_real_time,
1491 double* process_user_time,
1492 double* process_system_time) {
1493 struct tms ticks;
1494 clock_t real_ticks = times(&ticks);
1495
1496 if (real_ticks == (clock_t) (-1)) {
1497 return false;
1498 } else {
1499 double ticks_per_second = (double) clock_tics_per_sec;
1500 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1501 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1502 *process_real_time = ((double) real_ticks) / ticks_per_second;
1503
1504 return true;
1505 }
1506 }
1507
1508
1509 char * os::local_time_string(char *buf, size_t buflen) {
1510 struct tm t;
1511 time_t long_time;
1512 time(&long_time);
1513 localtime_r(&long_time, &t);
1514 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1515 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1516 t.tm_hour, t.tm_min, t.tm_sec);
1517 return buf;
1518 }
1519
1520 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1521 return localtime_r(clock, res);
1522 }
1523
1524 ////////////////////////////////////////////////////////////////////////////////
1525 // runtime exit support
1526
1527 // Note: os::shutdown() might be called very early during initialization, or
1528 // called from signal handler. Before adding something to os::shutdown(), make
1529 // sure it is async-safe and can handle partially initialized VM.
1530 void os::shutdown() {
1531
1532 // allow PerfMemory to attempt cleanup of any persistent resources
1533 perfMemory_exit();
1534
1535 // needs to remove object in file system
1536 AttachListener::abort();
1537
1538 // flush buffered output, finish log files
1539 ostream_abort();
1540
1541 // Check for abort hook
1542 abort_hook_t abort_hook = Arguments::abort_hook();
1543 if (abort_hook != NULL) {
1544 abort_hook();
1545 }
1546
1547 }
1548
1549 // Note: os::abort() might be called very early during initialization, or
1550 // called from signal handler. Before adding something to os::abort(), make
1551 // sure it is async-safe and can handle partially initialized VM.
1552 void os::abort(bool dump_core) {
1553 os::shutdown();
1554 if (dump_core) {
1555 #ifndef PRODUCT
1556 fdStream out(defaultStream::output_fd());
1557 out.print_raw("Current thread is ");
1558 char buf[16];
1559 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1560 out.print_raw_cr(buf);
1561 out.print_raw_cr("Dumping core ...");
1562 #endif
1563 ::abort(); // dump core
1564 }
1565
1566 ::exit(1);
1567 }
1568
1569 // Die immediately, no exit hook, no abort hook, no cleanup.
1570 void os::die() {
1571 // _exit() on LinuxThreads only kills current thread
1572 ::abort();
1573 }
1574
1575 // unused on linux for now.
1576 void os::set_error_file(const char *logfile) {}
1577
1578
1579 // This method is a copy of JDK's sysGetLastErrorString
1580 // from src/solaris/hpi/src/system_md.c
1581
1582 size_t os::lasterror(char *buf, size_t len) {
1583
1584 if (errno == 0) return 0;
1585
1586 const char *s = ::strerror(errno);
1587 size_t n = ::strlen(s);
1588 if (n >= len) {
1589 n = len - 1;
1590 }
1591 ::strncpy(buf, s, n);
1592 buf[n] = '\0';
1593 return n;
1594 }
1595
1596 intx os::current_thread_id() { return (intx)pthread_self(); }
1597 int os::current_process_id() {
1598
1599 // Under the old linux thread library, linux gives each thread
1600 // its own process id. Because of this each thread will return
1601 // a different pid if this method were to return the result
1602 // of getpid(2). Linux provides no api that returns the pid
1603 // of the launcher thread for the vm. This implementation
1604 // returns a unique pid, the pid of the launcher thread
1605 // that starts the vm 'process'.
1606
1607 // Under the NPTL, getpid() returns the same pid as the
1608 // launcher thread rather than a unique pid per thread.
1609 // Use gettid() if you want the old pre NPTL behaviour.
1610
1611 // if you are looking for the result of a call to getpid() that
1612 // returns a unique pid for the calling thread, then look at the
1613 // OSThread::thread_id() method in osThread_linux.hpp file
1614
1615 return (int)(_initial_pid ? _initial_pid : getpid());
1616 }
1617
1618 // DLL functions
1619
1620 const char* os::dll_file_extension() { return ".so"; }
1621
1622 // This must be hard coded because it's the system's temporary
1623 // directory not the java application's temp directory, ala java.io.tmpdir.
1624 const char* os::get_temp_directory() { return "/tmp"; }
1625
1626 static bool file_exists(const char* filename) {
1627 struct stat statbuf;
1628 if (filename == NULL || strlen(filename) == 0) {
1629 return false;
1630 }
1631 return os::stat(filename, &statbuf) == 0;
1632 }
1633
1634 bool os::dll_build_name(char* buffer, size_t buflen,
1635 const char* pname, const char* fname) {
1636 bool retval = false;
1637 // Copied from libhpi
1638 const size_t pnamelen = pname ? strlen(pname) : 0;
1639
1640 // Return error on buffer overflow.
1641 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1642 return retval;
1643 }
1644
1645 if (pnamelen == 0) {
1646 snprintf(buffer, buflen, "lib%s.so", fname);
1647 retval = true;
1648 } else if (strchr(pname, *os::path_separator()) != NULL) {
1649 int n;
1650 char** pelements = split_path(pname, &n);
1651 if (pelements == NULL) {
1652 return false;
1653 }
1654 for (int i = 0 ; i < n ; i++) {
1655 // Really shouldn't be NULL, but check can't hurt
1656 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1657 continue; // skip the empty path values
1658 }
1659 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1660 if (file_exists(buffer)) {
1661 retval = true;
1662 break;
1663 }
1664 }
1665 // release the storage
1666 for (int i = 0 ; i < n ; i++) {
1667 if (pelements[i] != NULL) {
1668 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1669 }
1670 }
1671 if (pelements != NULL) {
1672 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1673 }
1674 } else {
1675 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1676 retval = true;
1677 }
1678 return retval;
1679 }
1680
1681 // check if addr is inside libjvm.so
1682 bool os::address_is_in_vm(address addr) {
1683 static address libjvm_base_addr;
1684 Dl_info dlinfo;
1685
1686 if (libjvm_base_addr == NULL) {
1687 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1688 libjvm_base_addr = (address)dlinfo.dli_fbase;
1689 }
1690 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1691 }
1692
1693 if (dladdr((void *)addr, &dlinfo) != 0) {
1694 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1695 }
1696
1697 return false;
1698 }
1699
1700 bool os::dll_address_to_function_name(address addr, char *buf,
1701 int buflen, int *offset) {
1702 // buf is not optional, but offset is optional
1703 assert(buf != NULL, "sanity check");
1704
1705 Dl_info dlinfo;
1706
1707 if (dladdr((void*)addr, &dlinfo) != 0) {
1708 // see if we have a matching symbol
1709 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1710 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1711 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1712 }
1713 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1714 return true;
1715 }
1716 // no matching symbol so try for just file info
1717 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1718 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1719 buf, buflen, offset, dlinfo.dli_fname)) {
1720 return true;
1721 }
1722 }
1723 }
1724
1725 buf[0] = '\0';
1726 if (offset != NULL) *offset = -1;
1727 return false;
1728 }
1729
1730 struct _address_to_library_name {
1731 address addr; // input : memory address
1732 size_t buflen; // size of fname
1733 char* fname; // output: library name
1734 address base; // library base addr
1735 };
1736
1737 static int address_to_library_name_callback(struct dl_phdr_info *info,
1738 size_t size, void *data) {
1739 int i;
1740 bool found = false;
1741 address libbase = NULL;
1742 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1743
1744 // iterate through all loadable segments
1745 for (i = 0; i < info->dlpi_phnum; i++) {
1746 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1747 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1748 // base address of a library is the lowest address of its loaded
1749 // segments.
1750 if (libbase == NULL || libbase > segbase) {
1751 libbase = segbase;
1752 }
1753 // see if 'addr' is within current segment
1754 if (segbase <= d->addr &&
1755 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1756 found = true;
1757 }
1758 }
1759 }
1760
1761 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1762 // so dll_address_to_library_name() can fall through to use dladdr() which
1763 // can figure out executable name from argv[0].
1764 if (found && info->dlpi_name && info->dlpi_name[0]) {
1765 d->base = libbase;
1766 if (d->fname) {
1767 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1768 }
1769 return 1;
1770 }
1771 return 0;
1772 }
1773
1774 bool os::dll_address_to_library_name(address addr, char* buf,
1775 int buflen, int* offset) {
1776 // buf is not optional, but offset is optional
1777 assert(buf != NULL, "sanity check");
1778
1779 Dl_info dlinfo;
1780 struct _address_to_library_name data;
1781
1782 // There is a bug in old glibc dladdr() implementation that it could resolve
1783 // to wrong library name if the .so file has a base address != NULL. Here
1784 // we iterate through the program headers of all loaded libraries to find
1785 // out which library 'addr' really belongs to. This workaround can be
1786 // removed once the minimum requirement for glibc is moved to 2.3.x.
1787 data.addr = addr;
1788 data.fname = buf;
1789 data.buflen = buflen;
1790 data.base = NULL;
1791 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1792
1793 if (rslt) {
1794 // buf already contains library name
1795 if (offset) *offset = addr - data.base;
1796 return true;
1797 }
1798 if (dladdr((void*)addr, &dlinfo) != 0) {
1799 if (dlinfo.dli_fname != NULL) {
1800 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1801 }
1802 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1803 *offset = addr - (address)dlinfo.dli_fbase;
1804 }
1805 return true;
1806 }
1807
1808 buf[0] = '\0';
1809 if (offset) *offset = -1;
1810 return false;
1811 }
1812
1813 // Loads .dll/.so and
1814 // in case of error it checks if .dll/.so was built for the
1815 // same architecture as Hotspot is running on
1816
1817
1818 // Remember the stack's state. The Linux dynamic linker will change
1819 // the stack to 'executable' at most once, so we must safepoint only once.
1820 bool os::Linux::_stack_is_executable = false;
1821
1822 // VM operation that loads a library. This is necessary if stack protection
1823 // of the Java stacks can be lost during loading the library. If we
1824 // do not stop the Java threads, they can stack overflow before the stacks
1825 // are protected again.
1826 class VM_LinuxDllLoad: public VM_Operation {
1827 private:
1828 const char *_filename;
1829 char *_ebuf;
1830 int _ebuflen;
1831 void *_lib;
1832 public:
1833 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1834 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1835 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1836 void doit() {
1837 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1838 os::Linux::_stack_is_executable = true;
1839 }
1840 void* loaded_library() { return _lib; }
1841 };
1842
1843 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1844 {
1845 void * result = NULL;
1846 bool load_attempted = false;
1847
1848 // Check whether the library to load might change execution rights
1849 // of the stack. If they are changed, the protection of the stack
1850 // guard pages will be lost. We need a safepoint to fix this.
1851 //
1852 // See Linux man page execstack(8) for more info.
1853 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1854 ElfFile ef(filename);
1855 if (!ef.specifies_noexecstack()) {
1856 if (!is_init_completed()) {
1857 os::Linux::_stack_is_executable = true;
1858 // This is OK - No Java threads have been created yet, and hence no
1859 // stack guard pages to fix.
1860 //
1861 // This should happen only when you are building JDK7 using a very
1862 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1863 //
1864 // Dynamic loader will make all stacks executable after
1865 // this function returns, and will not do that again.
1866 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1867 } else {
1868 warning("You have loaded library %s which might have disabled stack guard. "
1869 "The VM will try to fix the stack guard now.\n"
1870 "It's highly recommended that you fix the library with "
1871 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1872 filename);
1873
1874 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1875 JavaThread *jt = JavaThread::current();
1876 if (jt->thread_state() != _thread_in_native) {
1877 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1878 // that requires ExecStack. Cannot enter safe point. Let's give up.
1879 warning("Unable to fix stack guard. Giving up.");
1880 } else {
1881 if (!LoadExecStackDllInVMThread) {
1882 // This is for the case where the DLL has an static
1883 // constructor function that executes JNI code. We cannot
1884 // load such DLLs in the VMThread.
1885 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1886 }
1887
1888 ThreadInVMfromNative tiv(jt);
1889 debug_only(VMNativeEntryWrapper vew;)
1890
1891 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1892 VMThread::execute(&op);
1893 if (LoadExecStackDllInVMThread) {
1894 result = op.loaded_library();
1895 }
1896 load_attempted = true;
1897 }
1898 }
1899 }
1900 }
1901
1902 if (!load_attempted) {
1903 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1904 }
1905
1906 if (result != NULL) {
1907 // Successful loading
1908 return result;
1909 }
1910
1911 Elf32_Ehdr elf_head;
1912 int diag_msg_max_length=ebuflen-strlen(ebuf);
1913 char* diag_msg_buf=ebuf+strlen(ebuf);
1914
1915 if (diag_msg_max_length==0) {
1916 // No more space in ebuf for additional diagnostics message
1917 return NULL;
1918 }
1919
1920
1921 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1922
1923 if (file_descriptor < 0) {
1924 // Can't open library, report dlerror() message
1925 return NULL;
1926 }
1927
1928 bool failed_to_read_elf_head=
1929 (sizeof(elf_head)!=
1930 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1931
1932 ::close(file_descriptor);
1933 if (failed_to_read_elf_head) {
1934 // file i/o error - report dlerror() msg
1935 return NULL;
1936 }
1937
1938 typedef struct {
1939 Elf32_Half code; // Actual value as defined in elf.h
1940 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1941 char elf_class; // 32 or 64 bit
1942 char endianess; // MSB or LSB
1943 char* name; // String representation
1944 } arch_t;
1945
1946 #ifndef EM_486
1947 #define EM_486 6 /* Intel 80486 */
1948 #endif
1949
1950 static const arch_t arch_array[]={
1951 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1952 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1953 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1954 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1955 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1956 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1957 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1958 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1959 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1960 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1961 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1962 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1963 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1964 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1965 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1966 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1967 };
1968
1969 #if (defined IA32)
1970 static Elf32_Half running_arch_code=EM_386;
1971 #elif (defined AMD64)
1972 static Elf32_Half running_arch_code=EM_X86_64;
1973 #elif (defined IA64)
1974 static Elf32_Half running_arch_code=EM_IA_64;
1975 #elif (defined __sparc) && (defined _LP64)
1976 static Elf32_Half running_arch_code=EM_SPARCV9;
1977 #elif (defined __sparc) && (!defined _LP64)
1978 static Elf32_Half running_arch_code=EM_SPARC;
1979 #elif (defined __powerpc64__)
1980 static Elf32_Half running_arch_code=EM_PPC64;
1981 #elif (defined __powerpc__)
1982 static Elf32_Half running_arch_code=EM_PPC;
1983 #elif (defined ARM)
1984 static Elf32_Half running_arch_code=EM_ARM;
1985 #elif (defined S390)
1986 static Elf32_Half running_arch_code=EM_S390;
1987 #elif (defined ALPHA)
1988 static Elf32_Half running_arch_code=EM_ALPHA;
1989 #elif (defined MIPSEL)
1990 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1991 #elif (defined PARISC)
1992 static Elf32_Half running_arch_code=EM_PARISC;
1993 #elif (defined MIPS)
1994 static Elf32_Half running_arch_code=EM_MIPS;
1995 #elif (defined M68K)
1996 static Elf32_Half running_arch_code=EM_68K;
1997 #else
1998 #error Method os::dll_load requires that one of following is defined:\
1999 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2000 #endif
2001
2002 // Identify compatability class for VM's architecture and library's architecture
2003 // Obtain string descriptions for architectures
2004
2005 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2006 int running_arch_index=-1;
2007
2008 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2009 if (running_arch_code == arch_array[i].code) {
2010 running_arch_index = i;
2011 }
2012 if (lib_arch.code == arch_array[i].code) {
2013 lib_arch.compat_class = arch_array[i].compat_class;
2014 lib_arch.name = arch_array[i].name;
2015 }
2016 }
2017
2018 assert(running_arch_index != -1,
2019 "Didn't find running architecture code (running_arch_code) in arch_array");
2020 if (running_arch_index == -1) {
2021 // Even though running architecture detection failed
2022 // we may still continue with reporting dlerror() message
2023 return NULL;
2024 }
2025
2026 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2027 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2028 return NULL;
2029 }
2030
2031 #ifndef S390
2032 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2033 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2034 return NULL;
2035 }
2036 #endif // !S390
2037
2038 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2039 if ( lib_arch.name!=NULL ) {
2040 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2041 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2042 lib_arch.name, arch_array[running_arch_index].name);
2043 } else {
2044 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2045 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2046 lib_arch.code,
2047 arch_array[running_arch_index].name);
2048 }
2049 }
2050
2051 return NULL;
2052 }
2053
2054 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2055 void * result = ::dlopen(filename, RTLD_LAZY);
2056 if (result == NULL) {
2057 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2058 ebuf[ebuflen-1] = '\0';
2059 }
2060 return result;
2061 }
2062
2063 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2064 void * result = NULL;
2065 if (LoadExecStackDllInVMThread) {
2066 result = dlopen_helper(filename, ebuf, ebuflen);
2067 }
2068
2069 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2070 // library that requires an executable stack, or which does not have this
2071 // stack attribute set, dlopen changes the stack attribute to executable. The
2072 // read protection of the guard pages gets lost.
2073 //
2074 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2075 // may have been queued at the same time.
2076
2077 if (!_stack_is_executable) {
2078 JavaThread *jt = Threads::first();
2079
2080 while (jt) {
2081 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2082 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2083 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2084 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2085 warning("Attempt to reguard stack yellow zone failed.");
2086 }
2087 }
2088 jt = jt->next();
2089 }
2090 }
2091
2092 return result;
2093 }
2094
2095 /*
2096 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2097 * chances are you might want to run the generated bits against glibc-2.0
2098 * libdl.so, so always use locking for any version of glibc.
2099 */
2100 void* os::dll_lookup(void* handle, const char* name) {
2101 pthread_mutex_lock(&dl_mutex);
2102 void* res = dlsym(handle, name);
2103 pthread_mutex_unlock(&dl_mutex);
2104 return res;
2105 }
2106
2107
2108 static bool _print_ascii_file(const char* filename, outputStream* st) {
2109 int fd = ::open(filename, O_RDONLY);
2110 if (fd == -1) {
2111 return false;
2112 }
2113
2114 char buf[32];
2115 int bytes;
2116 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2117 st->print_raw(buf, bytes);
2118 }
2119
2120 ::close(fd);
2121
2122 return true;
2123 }
2124
2125 void os::print_dll_info(outputStream *st) {
2126 st->print_cr("Dynamic libraries:");
2127
2128 char fname[32];
2129 pid_t pid = os::Linux::gettid();
2130
2131 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2132
2133 if (!_print_ascii_file(fname, st)) {
2134 st->print("Can not get library information for pid = %d\n", pid);
2135 }
2136 }
2137
2138 void os::print_os_info_brief(outputStream* st) {
2139 os::Linux::print_distro_info(st);
2140
2141 os::Posix::print_uname_info(st);
2142
2143 os::Linux::print_libversion_info(st);
2144
2145 }
2146
2147 void os::print_os_info(outputStream* st) {
2148 st->print("OS:");
2149
2150 os::Linux::print_distro_info(st);
2151
2152 os::Posix::print_uname_info(st);
2153
2154 // Print warning if unsafe chroot environment detected
2155 if (unsafe_chroot_detected) {
2156 st->print("WARNING!! ");
2157 st->print_cr(unstable_chroot_error);
2158 }
2159
2160 os::Linux::print_libversion_info(st);
2161
2162 os::Posix::print_rlimit_info(st);
2163
2164 os::Posix::print_load_average(st);
2165
2166 os::Linux::print_full_memory_info(st);
2167 }
2168
2169 // Try to identify popular distros.
2170 // Most Linux distributions have a /etc/XXX-release file, which contains
2171 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2172 // file that also contains the OS version string. Some have more than one
2173 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2174 // /etc/redhat-release.), so the order is important.
2175 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2176 // their own specific XXX-release file as well as a redhat-release file.
2177 // Because of this the XXX-release file needs to be searched for before the
2178 // redhat-release file.
2179 // Since Red Hat has a lsb-release file that is not very descriptive the
2180 // search for redhat-release needs to be before lsb-release.
2181 // Since the lsb-release file is the new standard it needs to be searched
2182 // before the older style release files.
2183 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2184 // next to last resort. The os-release file is a new standard that contains
2185 // distribution information and the system-release file seems to be an old
2186 // standard that has been replaced by the lsb-release and os-release files.
2187 // Searching for the debian_version file is the last resort. It contains
2188 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2189 // "Debian " is printed before the contents of the debian_version file.
2190 void os::Linux::print_distro_info(outputStream* st) {
2191 if (!_print_ascii_file("/etc/oracle-release", st) &&
2192 !_print_ascii_file("/etc/mandriva-release", st) &&
2193 !_print_ascii_file("/etc/mandrake-release", st) &&
2194 !_print_ascii_file("/etc/sun-release", st) &&
2195 !_print_ascii_file("/etc/redhat-release", st) &&
2196 !_print_ascii_file("/etc/lsb-release", st) &&
2197 !_print_ascii_file("/etc/SuSE-release", st) &&
2198 !_print_ascii_file("/etc/turbolinux-release", st) &&
2199 !_print_ascii_file("/etc/gentoo-release", st) &&
2200 !_print_ascii_file("/etc/ltib-release", st) &&
2201 !_print_ascii_file("/etc/angstrom-version", st) &&
2202 !_print_ascii_file("/etc/system-release", st) &&
2203 !_print_ascii_file("/etc/os-release", st)) {
2204
2205 if (file_exists("/etc/debian_version")) {
2206 st->print("Debian ");
2207 _print_ascii_file("/etc/debian_version", st);
2208 } else {
2209 st->print("Linux");
2210 }
2211 }
2212 st->cr();
2213 }
2214
2215 void os::Linux::print_libversion_info(outputStream* st) {
2216 // libc, pthread
2217 st->print("libc:");
2218 st->print(os::Linux::glibc_version()); st->print(" ");
2219 st->print(os::Linux::libpthread_version()); st->print(" ");
2220 if (os::Linux::is_LinuxThreads()) {
2221 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2222 }
2223 st->cr();
2224 }
2225
2226 void os::Linux::print_full_memory_info(outputStream* st) {
2227 st->print("\n/proc/meminfo:\n");
2228 _print_ascii_file("/proc/meminfo", st);
2229 st->cr();
2230 }
2231
2232 void os::print_memory_info(outputStream* st) {
2233
2234 st->print("Memory:");
2235 st->print(" %dk page", os::vm_page_size()>>10);
2236
2237 // values in struct sysinfo are "unsigned long"
2238 struct sysinfo si;
2239 sysinfo(&si);
2240
2241 st->print(", physical " UINT64_FORMAT "k",
2242 os::physical_memory() >> 10);
2243 st->print("(" UINT64_FORMAT "k free)",
2244 os::available_memory() >> 10);
2245 st->print(", swap " UINT64_FORMAT "k",
2246 ((jlong)si.totalswap * si.mem_unit) >> 10);
2247 st->print("(" UINT64_FORMAT "k free)",
2248 ((jlong)si.freeswap * si.mem_unit) >> 10);
2249 st->cr();
2250 }
2251
2252 void os::pd_print_cpu_info(outputStream* st) {
2253 st->print("\n/proc/cpuinfo:\n");
2254 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2255 st->print(" <Not Available>");
2256 }
2257 st->cr();
2258 }
2259
2260 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2261 // but they're the same for all the linux arch that we support
2262 // and they're the same for solaris but there's no common place to put this.
2263 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2264 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2265 "ILL_COPROC", "ILL_BADSTK" };
2266
2267 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2268 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2269 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2270
2271 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2272
2273 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2274
2275 void os::print_siginfo(outputStream* st, void* siginfo) {
2276 st->print("siginfo:");
2277
2278 const int buflen = 100;
2279 char buf[buflen];
2280 siginfo_t *si = (siginfo_t*)siginfo;
2281 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2282 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2283 st->print("si_errno=%s", buf);
2284 } else {
2285 st->print("si_errno=%d", si->si_errno);
2286 }
2287 const int c = si->si_code;
2288 assert(c > 0, "unexpected si_code");
2289 switch (si->si_signo) {
2290 case SIGILL:
2291 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2292 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2293 break;
2294 case SIGFPE:
2295 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2296 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2297 break;
2298 case SIGSEGV:
2299 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2300 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2301 break;
2302 case SIGBUS:
2303 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2304 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2305 break;
2306 default:
2307 st->print(", si_code=%d", si->si_code);
2308 // no si_addr
2309 }
2310
2311 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2312 UseSharedSpaces) {
2313 FileMapInfo* mapinfo = FileMapInfo::current_info();
2314 if (mapinfo->is_in_shared_space(si->si_addr)) {
2315 st->print("\n\nError accessing class data sharing archive." \
2316 " Mapped file inaccessible during execution, " \
2317 " possible disk/network problem.");
2318 }
2319 }
2320 st->cr();
2321 }
2322
2323
2324 static void print_signal_handler(outputStream* st, int sig,
2325 char* buf, size_t buflen);
2326
2327 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2328 st->print_cr("Signal Handlers:");
2329 print_signal_handler(st, SIGSEGV, buf, buflen);
2330 print_signal_handler(st, SIGBUS , buf, buflen);
2331 print_signal_handler(st, SIGFPE , buf, buflen);
2332 print_signal_handler(st, SIGPIPE, buf, buflen);
2333 print_signal_handler(st, SIGXFSZ, buf, buflen);
2334 print_signal_handler(st, SIGILL , buf, buflen);
2335 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2336 print_signal_handler(st, SR_signum, buf, buflen);
2337 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2338 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2339 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2340 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2341 }
2342
2343 static char saved_jvm_path[MAXPATHLEN] = {0};
2344
2345 // Find the full path to the current module, libjvm.so
2346 void os::jvm_path(char *buf, jint buflen) {
2347 // Error checking.
2348 if (buflen < MAXPATHLEN) {
2349 assert(false, "must use a large-enough buffer");
2350 buf[0] = '\0';
2351 return;
2352 }
2353 // Lazy resolve the path to current module.
2354 if (saved_jvm_path[0] != 0) {
2355 strcpy(buf, saved_jvm_path);
2356 return;
2357 }
2358
2359 char dli_fname[MAXPATHLEN];
2360 bool ret = dll_address_to_library_name(
2361 CAST_FROM_FN_PTR(address, os::jvm_path),
2362 dli_fname, sizeof(dli_fname), NULL);
2363 assert(ret, "cannot locate libjvm");
2364 char *rp = NULL;
2365 if (ret && dli_fname[0] != '\0') {
2366 rp = realpath(dli_fname, buf);
2367 }
2368 if (rp == NULL)
2369 return;
2370
2371 if (Arguments::created_by_gamma_launcher()) {
2372 // Support for the gamma launcher. Typical value for buf is
2373 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2374 // the right place in the string, then assume we are installed in a JDK and
2375 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2376 // up the path so it looks like libjvm.so is installed there (append a
2377 // fake suffix hotspot/libjvm.so).
2378 const char *p = buf + strlen(buf) - 1;
2379 for (int count = 0; p > buf && count < 5; ++count) {
2380 for (--p; p > buf && *p != '/'; --p)
2381 /* empty */ ;
2382 }
2383
2384 if (strncmp(p, "/jre/lib/", 9) != 0) {
2385 // Look for JAVA_HOME in the environment.
2386 char* java_home_var = ::getenv("JAVA_HOME");
2387 if (java_home_var != NULL && java_home_var[0] != 0) {
2388 char* jrelib_p;
2389 int len;
2390
2391 // Check the current module name "libjvm.so".
2392 p = strrchr(buf, '/');
2393 assert(strstr(p, "/libjvm") == p, "invalid library name");
2394
2395 rp = realpath(java_home_var, buf);
2396 if (rp == NULL)
2397 return;
2398
2399 // determine if this is a legacy image or modules image
2400 // modules image doesn't have "jre" subdirectory
2401 len = strlen(buf);
2402 jrelib_p = buf + len;
2403 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2404 if (0 != access(buf, F_OK)) {
2405 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2406 }
2407
2408 if (0 == access(buf, F_OK)) {
2409 // Use current module name "libjvm.so"
2410 len = strlen(buf);
2411 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2412 } else {
2413 // Go back to path of .so
2414 rp = realpath(dli_fname, buf);
2415 if (rp == NULL)
2416 return;
2417 }
2418 }
2419 }
2420 }
2421
2422 strcpy(saved_jvm_path, buf);
2423 }
2424
2425 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2426 // no prefix required, not even "_"
2427 }
2428
2429 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2430 // no suffix required
2431 }
2432
2433 ////////////////////////////////////////////////////////////////////////////////
2434 // sun.misc.Signal support
2435
2436 static volatile jint sigint_count = 0;
2437
2438 static void
2439 UserHandler(int sig, void *siginfo, void *context) {
2440 // 4511530 - sem_post is serialized and handled by the manager thread. When
2441 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2442 // don't want to flood the manager thread with sem_post requests.
2443 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2444 return;
2445
2446 // Ctrl-C is pressed during error reporting, likely because the error
2447 // handler fails to abort. Let VM die immediately.
2448 if (sig == SIGINT && is_error_reported()) {
2449 os::die();
2450 }
2451
2452 os::signal_notify(sig);
2453 }
2454
2455 void* os::user_handler() {
2456 return CAST_FROM_FN_PTR(void*, UserHandler);
2457 }
2458
2459 class Semaphore : public StackObj {
2460 public:
2461 Semaphore();
2462 ~Semaphore();
2463 void signal();
2464 void wait();
2465 bool trywait();
2466 bool timedwait(unsigned int sec, int nsec);
2467 private:
2468 sem_t _semaphore;
2469 };
2470
2471
2472 Semaphore::Semaphore() {
2473 sem_init(&_semaphore, 0, 0);
2474 }
2475
2476 Semaphore::~Semaphore() {
2477 sem_destroy(&_semaphore);
2478 }
2479
2480 void Semaphore::signal() {
2481 sem_post(&_semaphore);
2482 }
2483
2484 void Semaphore::wait() {
2485 sem_wait(&_semaphore);
2486 }
2487
2488 bool Semaphore::trywait() {
2489 return sem_trywait(&_semaphore) == 0;
2490 }
2491
2492 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2493 struct timespec ts;
2494 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2495
2496 while (1) {
2497 int result = sem_timedwait(&_semaphore, &ts);
2498 if (result == 0) {
2499 return true;
2500 } else if (errno == EINTR) {
2501 continue;
2502 } else if (errno == ETIMEDOUT) {
2503 return false;
2504 } else {
2505 return false;
2506 }
2507 }
2508 }
2509
2510 extern "C" {
2511 typedef void (*sa_handler_t)(int);
2512 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2513 }
2514
2515 void* os::signal(int signal_number, void* handler) {
2516 struct sigaction sigAct, oldSigAct;
2517
2518 sigfillset(&(sigAct.sa_mask));
2519 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2520 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2521
2522 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2523 // -1 means registration failed
2524 return (void *)-1;
2525 }
2526
2527 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2528 }
2529
2530 void os::signal_raise(int signal_number) {
2531 ::raise(signal_number);
2532 }
2533
2534 /*
2535 * The following code is moved from os.cpp for making this
2536 * code platform specific, which it is by its very nature.
2537 */
2538
2539 // Will be modified when max signal is changed to be dynamic
2540 int os::sigexitnum_pd() {
2541 return NSIG;
2542 }
2543
2544 // a counter for each possible signal value
2545 static volatile jint pending_signals[NSIG+1] = { 0 };
2546
2547 // Linux(POSIX) specific hand shaking semaphore.
2548 static sem_t sig_sem;
2549 static Semaphore sr_semaphore;
2550
2551 void os::signal_init_pd() {
2552 // Initialize signal structures
2553 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2554
2555 // Initialize signal semaphore
2556 ::sem_init(&sig_sem, 0, 0);
2557 }
2558
2559 void os::signal_notify(int sig) {
2560 Atomic::inc(&pending_signals[sig]);
2561 ::sem_post(&sig_sem);
2562 }
2563
2564 static int check_pending_signals(bool wait) {
2565 Atomic::store(0, &sigint_count);
2566 for (;;) {
2567 for (int i = 0; i < NSIG + 1; i++) {
2568 jint n = pending_signals[i];
2569 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2570 return i;
2571 }
2572 }
2573 if (!wait) {
2574 return -1;
2575 }
2576 JavaThread *thread = JavaThread::current();
2577 ThreadBlockInVM tbivm(thread);
2578
2579 bool threadIsSuspended;
2580 do {
2581 thread->set_suspend_equivalent();
2582 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2583 ::sem_wait(&sig_sem);
2584
2585 // were we externally suspended while we were waiting?
2586 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2587 if (threadIsSuspended) {
2588 //
2589 // The semaphore has been incremented, but while we were waiting
2590 // another thread suspended us. We don't want to continue running
2591 // while suspended because that would surprise the thread that
2592 // suspended us.
2593 //
2594 ::sem_post(&sig_sem);
2595
2596 thread->java_suspend_self();
2597 }
2598 } while (threadIsSuspended);
2599 }
2600 }
2601
2602 int os::signal_lookup() {
2603 return check_pending_signals(false);
2604 }
2605
2606 int os::signal_wait() {
2607 return check_pending_signals(true);
2608 }
2609
2610 ////////////////////////////////////////////////////////////////////////////////
2611 // Virtual Memory
2612
2613 int os::vm_page_size() {
2614 // Seems redundant as all get out
2615 assert(os::Linux::page_size() != -1, "must call os::init");
2616 return os::Linux::page_size();
2617 }
2618
2619 // Solaris allocates memory by pages.
2620 int os::vm_allocation_granularity() {
2621 assert(os::Linux::page_size() != -1, "must call os::init");
2622 return os::Linux::page_size();
2623 }
2624
2625 // Rationale behind this function:
2626 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2627 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2628 // samples for JITted code. Here we create private executable mapping over the code cache
2629 // and then we can use standard (well, almost, as mapping can change) way to provide
2630 // info for the reporting script by storing timestamp and location of symbol
2631 void linux_wrap_code(char* base, size_t size) {
2632 static volatile jint cnt = 0;
2633
2634 if (!UseOprofile) {
2635 return;
2636 }
2637
2638 char buf[PATH_MAX+1];
2639 int num = Atomic::add(1, &cnt);
2640
2641 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2642 os::get_temp_directory(), os::current_process_id(), num);
2643 unlink(buf);
2644
2645 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2646
2647 if (fd != -1) {
2648 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2649 if (rv != (off_t)-1) {
2650 if (::write(fd, "", 1) == 1) {
2651 mmap(base, size,
2652 PROT_READ|PROT_WRITE|PROT_EXEC,
2653 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2654 }
2655 }
2656 ::close(fd);
2657 unlink(buf);
2658 }
2659 }
2660
2661 static bool recoverable_mmap_error(int err) {
2662 // See if the error is one we can let the caller handle. This
2663 // list of errno values comes from JBS-6843484. I can't find a
2664 // Linux man page that documents this specific set of errno
2665 // values so while this list currently matches Solaris, it may
2666 // change as we gain experience with this failure mode.
2667 switch (err) {
2668 case EBADF:
2669 case EINVAL:
2670 case ENOTSUP:
2671 // let the caller deal with these errors
2672 return true;
2673
2674 default:
2675 // Any remaining errors on this OS can cause our reserved mapping
2676 // to be lost. That can cause confusion where different data
2677 // structures think they have the same memory mapped. The worst
2678 // scenario is if both the VM and a library think they have the
2679 // same memory mapped.
2680 return false;
2681 }
2682 }
2683
2684 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2685 int err) {
2686 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2687 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2688 strerror(err), err);
2689 }
2690
2691 static void warn_fail_commit_memory(char* addr, size_t size,
2692 size_t alignment_hint, bool exec,
2693 int err) {
2694 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2695 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2696 alignment_hint, exec, strerror(err), err);
2697 }
2698
2699 // NOTE: Linux kernel does not really reserve the pages for us.
2700 // All it does is to check if there are enough free pages
2701 // left at the time of mmap(). This could be a potential
2702 // problem.
2703 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2704 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2705 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2706 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2707 if (res != (uintptr_t) MAP_FAILED) {
2708 if (UseNUMAInterleaving) {
2709 numa_make_global(addr, size);
2710 }
2711 return 0;
2712 }
2713
2714 int err = errno; // save errno from mmap() call above
2715
2716 if (!recoverable_mmap_error(err)) {
2717 warn_fail_commit_memory(addr, size, exec, err);
2718 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2719 }
2720
2721 return err;
2722 }
2723
2724 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2725 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2726 }
2727
2728 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2729 const char* mesg) {
2730 assert(mesg != NULL, "mesg must be specified");
2731 int err = os::Linux::commit_memory_impl(addr, size, exec);
2732 if (err != 0) {
2733 // the caller wants all commit errors to exit with the specified mesg:
2734 warn_fail_commit_memory(addr, size, exec, err);
2735 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2736 }
2737 }
2738
2739 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2740 #ifndef MAP_HUGETLB
2741 #define MAP_HUGETLB 0x40000
2742 #endif
2743
2744 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2745 #ifndef MADV_HUGEPAGE
2746 #define MADV_HUGEPAGE 14
2747 #endif
2748
2749 int os::Linux::commit_memory_impl(char* addr, size_t size,
2750 size_t alignment_hint, bool exec) {
2751 int err = os::Linux::commit_memory_impl(addr, size, exec);
2752 if (err == 0) {
2753 realign_memory(addr, size, alignment_hint);
2754 }
2755 return err;
2756 }
2757
2758 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2759 bool exec) {
2760 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2761 }
2762
2763 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2764 size_t alignment_hint, bool exec,
2765 const char* mesg) {
2766 assert(mesg != NULL, "mesg must be specified");
2767 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2768 if (err != 0) {
2769 // the caller wants all commit errors to exit with the specified mesg:
2770 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2771 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2772 }
2773 }
2774
2775 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2776 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2777 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2778 // be supported or the memory may already be backed by huge pages.
2779 ::madvise(addr, bytes, MADV_HUGEPAGE);
2780 }
2781 }
2782
2783 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2784 // This method works by doing an mmap over an existing mmaping and effectively discarding
2785 // the existing pages. However it won't work for SHM-based large pages that cannot be
2786 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2787 // small pages on top of the SHM segment. This method always works for small pages, so we
2788 // allow that in any case.
2789 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2790 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2791 }
2792 }
2793
2794 void os::numa_make_global(char *addr, size_t bytes) {
2795 Linux::numa_interleave_memory(addr, bytes);
2796 }
2797
2798 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2799 // bind policy to MPOL_PREFERRED for the current thread.
2800 #define USE_MPOL_PREFERRED 0
2801
2802 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2803 // To make NUMA and large pages more robust when both enabled, we need to ease
2804 // the requirements on where the memory should be allocated. MPOL_BIND is the
2805 // default policy and it will force memory to be allocated on the specified
2806 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2807 // the specified node, but will not force it. Using this policy will prevent
2808 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2809 // free large pages.
2810 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2811 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2812 }
2813
2814 bool os::numa_topology_changed() { return false; }
2815
2816 size_t os::numa_get_groups_num() {
2817 int max_node = Linux::numa_max_node();
2818 return max_node > 0 ? max_node + 1 : 1;
2819 }
2820
2821 int os::numa_get_group_id() {
2822 int cpu_id = Linux::sched_getcpu();
2823 if (cpu_id != -1) {
2824 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2825 if (lgrp_id != -1) {
2826 return lgrp_id;
2827 }
2828 }
2829 return 0;
2830 }
2831
2832 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2833 for (size_t i = 0; i < size; i++) {
2834 ids[i] = i;
2835 }
2836 return size;
2837 }
2838
2839 bool os::get_page_info(char *start, page_info* info) {
2840 return false;
2841 }
2842
2843 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2844 return end;
2845 }
2846
2847
2848 int os::Linux::sched_getcpu_syscall(void) {
2849 unsigned int cpu;
2850 int retval = -1;
2851
2852 #if defined(IA32)
2853 # ifndef SYS_getcpu
2854 # define SYS_getcpu 318
2855 # endif
2856 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2857 #elif defined(AMD64)
2858 // Unfortunately we have to bring all these macros here from vsyscall.h
2859 // to be able to compile on old linuxes.
2860 # define __NR_vgetcpu 2
2861 # define VSYSCALL_START (-10UL << 20)
2862 # define VSYSCALL_SIZE 1024
2863 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2864 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2865 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2866 retval = vgetcpu(&cpu, NULL, NULL);
2867 #endif
2868
2869 return (retval == -1) ? retval : cpu;
2870 }
2871
2872 // Something to do with the numa-aware allocator needs these symbols
2873 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2874 extern "C" JNIEXPORT void numa_error(char *where) { }
2875 extern "C" JNIEXPORT int fork1() { return fork(); }
2876
2877
2878 // If we are running with libnuma version > 2, then we should
2879 // be trying to use symbols with versions 1.1
2880 // If we are running with earlier version, which did not have symbol versions,
2881 // we should use the base version.
2882 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2883 void *f = dlvsym(handle, name, "libnuma_1.1");
2884 if (f == NULL) {
2885 f = dlsym(handle, name);
2886 }
2887 return f;
2888 }
2889
2890 bool os::Linux::libnuma_init() {
2891 // sched_getcpu() should be in libc.
2892 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2893 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2894
2895 // If it's not, try a direct syscall.
2896 if (sched_getcpu() == -1)
2897 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2898
2899 if (sched_getcpu() != -1) { // Does it work?
2900 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2901 if (handle != NULL) {
2902 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2903 libnuma_dlsym(handle, "numa_node_to_cpus")));
2904 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2905 libnuma_dlsym(handle, "numa_max_node")));
2906 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2907 libnuma_dlsym(handle, "numa_available")));
2908 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2909 libnuma_dlsym(handle, "numa_tonode_memory")));
2910 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2911 libnuma_dlsym(handle, "numa_interleave_memory")));
2912 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2913 libnuma_dlsym(handle, "numa_set_bind_policy")));
2914
2915
2916 if (numa_available() != -1) {
2917 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2918 // Create a cpu -> node mapping
2919 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2920 rebuild_cpu_to_node_map();
2921 return true;
2922 }
2923 }
2924 }
2925 return false;
2926 }
2927
2928 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2929 // The table is later used in get_node_by_cpu().
2930 void os::Linux::rebuild_cpu_to_node_map() {
2931 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2932 // in libnuma (possible values are starting from 16,
2933 // and continuing up with every other power of 2, but less
2934 // than the maximum number of CPUs supported by kernel), and
2935 // is a subject to change (in libnuma version 2 the requirements
2936 // are more reasonable) we'll just hardcode the number they use
2937 // in the library.
2938 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2939
2940 size_t cpu_num = os::active_processor_count();
2941 size_t cpu_map_size = NCPUS / BitsPerCLong;
2942 size_t cpu_map_valid_size =
2943 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2944
2945 cpu_to_node()->clear();
2946 cpu_to_node()->at_grow(cpu_num - 1);
2947 size_t node_num = numa_get_groups_num();
2948
2949 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2950 for (size_t i = 0; i < node_num; i++) {
2951 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2952 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2953 if (cpu_map[j] != 0) {
2954 for (size_t k = 0; k < BitsPerCLong; k++) {
2955 if (cpu_map[j] & (1UL << k)) {
2956 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2957 }
2958 }
2959 }
2960 }
2961 }
2962 }
2963 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2964 }
2965
2966 int os::Linux::get_node_by_cpu(int cpu_id) {
2967 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2968 return cpu_to_node()->at(cpu_id);
2969 }
2970 return -1;
2971 }
2972
2973 GrowableArray<int>* os::Linux::_cpu_to_node;
2974 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2975 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2976 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2977 os::Linux::numa_available_func_t os::Linux::_numa_available;
2978 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2979 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2980 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2981 unsigned long* os::Linux::_numa_all_nodes;
2982
2983 bool os::pd_uncommit_memory(char* addr, size_t size) {
2984 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2985 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2986 return res != (uintptr_t) MAP_FAILED;
2987 }
2988
2989 static
2990 address get_stack_commited_bottom(address bottom, size_t size) {
2991 address nbot = bottom;
2992 address ntop = bottom + size;
2993
2994 size_t page_sz = os::vm_page_size();
2995 unsigned pages = size / page_sz;
2996
2997 unsigned char vec[1];
2998 unsigned imin = 1, imax = pages + 1, imid;
2999 int mincore_return_value;
3000
3001 while (imin < imax) {
3002 imid = (imax + imin) / 2;
3003 nbot = ntop - (imid * page_sz);
3004
3005 // Use a trick with mincore to check whether the page is mapped or not.
3006 // mincore sets vec to 1 if page resides in memory and to 0 if page
3007 // is swapped output but if page we are asking for is unmapped
3008 // it returns -1,ENOMEM
3009 mincore_return_value = mincore(nbot, page_sz, vec);
3010
3011 if (mincore_return_value == -1) {
3012 // Page is not mapped go up
3013 // to find first mapped page
3014 if (errno != EAGAIN) {
3015 assert(errno == ENOMEM, "Unexpected mincore errno");
3016 imax = imid;
3017 }
3018 } else {
3019 // Page is mapped go down
3020 // to find first not mapped page
3021 imin = imid + 1;
3022 }
3023 }
3024
3025 nbot = nbot + page_sz;
3026
3027 // Adjust stack bottom one page up if last checked page is not mapped
3028 if (mincore_return_value == -1) {
3029 nbot = nbot + page_sz;
3030 }
3031
3032 return nbot;
3033 }
3034
3035
3036 // Linux uses a growable mapping for the stack, and if the mapping for
3037 // the stack guard pages is not removed when we detach a thread the
3038 // stack cannot grow beyond the pages where the stack guard was
3039 // mapped. If at some point later in the process the stack expands to
3040 // that point, the Linux kernel cannot expand the stack any further
3041 // because the guard pages are in the way, and a segfault occurs.
3042 //
3043 // However, it's essential not to split the stack region by unmapping
3044 // a region (leaving a hole) that's already part of the stack mapping,
3045 // so if the stack mapping has already grown beyond the guard pages at
3046 // the time we create them, we have to truncate the stack mapping.
3047 // So, we need to know the extent of the stack mapping when
3048 // create_stack_guard_pages() is called.
3049
3050 // We only need this for stacks that are growable: at the time of
3051 // writing thread stacks don't use growable mappings (i.e. those
3052 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3053 // only applies to the main thread.
3054
3055 // If the (growable) stack mapping already extends beyond the point
3056 // where we're going to put our guard pages, truncate the mapping at
3057 // that point by munmap()ping it. This ensures that when we later
3058 // munmap() the guard pages we don't leave a hole in the stack
3059 // mapping. This only affects the main/initial thread
3060
3061 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3062
3063 if (os::Linux::is_initial_thread()) {
3064 // As we manually grow stack up to bottom inside create_attached_thread(),
3065 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3066 // we don't need to do anything special.
3067 // Check it first, before calling heavy function.
3068 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3069 unsigned char vec[1];
3070
3071 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3072 // Fallback to slow path on all errors, including EAGAIN
3073 stack_extent = (uintptr_t) get_stack_commited_bottom(
3074 os::Linux::initial_thread_stack_bottom(),
3075 (size_t)addr - stack_extent);
3076 }
3077
3078 if (stack_extent < (uintptr_t)addr) {
3079 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3080 }
3081 }
3082
3083 return os::commit_memory(addr, size, !ExecMem);
3084 }
3085
3086 // If this is a growable mapping, remove the guard pages entirely by
3087 // munmap()ping them. If not, just call uncommit_memory(). This only
3088 // affects the main/initial thread, but guard against future OS changes
3089 // It's safe to always unmap guard pages for initial thread because we
3090 // always place it right after end of the mapped region
3091
3092 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3093 uintptr_t stack_extent, stack_base;
3094
3095 if (os::Linux::is_initial_thread()) {
3096 return ::munmap(addr, size) == 0;
3097 }
3098
3099 return os::uncommit_memory(addr, size);
3100 }
3101
3102 static address _highest_vm_reserved_address = NULL;
3103
3104 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3105 // at 'requested_addr'. If there are existing memory mappings at the same
3106 // location, however, they will be overwritten. If 'fixed' is false,
3107 // 'requested_addr' is only treated as a hint, the return value may or
3108 // may not start from the requested address. Unlike Linux mmap(), this
3109 // function returns NULL to indicate failure.
3110 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3111 char * addr;
3112 int flags;
3113
3114 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3115 if (fixed) {
3116 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3117 flags |= MAP_FIXED;
3118 }
3119
3120 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3121 // touch an uncommitted page. Otherwise, the read/write might
3122 // succeed if we have enough swap space to back the physical page.
3123 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3124 flags, -1, 0);
3125
3126 if (addr != MAP_FAILED) {
3127 // anon_mmap() should only get called during VM initialization,
3128 // don't need lock (actually we can skip locking even it can be called
3129 // from multiple threads, because _highest_vm_reserved_address is just a
3130 // hint about the upper limit of non-stack memory regions.)
3131 if ((address)addr + bytes > _highest_vm_reserved_address) {
3132 _highest_vm_reserved_address = (address)addr + bytes;
3133 }
3134 }
3135
3136 return addr == MAP_FAILED ? NULL : addr;
3137 }
3138
3139 // Don't update _highest_vm_reserved_address, because there might be memory
3140 // regions above addr + size. If so, releasing a memory region only creates
3141 // a hole in the address space, it doesn't help prevent heap-stack collision.
3142 //
3143 static int anon_munmap(char * addr, size_t size) {
3144 return ::munmap(addr, size) == 0;
3145 }
3146
3147 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3148 size_t alignment_hint) {
3149 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3150 }
3151
3152 bool os::pd_release_memory(char* addr, size_t size) {
3153 return anon_munmap(addr, size);
3154 }
3155
3156 static address highest_vm_reserved_address() {
3157 return _highest_vm_reserved_address;
3158 }
3159
3160 static bool linux_mprotect(char* addr, size_t size, int prot) {
3161 // Linux wants the mprotect address argument to be page aligned.
3162 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3163
3164 // According to SUSv3, mprotect() should only be used with mappings
3165 // established by mmap(), and mmap() always maps whole pages. Unaligned
3166 // 'addr' likely indicates problem in the VM (e.g. trying to change
3167 // protection of malloc'ed or statically allocated memory). Check the
3168 // caller if you hit this assert.
3169 assert(addr == bottom, "sanity check");
3170
3171 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3172 return ::mprotect(bottom, size, prot) == 0;
3173 }
3174
3175 // Set protections specified
3176 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3177 bool is_committed) {
3178 unsigned int p = 0;
3179 switch (prot) {
3180 case MEM_PROT_NONE: p = PROT_NONE; break;
3181 case MEM_PROT_READ: p = PROT_READ; break;
3182 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3183 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3184 default:
3185 ShouldNotReachHere();
3186 }
3187 // is_committed is unused.
3188 return linux_mprotect(addr, bytes, p);
3189 }
3190
3191 bool os::guard_memory(char* addr, size_t size) {
3192 return linux_mprotect(addr, size, PROT_NONE);
3193 }
3194
3195 bool os::unguard_memory(char* addr, size_t size) {
3196 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3197 }
3198
3199 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3200 bool result = false;
3201 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3202 MAP_ANONYMOUS|MAP_PRIVATE,
3203 -1, 0);
3204 if (p != MAP_FAILED) {
3205 void *aligned_p = align_ptr_up(p, page_size);
3206
3207 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3208
3209 munmap(p, page_size * 2);
3210 }
3211
3212 if (warn && !result) {
3213 warning("TransparentHugePages is not supported by the operating system.");
3214 }
3215
3216 return result;
3217 }
3218
3219 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3220 bool result = false;
3221 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3222 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3223 -1, 0);
3224
3225 if (p != MAP_FAILED) {
3226 // We don't know if this really is a huge page or not.
3227 FILE *fp = fopen("/proc/self/maps", "r");
3228 if (fp) {
3229 while (!feof(fp)) {
3230 char chars[257];
3231 long x = 0;
3232 if (fgets(chars, sizeof(chars), fp)) {
3233 if (sscanf(chars, "%lx-%*x", &x) == 1
3234 && x == (long)p) {
3235 if (strstr (chars, "hugepage")) {
3236 result = true;
3237 break;
3238 }
3239 }
3240 }
3241 }
3242 fclose(fp);
3243 }
3244 munmap(p, page_size);
3245 }
3246
3247 if (warn && !result) {
3248 warning("HugeTLBFS is not supported by the operating system.");
3249 }
3250
3251 return result;
3252 }
3253
3254 /*
3255 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3256 *
3257 * From the coredump_filter documentation:
3258 *
3259 * - (bit 0) anonymous private memory
3260 * - (bit 1) anonymous shared memory
3261 * - (bit 2) file-backed private memory
3262 * - (bit 3) file-backed shared memory
3263 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3264 * effective only if the bit 2 is cleared)
3265 * - (bit 5) hugetlb private memory
3266 * - (bit 6) hugetlb shared memory
3267 */
3268 static void set_coredump_filter(void) {
3269 FILE *f;
3270 long cdm;
3271
3272 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3273 return;
3274 }
3275
3276 if (fscanf(f, "%lx", &cdm) != 1) {
3277 fclose(f);
3278 return;
3279 }
3280
3281 rewind(f);
3282
3283 if ((cdm & LARGEPAGES_BIT) == 0) {
3284 cdm |= LARGEPAGES_BIT;
3285 fprintf(f, "%#lx", cdm);
3286 }
3287
3288 fclose(f);
3289 }
3290
3291 // Large page support
3292
3293 static size_t _large_page_size = 0;
3294
3295 size_t os::Linux::find_large_page_size() {
3296 size_t large_page_size = 0;
3297
3298 // large_page_size on Linux is used to round up heap size. x86 uses either
3299 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3300 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3301 // page as large as 256M.
3302 //
3303 // Here we try to figure out page size by parsing /proc/meminfo and looking
3304 // for a line with the following format:
3305 // Hugepagesize: 2048 kB
3306 //
3307 // If we can't determine the value (e.g. /proc is not mounted, or the text
3308 // format has been changed), we'll use the largest page size supported by
3309 // the processor.
3310
3311 #ifndef ZERO
3312 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3313 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3314 #endif // ZERO
3315
3316 FILE *fp = fopen("/proc/meminfo", "r");
3317 if (fp) {
3318 while (!feof(fp)) {
3319 int x = 0;
3320 char buf[16];
3321 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3322 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3323 large_page_size = x * K;
3324 break;
3325 }
3326 } else {
3327 // skip to next line
3328 for (;;) {
3329 int ch = fgetc(fp);
3330 if (ch == EOF || ch == (int)'\n') break;
3331 }
3332 }
3333 }
3334 fclose(fp);
3335 }
3336
3337 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3338 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3339 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3340 proper_unit_for_byte_size(large_page_size));
3341 }
3342
3343 return large_page_size;
3344 }
3345
3346 size_t os::Linux::setup_large_page_size() {
3347 _large_page_size = Linux::find_large_page_size();
3348 const size_t default_page_size = (size_t)Linux::page_size();
3349 if (_large_page_size > default_page_size) {
3350 _page_sizes[0] = _large_page_size;
3351 _page_sizes[1] = default_page_size;
3352 _page_sizes[2] = 0;
3353 }
3354
3355 return _large_page_size;
3356 }
3357
3358 bool os::Linux::setup_large_page_type(size_t page_size) {
3359 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3360 FLAG_IS_DEFAULT(UseSHM) &&
3361 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3362
3363 // The type of large pages has not been specified by the user.
3364
3365 // Try UseHugeTLBFS and then UseSHM.
3366 UseHugeTLBFS = UseSHM = true;
3367
3368 // Don't try UseTransparentHugePages since there are known
3369 // performance issues with it turned on. This might change in the future.
3370 UseTransparentHugePages = false;
3371 }
3372
3373 if (UseTransparentHugePages) {
3374 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3375 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3376 UseHugeTLBFS = false;
3377 UseSHM = false;
3378 return true;
3379 }
3380 UseTransparentHugePages = false;
3381 }
3382
3383 if (UseHugeTLBFS) {
3384 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3385 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3386 UseSHM = false;
3387 return true;
3388 }
3389 UseHugeTLBFS = false;
3390 }
3391
3392 return UseSHM;
3393 }
3394
3395 void os::large_page_init() {
3396 if (!UseLargePages &&
3397 !UseTransparentHugePages &&
3398 !UseHugeTLBFS &&
3399 !UseSHM) {
3400 // Not using large pages.
3401 return;
3402 }
3403
3404 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3405 // The user explicitly turned off large pages.
3406 // Ignore the rest of the large pages flags.
3407 UseTransparentHugePages = false;
3408 UseHugeTLBFS = false;
3409 UseSHM = false;
3410 return;
3411 }
3412
3413 size_t large_page_size = Linux::setup_large_page_size();
3414 UseLargePages = Linux::setup_large_page_type(large_page_size);
3415
3416 set_coredump_filter();
3417 }
3418
3419 #ifndef SHM_HUGETLB
3420 #define SHM_HUGETLB 04000
3421 #endif
3422
3423 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3424 // "exec" is passed in but not used. Creating the shared image for
3425 // the code cache doesn't have an SHM_X executable permission to check.
3426 assert(UseLargePages && UseSHM, "only for SHM large pages");
3427 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3428
3429 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3430 return NULL; // Fallback to small pages.
3431 }
3432
3433 key_t key = IPC_PRIVATE;
3434 char *addr;
3435
3436 bool warn_on_failure = UseLargePages &&
3437 (!FLAG_IS_DEFAULT(UseLargePages) ||
3438 !FLAG_IS_DEFAULT(UseSHM) ||
3439 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3440 );
3441 char msg[128];
3442
3443 // Create a large shared memory region to attach to based on size.
3444 // Currently, size is the total size of the heap
3445 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3446 if (shmid == -1) {
3447 // Possible reasons for shmget failure:
3448 // 1. shmmax is too small for Java heap.
3449 // > check shmmax value: cat /proc/sys/kernel/shmmax
3450 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3451 // 2. not enough large page memory.
3452 // > check available large pages: cat /proc/meminfo
3453 // > increase amount of large pages:
3454 // echo new_value > /proc/sys/vm/nr_hugepages
3455 // Note 1: different Linux may use different name for this property,
3456 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3457 // Note 2: it's possible there's enough physical memory available but
3458 // they are so fragmented after a long run that they can't
3459 // coalesce into large pages. Try to reserve large pages when
3460 // the system is still "fresh".
3461 if (warn_on_failure) {
3462 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3463 warning(msg);
3464 }
3465 return NULL;
3466 }
3467
3468 // attach to the region
3469 addr = (char*)shmat(shmid, req_addr, 0);
3470 int err = errno;
3471
3472 // Remove shmid. If shmat() is successful, the actual shared memory segment
3473 // will be deleted when it's detached by shmdt() or when the process
3474 // terminates. If shmat() is not successful this will remove the shared
3475 // segment immediately.
3476 shmctl(shmid, IPC_RMID, NULL);
3477
3478 if ((intptr_t)addr == -1) {
3479 if (warn_on_failure) {
3480 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3481 warning(msg);
3482 }
3483 return NULL;
3484 }
3485
3486 return addr;
3487 }
3488
3489 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3490 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3491
3492 bool warn_on_failure = UseLargePages &&
3493 (!FLAG_IS_DEFAULT(UseLargePages) ||
3494 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3495 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3496
3497 if (warn_on_failure) {
3498 char msg[128];
3499 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3500 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3501 warning(msg);
3502 }
3503 }
3504
3505 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3506 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3507 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3508 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3509
3510 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3511 char* addr = (char*)::mmap(req_addr, bytes, prot,
3512 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3513 -1, 0);
3514
3515 if (addr == MAP_FAILED) {
3516 warn_on_large_pages_failure(req_addr, bytes, errno);
3517 return NULL;
3518 }
3519
3520 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3521
3522 return addr;
3523 }
3524
3525 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3526 size_t large_page_size = os::large_page_size();
3527
3528 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3529
3530 // Allocate small pages.
3531
3532 char* start;
3533 if (req_addr != NULL) {
3534 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3535 assert(is_size_aligned(bytes, alignment), "Must be");
3536 start = os::reserve_memory(bytes, req_addr);
3537 assert(start == NULL || start == req_addr, "Must be");
3538 } else {
3539 start = os::reserve_memory_aligned(bytes, alignment);
3540 }
3541
3542 if (start == NULL) {
3543 return NULL;
3544 }
3545
3546 assert(is_ptr_aligned(start, alignment), "Must be");
3547
3548 // os::reserve_memory_special will record this memory area.
3549 // Need to release it here to prevent overlapping reservations.
3550 MemTracker::record_virtual_memory_release((address)start, bytes);
3551
3552 char* end = start + bytes;
3553
3554 // Find the regions of the allocated chunk that can be promoted to large pages.
3555 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3556 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3557
3558 size_t lp_bytes = lp_end - lp_start;
3559
3560 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3561
3562 if (lp_bytes == 0) {
3563 // The mapped region doesn't even span the start and the end of a large page.
3564 // Fall back to allocate a non-special area.
3565 ::munmap(start, end - start);
3566 return NULL;
3567 }
3568
3569 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3570
3571
3572 void* result;
3573
3574 if (start != lp_start) {
3575 result = ::mmap(start, lp_start - start, prot,
3576 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3577 -1, 0);
3578 if (result == MAP_FAILED) {
3579 ::munmap(lp_start, end - lp_start);
3580 return NULL;
3581 }
3582 }
3583
3584 result = ::mmap(lp_start, lp_bytes, prot,
3585 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3586 -1, 0);
3587 if (result == MAP_FAILED) {
3588 warn_on_large_pages_failure(req_addr, bytes, errno);
3589 // If the mmap above fails, the large pages region will be unmapped and we
3590 // have regions before and after with small pages. Release these regions.
3591 //
3592 // | mapped | unmapped | mapped |
3593 // ^ ^ ^ ^
3594 // start lp_start lp_end end
3595 //
3596 ::munmap(start, lp_start - start);
3597 ::munmap(lp_end, end - lp_end);
3598 return NULL;
3599 }
3600
3601 if (lp_end != end) {
3602 result = ::mmap(lp_end, end - lp_end, prot,
3603 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3604 -1, 0);
3605 if (result == MAP_FAILED) {
3606 ::munmap(start, lp_end - start);
3607 return NULL;
3608 }
3609 }
3610
3611 return start;
3612 }
3613
3614 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3615 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3616 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3617 assert(is_power_of_2(alignment), "Must be");
3618 assert(is_power_of_2(os::large_page_size()), "Must be");
3619 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3620
3621 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3622 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3623 } else {
3624 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3625 }
3626 }
3627
3628 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3629 assert(UseLargePages, "only for large pages");
3630
3631 char* addr;
3632 if (UseSHM) {
3633 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3634 } else {
3635 assert(UseHugeTLBFS, "must be");
3636 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3637 }
3638
3639 if (addr != NULL) {
3640 if (UseNUMAInterleaving) {
3641 numa_make_global(addr, bytes);
3642 }
3643
3644 // The memory is committed
3645 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3646 }
3647
3648 return addr;
3649 }
3650
3651 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3652 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3653 return shmdt(base) == 0;
3654 }
3655
3656 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3657 return pd_release_memory(base, bytes);
3658 }
3659
3660 bool os::release_memory_special(char* base, size_t bytes) {
3661 assert(UseLargePages, "only for large pages");
3662
3663 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3664
3665 bool res;
3666 if (UseSHM) {
3667 res = os::Linux::release_memory_special_shm(base, bytes);
3668 } else {
3669 assert(UseHugeTLBFS, "must be");
3670 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3671 }
3672
3673 if (res) {
3674 tkr.record((address)base, bytes);
3675 } else {
3676 tkr.discard();
3677 }
3678
3679 return res;
3680 }
3681
3682 size_t os::large_page_size() {
3683 return _large_page_size;
3684 }
3685
3686 // With SysV SHM the entire memory region must be allocated as shared
3687 // memory.
3688 // HugeTLBFS allows application to commit large page memory on demand.
3689 // However, when committing memory with HugeTLBFS fails, the region
3690 // that was supposed to be committed will lose the old reservation
3691 // and allow other threads to steal that memory region. Because of this
3692 // behavior we can't commit HugeTLBFS memory.
3693 bool os::can_commit_large_page_memory() {
3694 return UseTransparentHugePages;
3695 }
3696
3697 bool os::can_execute_large_page_memory() {
3698 return UseTransparentHugePages || UseHugeTLBFS;
3699 }
3700
3701 // Reserve memory at an arbitrary address, only if that area is
3702 // available (and not reserved for something else).
3703
3704 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3705 const int max_tries = 10;
3706 char* base[max_tries];
3707 size_t size[max_tries];
3708 const size_t gap = 0x000000;
3709
3710 // Assert only that the size is a multiple of the page size, since
3711 // that's all that mmap requires, and since that's all we really know
3712 // about at this low abstraction level. If we need higher alignment,
3713 // we can either pass an alignment to this method or verify alignment
3714 // in one of the methods further up the call chain. See bug 5044738.
3715 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3716
3717 // Repeatedly allocate blocks until the block is allocated at the
3718 // right spot. Give up after max_tries. Note that reserve_memory() will
3719 // automatically update _highest_vm_reserved_address if the call is
3720 // successful. The variable tracks the highest memory address every reserved
3721 // by JVM. It is used to detect heap-stack collision if running with
3722 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3723 // space than needed, it could confuse the collision detecting code. To
3724 // solve the problem, save current _highest_vm_reserved_address and
3725 // calculate the correct value before return.
3726 address old_highest = _highest_vm_reserved_address;
3727
3728 // Linux mmap allows caller to pass an address as hint; give it a try first,
3729 // if kernel honors the hint then we can return immediately.
3730 char * addr = anon_mmap(requested_addr, bytes, false);
3731 if (addr == requested_addr) {
3732 return requested_addr;
3733 }
3734
3735 if (addr != NULL) {
3736 // mmap() is successful but it fails to reserve at the requested address
3737 anon_munmap(addr, bytes);
3738 }
3739
3740 int i;
3741 for (i = 0; i < max_tries; ++i) {
3742 base[i] = reserve_memory(bytes);
3743
3744 if (base[i] != NULL) {
3745 // Is this the block we wanted?
3746 if (base[i] == requested_addr) {
3747 size[i] = bytes;
3748 break;
3749 }
3750
3751 // Does this overlap the block we wanted? Give back the overlapped
3752 // parts and try again.
3753
3754 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3755 if (top_overlap >= 0 && top_overlap < bytes) {
3756 unmap_memory(base[i], top_overlap);
3757 base[i] += top_overlap;
3758 size[i] = bytes - top_overlap;
3759 } else {
3760 size_t bottom_overlap = base[i] + bytes - requested_addr;
3761 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3762 unmap_memory(requested_addr, bottom_overlap);
3763 size[i] = bytes - bottom_overlap;
3764 } else {
3765 size[i] = bytes;
3766 }
3767 }
3768 }
3769 }
3770
3771 // Give back the unused reserved pieces.
3772
3773 for (int j = 0; j < i; ++j) {
3774 if (base[j] != NULL) {
3775 unmap_memory(base[j], size[j]);
3776 }
3777 }
3778
3779 if (i < max_tries) {
3780 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3781 return requested_addr;
3782 } else {
3783 _highest_vm_reserved_address = old_highest;
3784 return NULL;
3785 }
3786 }
3787
3788 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3789 return ::read(fd, buf, nBytes);
3790 }
3791
3792 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3793 // Solaris uses poll(), linux uses park().
3794 // Poll() is likely a better choice, assuming that Thread.interrupt()
3795 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3796 // SIGSEGV, see 4355769.
3797
3798 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3799 assert(thread == Thread::current(), "thread consistency check");
3800
3801 ParkEvent * const slp = thread->_SleepEvent ;
3802 slp->reset() ;
3803 OrderAccess::fence() ;
3804
3805 if (interruptible) {
3806 jlong prevtime = javaTimeNanos();
3807
3808 for (;;) {
3809 if (os::is_interrupted(thread, true)) {
3810 return OS_INTRPT;
3811 }
3812
3813 jlong newtime = javaTimeNanos();
3814
3815 if (newtime - prevtime < 0) {
3816 // time moving backwards, should only happen if no monotonic clock
3817 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3818 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3819 } else {
3820 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3821 }
3822
3823 if(millis <= 0) {
3824 return OS_OK;
3825 }
3826
3827 prevtime = newtime;
3828
3829 {
3830 assert(thread->is_Java_thread(), "sanity check");
3831 JavaThread *jt = (JavaThread *) thread;
3832 ThreadBlockInVM tbivm(jt);
3833 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3834
3835 jt->set_suspend_equivalent();
3836 // cleared by handle_special_suspend_equivalent_condition() or
3837 // java_suspend_self() via check_and_wait_while_suspended()
3838
3839 slp->park(millis);
3840
3841 // were we externally suspended while we were waiting?
3842 jt->check_and_wait_while_suspended();
3843 }
3844 }
3845 } else {
3846 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3847 jlong prevtime = javaTimeNanos();
3848
3849 for (;;) {
3850 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3851 // the 1st iteration ...
3852 jlong newtime = javaTimeNanos();
3853
3854 if (newtime - prevtime < 0) {
3855 // time moving backwards, should only happen if no monotonic clock
3856 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3857 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3858 } else {
3859 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3860 }
3861
3862 if(millis <= 0) break ;
3863
3864 prevtime = newtime;
3865 slp->park(millis);
3866 }
3867 return OS_OK ;
3868 }
3869 }
3870
3871 //
3872 // Short sleep, direct OS call.
3873 //
3874 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3875 // sched_yield(2) will actually give up the CPU:
3876 //
3877 // * Alone on this pariticular CPU, keeps running.
3878 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3879 // (pre 2.6.39).
3880 //
3881 // So calling this with 0 is an alternative.
3882 //
3883 void os::naked_short_sleep(jlong ms) {
3884 struct timespec req;
3885
3886 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3887 req.tv_sec = 0;
3888 if (ms > 0) {
3889 req.tv_nsec = (ms % 1000) * 1000000;
3890 }
3891 else {
3892 req.tv_nsec = 1;
3893 }
3894
3895 nanosleep(&req, NULL);
3896
3897 return;
3898 }
3899
3900 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3901 void os::infinite_sleep() {
3902 while (true) { // sleep forever ...
3903 ::sleep(100); // ... 100 seconds at a time
3904 }
3905 }
3906
3907 // Used to convert frequent JVM_Yield() to nops
3908 bool os::dont_yield() {
3909 return DontYieldALot;
3910 }
3911
3912 void os::yield() {
3913 sched_yield();
3914 }
3915
3916 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3917
3918 void os::yield_all(int attempts) {
3919 // Yields to all threads, including threads with lower priorities
3920 // Threads on Linux are all with same priority. The Solaris style
3921 // os::yield_all() with nanosleep(1ms) is not necessary.
3922 sched_yield();
3923 }
3924
3925 // Called from the tight loops to possibly influence time-sharing heuristics
3926 void os::loop_breaker(int attempts) {
3927 os::yield_all(attempts);
3928 }
3929
3930 ////////////////////////////////////////////////////////////////////////////////
3931 // thread priority support
3932
3933 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3934 // only supports dynamic priority, static priority must be zero. For real-time
3935 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3936 // However, for large multi-threaded applications, SCHED_RR is not only slower
3937 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3938 // of 5 runs - Sep 2005).
3939 //
3940 // The following code actually changes the niceness of kernel-thread/LWP. It
3941 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3942 // not the entire user process, and user level threads are 1:1 mapped to kernel
3943 // threads. It has always been the case, but could change in the future. For
3944 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3945 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3946
3947 int os::java_to_os_priority[CriticalPriority + 1] = {
3948 19, // 0 Entry should never be used
3949
3950 4, // 1 MinPriority
3951 3, // 2
3952 2, // 3
3953
3954 1, // 4
3955 0, // 5 NormPriority
3956 -1, // 6
3957
3958 -2, // 7
3959 -3, // 8
3960 -4, // 9 NearMaxPriority
3961
3962 -5, // 10 MaxPriority
3963
3964 -5 // 11 CriticalPriority
3965 };
3966
3967 static int prio_init() {
3968 if (ThreadPriorityPolicy == 1) {
3969 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3970 // if effective uid is not root. Perhaps, a more elegant way of doing
3971 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3972 if (geteuid() != 0) {
3973 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3974 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3975 }
3976 ThreadPriorityPolicy = 0;
3977 }
3978 }
3979 if (UseCriticalJavaThreadPriority) {
3980 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3981 }
3982 return 0;
3983 }
3984
3985 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3986 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3987
3988 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3989 return (ret == 0) ? OS_OK : OS_ERR;
3990 }
3991
3992 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3993 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3994 *priority_ptr = java_to_os_priority[NormPriority];
3995 return OS_OK;
3996 }
3997
3998 errno = 0;
3999 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4000 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4001 }
4002
4003 // Hint to the underlying OS that a task switch would not be good.
4004 // Void return because it's a hint and can fail.
4005 void os::hint_no_preempt() {}
4006
4007 ////////////////////////////////////////////////////////////////////////////////
4008 // suspend/resume support
4009
4010 // the low-level signal-based suspend/resume support is a remnant from the
4011 // old VM-suspension that used to be for java-suspension, safepoints etc,
4012 // within hotspot. Now there is a single use-case for this:
4013 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4014 // that runs in the watcher thread.
4015 // The remaining code is greatly simplified from the more general suspension
4016 // code that used to be used.
4017 //
4018 // The protocol is quite simple:
4019 // - suspend:
4020 // - sends a signal to the target thread
4021 // - polls the suspend state of the osthread using a yield loop
4022 // - target thread signal handler (SR_handler) sets suspend state
4023 // and blocks in sigsuspend until continued
4024 // - resume:
4025 // - sets target osthread state to continue
4026 // - sends signal to end the sigsuspend loop in the SR_handler
4027 //
4028 // Note that the SR_lock plays no role in this suspend/resume protocol.
4029 //
4030
4031 static void resume_clear_context(OSThread *osthread) {
4032 osthread->set_ucontext(NULL);
4033 osthread->set_siginfo(NULL);
4034 }
4035
4036 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4037 osthread->set_ucontext(context);
4038 osthread->set_siginfo(siginfo);
4039 }
4040
4041 //
4042 // Handler function invoked when a thread's execution is suspended or
4043 // resumed. We have to be careful that only async-safe functions are
4044 // called here (Note: most pthread functions are not async safe and
4045 // should be avoided.)
4046 //
4047 // Note: sigwait() is a more natural fit than sigsuspend() from an
4048 // interface point of view, but sigwait() prevents the signal hander
4049 // from being run. libpthread would get very confused by not having
4050 // its signal handlers run and prevents sigwait()'s use with the
4051 // mutex granting granting signal.
4052 //
4053 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4054 //
4055 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4056 // Save and restore errno to avoid confusing native code with EINTR
4057 // after sigsuspend.
4058 int old_errno = errno;
4059
4060 Thread* thread = Thread::current();
4061 OSThread* osthread = thread->osthread();
4062 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4063
4064 os::SuspendResume::State current = osthread->sr.state();
4065 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4066 suspend_save_context(osthread, siginfo, context);
4067
4068 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4069 os::SuspendResume::State state = osthread->sr.suspended();
4070 if (state == os::SuspendResume::SR_SUSPENDED) {
4071 sigset_t suspend_set; // signals for sigsuspend()
4072
4073 // get current set of blocked signals and unblock resume signal
4074 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4075 sigdelset(&suspend_set, SR_signum);
4076
4077 sr_semaphore.signal();
4078 // wait here until we are resumed
4079 while (1) {
4080 sigsuspend(&suspend_set);
4081
4082 os::SuspendResume::State result = osthread->sr.running();
4083 if (result == os::SuspendResume::SR_RUNNING) {
4084 sr_semaphore.signal();
4085 break;
4086 }
4087 }
4088
4089 } else if (state == os::SuspendResume::SR_RUNNING) {
4090 // request was cancelled, continue
4091 } else {
4092 ShouldNotReachHere();
4093 }
4094
4095 resume_clear_context(osthread);
4096 } else if (current == os::SuspendResume::SR_RUNNING) {
4097 // request was cancelled, continue
4098 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4099 // ignore
4100 } else {
4101 // ignore
4102 }
4103
4104 errno = old_errno;
4105 }
4106
4107
4108 static int SR_initialize() {
4109 struct sigaction act;
4110 char *s;
4111 /* Get signal number to use for suspend/resume */
4112 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4113 int sig = ::strtol(s, 0, 10);
4114 if (sig > 0 || sig < _NSIG) {
4115 SR_signum = sig;
4116 }
4117 }
4118
4119 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4120 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4121
4122 sigemptyset(&SR_sigset);
4123 sigaddset(&SR_sigset, SR_signum);
4124
4125 /* Set up signal handler for suspend/resume */
4126 act.sa_flags = SA_RESTART|SA_SIGINFO;
4127 act.sa_handler = (void (*)(int)) SR_handler;
4128
4129 // SR_signum is blocked by default.
4130 // 4528190 - We also need to block pthread restart signal (32 on all
4131 // supported Linux platforms). Note that LinuxThreads need to block
4132 // this signal for all threads to work properly. So we don't have
4133 // to use hard-coded signal number when setting up the mask.
4134 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4135
4136 if (sigaction(SR_signum, &act, 0) == -1) {
4137 return -1;
4138 }
4139
4140 // Save signal flag
4141 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4142 return 0;
4143 }
4144
4145 static int sr_notify(OSThread* osthread) {
4146 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4147 assert_status(status == 0, status, "pthread_kill");
4148 return status;
4149 }
4150
4151 // "Randomly" selected value for how long we want to spin
4152 // before bailing out on suspending a thread, also how often
4153 // we send a signal to a thread we want to resume
4154 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4155 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4156
4157 // returns true on success and false on error - really an error is fatal
4158 // but this seems the normal response to library errors
4159 static bool do_suspend(OSThread* osthread) {
4160 assert(osthread->sr.is_running(), "thread should be running");
4161 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4162
4163 // mark as suspended and send signal
4164 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4165 // failed to switch, state wasn't running?
4166 ShouldNotReachHere();
4167 return false;
4168 }
4169
4170 if (sr_notify(osthread) != 0) {
4171 ShouldNotReachHere();
4172 }
4173
4174 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4175 while (true) {
4176 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4177 break;
4178 } else {
4179 // timeout
4180 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4181 if (cancelled == os::SuspendResume::SR_RUNNING) {
4182 return false;
4183 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4184 // make sure that we consume the signal on the semaphore as well
4185 sr_semaphore.wait();
4186 break;
4187 } else {
4188 ShouldNotReachHere();
4189 return false;
4190 }
4191 }
4192 }
4193
4194 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4195 return true;
4196 }
4197
4198 static void do_resume(OSThread* osthread) {
4199 assert(osthread->sr.is_suspended(), "thread should be suspended");
4200 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4201
4202 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4203 // failed to switch to WAKEUP_REQUEST
4204 ShouldNotReachHere();
4205 return;
4206 }
4207
4208 while (true) {
4209 if (sr_notify(osthread) == 0) {
4210 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4211 if (osthread->sr.is_running()) {
4212 return;
4213 }
4214 }
4215 } else {
4216 ShouldNotReachHere();
4217 }
4218 }
4219
4220 guarantee(osthread->sr.is_running(), "Must be running!");
4221 }
4222
4223 ////////////////////////////////////////////////////////////////////////////////
4224 // interrupt support
4225
4226 void os::interrupt(Thread* thread) {
4227 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4228 "possibility of dangling Thread pointer");
4229
4230 OSThread* osthread = thread->osthread();
4231
4232 if (!osthread->interrupted()) {
4233 osthread->set_interrupted(true);
4234 // More than one thread can get here with the same value of osthread,
4235 // resulting in multiple notifications. We do, however, want the store
4236 // to interrupted() to be visible to other threads before we execute unpark().
4237 OrderAccess::fence();
4238 ParkEvent * const slp = thread->_SleepEvent ;
4239 if (slp != NULL) slp->unpark() ;
4240 }
4241
4242 // For JSR166. Unpark even if interrupt status already was set
4243 if (thread->is_Java_thread())
4244 ((JavaThread*)thread)->parker()->unpark();
4245
4246 ParkEvent * ev = thread->_ParkEvent ;
4247 if (ev != NULL) ev->unpark() ;
4248
4249 }
4250
4251 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4252 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4253 "possibility of dangling Thread pointer");
4254
4255 OSThread* osthread = thread->osthread();
4256
4257 bool interrupted = osthread->interrupted();
4258
4259 if (interrupted && clear_interrupted) {
4260 osthread->set_interrupted(false);
4261 // consider thread->_SleepEvent->reset() ... optional optimization
4262 }
4263
4264 return interrupted;
4265 }
4266
4267 ///////////////////////////////////////////////////////////////////////////////////
4268 // signal handling (except suspend/resume)
4269
4270 // This routine may be used by user applications as a "hook" to catch signals.
4271 // The user-defined signal handler must pass unrecognized signals to this
4272 // routine, and if it returns true (non-zero), then the signal handler must
4273 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4274 // routine will never retun false (zero), but instead will execute a VM panic
4275 // routine kill the process.
4276 //
4277 // If this routine returns false, it is OK to call it again. This allows
4278 // the user-defined signal handler to perform checks either before or after
4279 // the VM performs its own checks. Naturally, the user code would be making
4280 // a serious error if it tried to handle an exception (such as a null check
4281 // or breakpoint) that the VM was generating for its own correct operation.
4282 //
4283 // This routine may recognize any of the following kinds of signals:
4284 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4285 // It should be consulted by handlers for any of those signals.
4286 //
4287 // The caller of this routine must pass in the three arguments supplied
4288 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4289 // field of the structure passed to sigaction(). This routine assumes that
4290 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4291 //
4292 // Note that the VM will print warnings if it detects conflicting signal
4293 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4294 //
4295 extern "C" JNIEXPORT int
4296 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4297 void* ucontext, int abort_if_unrecognized);
4298
4299 void signalHandler(int sig, siginfo_t* info, void* uc) {
4300 assert(info != NULL && uc != NULL, "it must be old kernel");
4301 int orig_errno = errno; // Preserve errno value over signal handler.
4302 JVM_handle_linux_signal(sig, info, uc, true);
4303 errno = orig_errno;
4304 }
4305
4306
4307 // This boolean allows users to forward their own non-matching signals
4308 // to JVM_handle_linux_signal, harmlessly.
4309 bool os::Linux::signal_handlers_are_installed = false;
4310
4311 // For signal-chaining
4312 struct sigaction os::Linux::sigact[MAXSIGNUM];
4313 unsigned int os::Linux::sigs = 0;
4314 bool os::Linux::libjsig_is_loaded = false;
4315 typedef struct sigaction *(*get_signal_t)(int);
4316 get_signal_t os::Linux::get_signal_action = NULL;
4317
4318 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4319 struct sigaction *actp = NULL;
4320
4321 if (libjsig_is_loaded) {
4322 // Retrieve the old signal handler from libjsig
4323 actp = (*get_signal_action)(sig);
4324 }
4325 if (actp == NULL) {
4326 // Retrieve the preinstalled signal handler from jvm
4327 actp = get_preinstalled_handler(sig);
4328 }
4329
4330 return actp;
4331 }
4332
4333 static bool call_chained_handler(struct sigaction *actp, int sig,
4334 siginfo_t *siginfo, void *context) {
4335 // Call the old signal handler
4336 if (actp->sa_handler == SIG_DFL) {
4337 // It's more reasonable to let jvm treat it as an unexpected exception
4338 // instead of taking the default action.
4339 return false;
4340 } else if (actp->sa_handler != SIG_IGN) {
4341 if ((actp->sa_flags & SA_NODEFER) == 0) {
4342 // automaticlly block the signal
4343 sigaddset(&(actp->sa_mask), sig);
4344 }
4345
4346 sa_handler_t hand;
4347 sa_sigaction_t sa;
4348 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4349 // retrieve the chained handler
4350 if (siginfo_flag_set) {
4351 sa = actp->sa_sigaction;
4352 } else {
4353 hand = actp->sa_handler;
4354 }
4355
4356 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4357 actp->sa_handler = SIG_DFL;
4358 }
4359
4360 // try to honor the signal mask
4361 sigset_t oset;
4362 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4363
4364 // call into the chained handler
4365 if (siginfo_flag_set) {
4366 (*sa)(sig, siginfo, context);
4367 } else {
4368 (*hand)(sig);
4369 }
4370
4371 // restore the signal mask
4372 pthread_sigmask(SIG_SETMASK, &oset, 0);
4373 }
4374 // Tell jvm's signal handler the signal is taken care of.
4375 return true;
4376 }
4377
4378 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4379 bool chained = false;
4380 // signal-chaining
4381 if (UseSignalChaining) {
4382 struct sigaction *actp = get_chained_signal_action(sig);
4383 if (actp != NULL) {
4384 chained = call_chained_handler(actp, sig, siginfo, context);
4385 }
4386 }
4387 return chained;
4388 }
4389
4390 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4391 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4392 return &sigact[sig];
4393 }
4394 return NULL;
4395 }
4396
4397 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4398 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4399 sigact[sig] = oldAct;
4400 sigs |= (unsigned int)1 << sig;
4401 }
4402
4403 // for diagnostic
4404 int os::Linux::sigflags[MAXSIGNUM];
4405
4406 int os::Linux::get_our_sigflags(int sig) {
4407 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4408 return sigflags[sig];
4409 }
4410
4411 void os::Linux::set_our_sigflags(int sig, int flags) {
4412 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4413 sigflags[sig] = flags;
4414 }
4415
4416 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4417 // Check for overwrite.
4418 struct sigaction oldAct;
4419 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4420
4421 void* oldhand = oldAct.sa_sigaction
4422 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4423 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4424 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4425 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4426 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4427 if (AllowUserSignalHandlers || !set_installed) {
4428 // Do not overwrite; user takes responsibility to forward to us.
4429 return;
4430 } else if (UseSignalChaining) {
4431 // save the old handler in jvm
4432 save_preinstalled_handler(sig, oldAct);
4433 // libjsig also interposes the sigaction() call below and saves the
4434 // old sigaction on it own.
4435 } else {
4436 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4437 "%#lx for signal %d.", (long)oldhand, sig));
4438 }
4439 }
4440
4441 struct sigaction sigAct;
4442 sigfillset(&(sigAct.sa_mask));
4443 sigAct.sa_handler = SIG_DFL;
4444 if (!set_installed) {
4445 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4446 } else {
4447 sigAct.sa_sigaction = signalHandler;
4448 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4449 }
4450 // Save flags, which are set by ours
4451 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4452 sigflags[sig] = sigAct.sa_flags;
4453
4454 int ret = sigaction(sig, &sigAct, &oldAct);
4455 assert(ret == 0, "check");
4456
4457 void* oldhand2 = oldAct.sa_sigaction
4458 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4459 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4460 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4461 }
4462
4463 // install signal handlers for signals that HotSpot needs to
4464 // handle in order to support Java-level exception handling.
4465
4466 void os::Linux::install_signal_handlers() {
4467 if (!signal_handlers_are_installed) {
4468 signal_handlers_are_installed = true;
4469
4470 // signal-chaining
4471 typedef void (*signal_setting_t)();
4472 signal_setting_t begin_signal_setting = NULL;
4473 signal_setting_t end_signal_setting = NULL;
4474 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4475 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4476 if (begin_signal_setting != NULL) {
4477 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4478 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4479 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4480 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4481 libjsig_is_loaded = true;
4482 assert(UseSignalChaining, "should enable signal-chaining");
4483 }
4484 if (libjsig_is_loaded) {
4485 // Tell libjsig jvm is setting signal handlers
4486 (*begin_signal_setting)();
4487 }
4488
4489 set_signal_handler(SIGSEGV, true);
4490 set_signal_handler(SIGPIPE, true);
4491 set_signal_handler(SIGBUS, true);
4492 set_signal_handler(SIGILL, true);
4493 set_signal_handler(SIGFPE, true);
4494 set_signal_handler(SIGXFSZ, true);
4495
4496 if (libjsig_is_loaded) {
4497 // Tell libjsig jvm finishes setting signal handlers
4498 (*end_signal_setting)();
4499 }
4500
4501 // We don't activate signal checker if libjsig is in place, we trust ourselves
4502 // and if UserSignalHandler is installed all bets are off.
4503 // Log that signal checking is off only if -verbose:jni is specified.
4504 if (CheckJNICalls) {
4505 if (libjsig_is_loaded) {
4506 if (PrintJNIResolving) {
4507 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4508 }
4509 check_signals = false;
4510 }
4511 if (AllowUserSignalHandlers) {
4512 if (PrintJNIResolving) {
4513 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4514 }
4515 check_signals = false;
4516 }
4517 }
4518 }
4519 }
4520
4521 // This is the fastest way to get thread cpu time on Linux.
4522 // Returns cpu time (user+sys) for any thread, not only for current.
4523 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4524 // It might work on 2.6.10+ with a special kernel/glibc patch.
4525 // For reference, please, see IEEE Std 1003.1-2004:
4526 // http://www.unix.org/single_unix_specification
4527
4528 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4529 struct timespec tp;
4530 int rc = os::Linux::clock_gettime(clockid, &tp);
4531 assert(rc == 0, "clock_gettime is expected to return 0 code");
4532
4533 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4534 }
4535
4536 /////
4537 // glibc on Linux platform uses non-documented flag
4538 // to indicate, that some special sort of signal
4539 // trampoline is used.
4540 // We will never set this flag, and we should
4541 // ignore this flag in our diagnostic
4542 #ifdef SIGNIFICANT_SIGNAL_MASK
4543 #undef SIGNIFICANT_SIGNAL_MASK
4544 #endif
4545 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4546
4547 static const char* get_signal_handler_name(address handler,
4548 char* buf, int buflen) {
4549 int offset;
4550 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4551 if (found) {
4552 // skip directory names
4553 const char *p1, *p2;
4554 p1 = buf;
4555 size_t len = strlen(os::file_separator());
4556 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4557 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4558 } else {
4559 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4560 }
4561 return buf;
4562 }
4563
4564 static void print_signal_handler(outputStream* st, int sig,
4565 char* buf, size_t buflen) {
4566 struct sigaction sa;
4567
4568 sigaction(sig, NULL, &sa);
4569
4570 // See comment for SIGNIFICANT_SIGNAL_MASK define
4571 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4572
4573 st->print("%s: ", os::exception_name(sig, buf, buflen));
4574
4575 address handler = (sa.sa_flags & SA_SIGINFO)
4576 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4577 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4578
4579 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4580 st->print("SIG_DFL");
4581 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4582 st->print("SIG_IGN");
4583 } else {
4584 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4585 }
4586
4587 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4588
4589 address rh = VMError::get_resetted_sighandler(sig);
4590 // May be, handler was resetted by VMError?
4591 if(rh != NULL) {
4592 handler = rh;
4593 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4594 }
4595
4596 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4597
4598 // Check: is it our handler?
4599 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4600 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4601 // It is our signal handler
4602 // check for flags, reset system-used one!
4603 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4604 st->print(
4605 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4606 os::Linux::get_our_sigflags(sig));
4607 }
4608 }
4609 st->cr();
4610 }
4611
4612
4613 #define DO_SIGNAL_CHECK(sig) \
4614 if (!sigismember(&check_signal_done, sig)) \
4615 os::Linux::check_signal_handler(sig)
4616
4617 // This method is a periodic task to check for misbehaving JNI applications
4618 // under CheckJNI, we can add any periodic checks here
4619
4620 void os::run_periodic_checks() {
4621
4622 if (check_signals == false) return;
4623
4624 // SEGV and BUS if overridden could potentially prevent
4625 // generation of hs*.log in the event of a crash, debugging
4626 // such a case can be very challenging, so we absolutely
4627 // check the following for a good measure:
4628 DO_SIGNAL_CHECK(SIGSEGV);
4629 DO_SIGNAL_CHECK(SIGILL);
4630 DO_SIGNAL_CHECK(SIGFPE);
4631 DO_SIGNAL_CHECK(SIGBUS);
4632 DO_SIGNAL_CHECK(SIGPIPE);
4633 DO_SIGNAL_CHECK(SIGXFSZ);
4634
4635
4636 // ReduceSignalUsage allows the user to override these handlers
4637 // see comments at the very top and jvm_solaris.h
4638 if (!ReduceSignalUsage) {
4639 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4640 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4641 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4642 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4643 }
4644
4645 DO_SIGNAL_CHECK(SR_signum);
4646 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4647 }
4648
4649 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4650
4651 static os_sigaction_t os_sigaction = NULL;
4652
4653 void os::Linux::check_signal_handler(int sig) {
4654 char buf[O_BUFLEN];
4655 address jvmHandler = NULL;
4656
4657
4658 struct sigaction act;
4659 if (os_sigaction == NULL) {
4660 // only trust the default sigaction, in case it has been interposed
4661 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4662 if (os_sigaction == NULL) return;
4663 }
4664
4665 os_sigaction(sig, (struct sigaction*)NULL, &act);
4666
4667
4668 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4669
4670 address thisHandler = (act.sa_flags & SA_SIGINFO)
4671 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4672 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4673
4674
4675 switch(sig) {
4676 case SIGSEGV:
4677 case SIGBUS:
4678 case SIGFPE:
4679 case SIGPIPE:
4680 case SIGILL:
4681 case SIGXFSZ:
4682 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4683 break;
4684
4685 case SHUTDOWN1_SIGNAL:
4686 case SHUTDOWN2_SIGNAL:
4687 case SHUTDOWN3_SIGNAL:
4688 case BREAK_SIGNAL:
4689 jvmHandler = (address)user_handler();
4690 break;
4691
4692 case INTERRUPT_SIGNAL:
4693 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4694 break;
4695
4696 default:
4697 if (sig == SR_signum) {
4698 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4699 } else {
4700 return;
4701 }
4702 break;
4703 }
4704
4705 if (thisHandler != jvmHandler) {
4706 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4707 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4708 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4709 // No need to check this sig any longer
4710 sigaddset(&check_signal_done, sig);
4711 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4712 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4713 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4714 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4715 // No need to check this sig any longer
4716 sigaddset(&check_signal_done, sig);
4717 }
4718
4719 // Dump all the signal
4720 if (sigismember(&check_signal_done, sig)) {
4721 print_signal_handlers(tty, buf, O_BUFLEN);
4722 }
4723 }
4724
4725 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4726
4727 extern bool signal_name(int signo, char* buf, size_t len);
4728
4729 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4730 if (0 < exception_code && exception_code <= SIGRTMAX) {
4731 // signal
4732 if (!signal_name(exception_code, buf, size)) {
4733 jio_snprintf(buf, size, "SIG%d", exception_code);
4734 }
4735 return buf;
4736 } else {
4737 return NULL;
4738 }
4739 }
4740
4741 // this is called _before_ the most of global arguments have been parsed
4742 void os::init(void) {
4743 char dummy; /* used to get a guess on initial stack address */
4744 // first_hrtime = gethrtime();
4745
4746 // With LinuxThreads the JavaMain thread pid (primordial thread)
4747 // is different than the pid of the java launcher thread.
4748 // So, on Linux, the launcher thread pid is passed to the VM
4749 // via the sun.java.launcher.pid property.
4750 // Use this property instead of getpid() if it was correctly passed.
4751 // See bug 6351349.
4752 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4753
4754 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4755
4756 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4757
4758 init_random(1234567);
4759
4760 ThreadCritical::initialize();
4761
4762 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4763 if (Linux::page_size() == -1) {
4764 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4765 strerror(errno)));
4766 }
4767 init_page_sizes((size_t) Linux::page_size());
4768
4769 Linux::initialize_system_info();
4770
4771 // main_thread points to the aboriginal thread
4772 Linux::_main_thread = pthread_self();
4773
4774 Linux::clock_init();
4775 initial_time_count = javaTimeNanos();
4776
4777 // pthread_condattr initialization for monotonic clock
4778 int status;
4779 pthread_condattr_t* _condattr = os::Linux::condAttr();
4780 if ((status = pthread_condattr_init(_condattr)) != 0) {
4781 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4782 }
4783 // Only set the clock if CLOCK_MONOTONIC is available
4784 if (Linux::supports_monotonic_clock()) {
4785 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4786 if (status == EINVAL) {
4787 warning("Unable to use monotonic clock with relative timed-waits" \
4788 " - changes to the time-of-day clock may have adverse affects");
4789 } else {
4790 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4791 }
4792 }
4793 }
4794 // else it defaults to CLOCK_REALTIME
4795
4796 pthread_mutex_init(&dl_mutex, NULL);
4797
4798 // If the pagesize of the VM is greater than 8K determine the appropriate
4799 // number of initial guard pages. The user can change this with the
4800 // command line arguments, if needed.
4801 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4802 StackYellowPages = 1;
4803 StackRedPages = 1;
4804 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4805 }
4806 }
4807
4808 // To install functions for atexit system call
4809 extern "C" {
4810 static void perfMemory_exit_helper() {
4811 perfMemory_exit();
4812 }
4813 }
4814
4815 // this is called _after_ the global arguments have been parsed
4816 jint os::init_2(void)
4817 {
4818 Linux::fast_thread_clock_init();
4819
4820 // Allocate a single page and mark it as readable for safepoint polling
4821 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4822 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4823
4824 os::set_polling_page( polling_page );
4825
4826 #ifndef PRODUCT
4827 if(Verbose && PrintMiscellaneous)
4828 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4829 #endif
4830
4831 if (!UseMembar) {
4832 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4833 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4834 os::set_memory_serialize_page( mem_serialize_page );
4835
4836 #ifndef PRODUCT
4837 if(Verbose && PrintMiscellaneous)
4838 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4839 #endif
4840 }
4841
4842 // initialize suspend/resume support - must do this before signal_sets_init()
4843 if (SR_initialize() != 0) {
4844 perror("SR_initialize failed");
4845 return JNI_ERR;
4846 }
4847
4848 Linux::signal_sets_init();
4849 Linux::install_signal_handlers();
4850
4851 // Check minimum allowable stack size for thread creation and to initialize
4852 // the java system classes, including StackOverflowError - depends on page
4853 // size. Add a page for compiler2 recursion in main thread.
4854 // Add in 2*BytesPerWord times page size to account for VM stack during
4855 // class initialization depending on 32 or 64 bit VM.
4856 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4857 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4858 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4859
4860 size_t threadStackSizeInBytes = ThreadStackSize * K;
4861 if (threadStackSizeInBytes != 0 &&
4862 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4863 tty->print_cr("\nThe stack size specified is too small, "
4864 "Specify at least %dk",
4865 os::Linux::min_stack_allowed/ K);
4866 return JNI_ERR;
4867 }
4868
4869 // Make the stack size a multiple of the page size so that
4870 // the yellow/red zones can be guarded.
4871 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4872 vm_page_size()));
4873
4874 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4875
4876 #if defined(IA32)
4877 workaround_expand_exec_shield_cs_limit();
4878 #endif
4879
4880 Linux::libpthread_init();
4881 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4882 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4883 Linux::glibc_version(), Linux::libpthread_version(),
4884 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4885 }
4886
4887 if (UseNUMA) {
4888 if (!Linux::libnuma_init()) {
4889 UseNUMA = false;
4890 } else {
4891 if ((Linux::numa_max_node() < 1)) {
4892 // There's only one node(they start from 0), disable NUMA.
4893 UseNUMA = false;
4894 }
4895 }
4896 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4897 // we can make the adaptive lgrp chunk resizing work. If the user specified
4898 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4899 // disable adaptive resizing.
4900 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4901 if (FLAG_IS_DEFAULT(UseNUMA)) {
4902 UseNUMA = false;
4903 } else {
4904 if (FLAG_IS_DEFAULT(UseLargePages) &&
4905 FLAG_IS_DEFAULT(UseSHM) &&
4906 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4907 UseLargePages = false;
4908 } else {
4909 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4910 UseAdaptiveSizePolicy = false;
4911 UseAdaptiveNUMAChunkSizing = false;
4912 }
4913 }
4914 }
4915 if (!UseNUMA && ForceNUMA) {
4916 UseNUMA = true;
4917 }
4918 }
4919
4920 if (MaxFDLimit) {
4921 // set the number of file descriptors to max. print out error
4922 // if getrlimit/setrlimit fails but continue regardless.
4923 struct rlimit nbr_files;
4924 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4925 if (status != 0) {
4926 if (PrintMiscellaneous && (Verbose || WizardMode))
4927 perror("os::init_2 getrlimit failed");
4928 } else {
4929 nbr_files.rlim_cur = nbr_files.rlim_max;
4930 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4931 if (status != 0) {
4932 if (PrintMiscellaneous && (Verbose || WizardMode))
4933 perror("os::init_2 setrlimit failed");
4934 }
4935 }
4936 }
4937
4938 // Initialize lock used to serialize thread creation (see os::create_thread)
4939 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4940
4941 // at-exit methods are called in the reverse order of their registration.
4942 // atexit functions are called on return from main or as a result of a
4943 // call to exit(3C). There can be only 32 of these functions registered
4944 // and atexit() does not set errno.
4945
4946 if (PerfAllowAtExitRegistration) {
4947 // only register atexit functions if PerfAllowAtExitRegistration is set.
4948 // atexit functions can be delayed until process exit time, which
4949 // can be problematic for embedded VM situations. Embedded VMs should
4950 // call DestroyJavaVM() to assure that VM resources are released.
4951
4952 // note: perfMemory_exit_helper atexit function may be removed in
4953 // the future if the appropriate cleanup code can be added to the
4954 // VM_Exit VMOperation's doit method.
4955 if (atexit(perfMemory_exit_helper) != 0) {
4956 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4957 }
4958 }
4959
4960 // initialize thread priority policy
4961 prio_init();
4962
4963 return JNI_OK;
4964 }
4965
4966 // this is called at the end of vm_initialization
4967 void os::init_3(void)
4968 {
4969 #ifdef JAVASE_EMBEDDED
4970 // Start the MemNotifyThread
4971 if (LowMemoryProtection) {
4972 MemNotifyThread::start();
4973 }
4974 return;
4975 #endif
4976 }
4977
4978 // Mark the polling page as unreadable
4979 void os::make_polling_page_unreadable(void) {
4980 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4981 fatal("Could not disable polling page");
4982 };
4983
4984 // Mark the polling page as readable
4985 void os::make_polling_page_readable(void) {
4986 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4987 fatal("Could not enable polling page");
4988 }
4989 };
4990
4991 int os::active_processor_count() {
4992 // Linux doesn't yet have a (official) notion of processor sets,
4993 // so just return the number of online processors.
4994 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4995 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4996 return online_cpus;
4997 }
4998
4999 void os::set_native_thread_name(const char *name) {
5000 // Not yet implemented.
5001 return;
5002 }
5003
5004 bool os::distribute_processes(uint length, uint* distribution) {
5005 // Not yet implemented.
5006 return false;
5007 }
5008
5009 bool os::bind_to_processor(uint processor_id) {
5010 // Not yet implemented.
5011 return false;
5012 }
5013
5014 ///
5015
5016 void os::SuspendedThreadTask::internal_do_task() {
5017 if (do_suspend(_thread->osthread())) {
5018 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5019 do_task(context);
5020 do_resume(_thread->osthread());
5021 }
5022 }
5023
5024 class PcFetcher : public os::SuspendedThreadTask {
5025 public:
5026 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5027 ExtendedPC result();
5028 protected:
5029 void do_task(const os::SuspendedThreadTaskContext& context);
5030 private:
5031 ExtendedPC _epc;
5032 };
5033
5034 ExtendedPC PcFetcher::result() {
5035 guarantee(is_done(), "task is not done yet.");
5036 return _epc;
5037 }
5038
5039 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5040 Thread* thread = context.thread();
5041 OSThread* osthread = thread->osthread();
5042 if (osthread->ucontext() != NULL) {
5043 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5044 } else {
5045 // NULL context is unexpected, double-check this is the VMThread
5046 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5047 }
5048 }
5049
5050 // Suspends the target using the signal mechanism and then grabs the PC before
5051 // resuming the target. Used by the flat-profiler only
5052 ExtendedPC os::get_thread_pc(Thread* thread) {
5053 // Make sure that it is called by the watcher for the VMThread
5054 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5055 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5056
5057 PcFetcher fetcher(thread);
5058 fetcher.run();
5059 return fetcher.result();
5060 }
5061
5062 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5063 {
5064 if (is_NPTL()) {
5065 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5066 } else {
5067 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5068 // word back to default 64bit precision if condvar is signaled. Java
5069 // wants 53bit precision. Save and restore current value.
5070 int fpu = get_fpu_control_word();
5071 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5072 set_fpu_control_word(fpu);
5073 return status;
5074 }
5075 }
5076
5077 ////////////////////////////////////////////////////////////////////////////////
5078 // debug support
5079
5080 bool os::find(address addr, outputStream* st) {
5081 Dl_info dlinfo;
5082 memset(&dlinfo, 0, sizeof(dlinfo));
5083 if (dladdr(addr, &dlinfo) != 0) {
5084 st->print(PTR_FORMAT ": ", addr);
5085 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5086 st->print("%s+%#x", dlinfo.dli_sname,
5087 addr - (intptr_t)dlinfo.dli_saddr);
5088 } else if (dlinfo.dli_fbase != NULL) {
5089 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5090 } else {
5091 st->print("<absolute address>");
5092 }
5093 if (dlinfo.dli_fname != NULL) {
5094 st->print(" in %s", dlinfo.dli_fname);
5095 }
5096 if (dlinfo.dli_fbase != NULL) {
5097 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5098 }
5099 st->cr();
5100
5101 if (Verbose) {
5102 // decode some bytes around the PC
5103 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5104 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5105 address lowest = (address) dlinfo.dli_sname;
5106 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5107 if (begin < lowest) begin = lowest;
5108 Dl_info dlinfo2;
5109 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5110 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5111 end = (address) dlinfo2.dli_saddr;
5112 Disassembler::decode(begin, end, st);
5113 }
5114 return true;
5115 }
5116 return false;
5117 }
5118
5119 ////////////////////////////////////////////////////////////////////////////////
5120 // misc
5121
5122 // This does not do anything on Linux. This is basically a hook for being
5123 // able to use structured exception handling (thread-local exception filters)
5124 // on, e.g., Win32.
5125 void
5126 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5127 JavaCallArguments* args, Thread* thread) {
5128 f(value, method, args, thread);
5129 }
5130
5131 void os::print_statistics() {
5132 }
5133
5134 int os::message_box(const char* title, const char* message) {
5135 int i;
5136 fdStream err(defaultStream::error_fd());
5137 for (i = 0; i < 78; i++) err.print_raw("=");
5138 err.cr();
5139 err.print_raw_cr(title);
5140 for (i = 0; i < 78; i++) err.print_raw("-");
5141 err.cr();
5142 err.print_raw_cr(message);
5143 for (i = 0; i < 78; i++) err.print_raw("=");
5144 err.cr();
5145
5146 char buf[16];
5147 // Prevent process from exiting upon "read error" without consuming all CPU
5148 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5149
5150 return buf[0] == 'y' || buf[0] == 'Y';
5151 }
5152
5153 int os::stat(const char *path, struct stat *sbuf) {
5154 char pathbuf[MAX_PATH];
5155 if (strlen(path) > MAX_PATH - 1) {
5156 errno = ENAMETOOLONG;
5157 return -1;
5158 }
5159 os::native_path(strcpy(pathbuf, path));
5160 return ::stat(pathbuf, sbuf);
5161 }
5162
5163 bool os::check_heap(bool force) {
5164 return true;
5165 }
5166
5167 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5168 return ::vsnprintf(buf, count, format, args);
5169 }
5170
5171 // Is a (classpath) directory empty?
5172 bool os::dir_is_empty(const char* path) {
5173 DIR *dir = NULL;
5174 struct dirent *ptr;
5175
5176 dir = opendir(path);
5177 if (dir == NULL) return true;
5178
5179 /* Scan the directory */
5180 bool result = true;
5181 char buf[sizeof(struct dirent) + MAX_PATH];
5182 while (result && (ptr = ::readdir(dir)) != NULL) {
5183 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5184 result = false;
5185 }
5186 }
5187 closedir(dir);
5188 return result;
5189 }
5190
5191 // This code originates from JDK's sysOpen and open64_w
5192 // from src/solaris/hpi/src/system_md.c
5193
5194 #ifndef O_DELETE
5195 #define O_DELETE 0x10000
5196 #endif
5197
5198 // Open a file. Unlink the file immediately after open returns
5199 // if the specified oflag has the O_DELETE flag set.
5200 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5201
5202 int os::open(const char *path, int oflag, int mode) {
5203
5204 if (strlen(path) > MAX_PATH - 1) {
5205 errno = ENAMETOOLONG;
5206 return -1;
5207 }
5208 int fd;
5209 int o_delete = (oflag & O_DELETE);
5210 oflag = oflag & ~O_DELETE;
5211
5212 fd = ::open64(path, oflag, mode);
5213 if (fd == -1) return -1;
5214
5215 //If the open succeeded, the file might still be a directory
5216 {
5217 struct stat64 buf64;
5218 int ret = ::fstat64(fd, &buf64);
5219 int st_mode = buf64.st_mode;
5220
5221 if (ret != -1) {
5222 if ((st_mode & S_IFMT) == S_IFDIR) {
5223 errno = EISDIR;
5224 ::close(fd);
5225 return -1;
5226 }
5227 } else {
5228 ::close(fd);
5229 return -1;
5230 }
5231 }
5232
5233 /*
5234 * All file descriptors that are opened in the JVM and not
5235 * specifically destined for a subprocess should have the
5236 * close-on-exec flag set. If we don't set it, then careless 3rd
5237 * party native code might fork and exec without closing all
5238 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5239 * UNIXProcess.c), and this in turn might:
5240 *
5241 * - cause end-of-file to fail to be detected on some file
5242 * descriptors, resulting in mysterious hangs, or
5243 *
5244 * - might cause an fopen in the subprocess to fail on a system
5245 * suffering from bug 1085341.
5246 *
5247 * (Yes, the default setting of the close-on-exec flag is a Unix
5248 * design flaw)
5249 *
5250 * See:
5251 * 1085341: 32-bit stdio routines should support file descriptors >255
5252 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5253 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5254 */
5255 #ifdef FD_CLOEXEC
5256 {
5257 int flags = ::fcntl(fd, F_GETFD);
5258 if (flags != -1)
5259 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5260 }
5261 #endif
5262
5263 if (o_delete != 0) {
5264 ::unlink(path);
5265 }
5266 return fd;
5267 }
5268
5269
5270 // create binary file, rewriting existing file if required
5271 int os::create_binary_file(const char* path, bool rewrite_existing) {
5272 int oflags = O_WRONLY | O_CREAT;
5273 if (!rewrite_existing) {
5274 oflags |= O_EXCL;
5275 }
5276 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5277 }
5278
5279 // return current position of file pointer
5280 jlong os::current_file_offset(int fd) {
5281 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5282 }
5283
5284 // move file pointer to the specified offset
5285 jlong os::seek_to_file_offset(int fd, jlong offset) {
5286 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5287 }
5288
5289 // This code originates from JDK's sysAvailable
5290 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5291
5292 int os::available(int fd, jlong *bytes) {
5293 jlong cur, end;
5294 int mode;
5295 struct stat64 buf64;
5296
5297 if (::fstat64(fd, &buf64) >= 0) {
5298 mode = buf64.st_mode;
5299 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5300 /*
5301 * XXX: is the following call interruptible? If so, this might
5302 * need to go through the INTERRUPT_IO() wrapper as for other
5303 * blocking, interruptible calls in this file.
5304 */
5305 int n;
5306 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5307 *bytes = n;
5308 return 1;
5309 }
5310 }
5311 }
5312 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5313 return 0;
5314 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5315 return 0;
5316 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5317 return 0;
5318 }
5319 *bytes = end - cur;
5320 return 1;
5321 }
5322
5323 int os::socket_available(int fd, jint *pbytes) {
5324 // Linux doc says EINTR not returned, unlike Solaris
5325 int ret = ::ioctl(fd, FIONREAD, pbytes);
5326
5327 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5328 // is expected to return 0 on failure and 1 on success to the jdk.
5329 return (ret < 0) ? 0 : 1;
5330 }
5331
5332 // Map a block of memory.
5333 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5334 char *addr, size_t bytes, bool read_only,
5335 bool allow_exec) {
5336 int prot;
5337 int flags = MAP_PRIVATE;
5338
5339 if (read_only) {
5340 prot = PROT_READ;
5341 } else {
5342 prot = PROT_READ | PROT_WRITE;
5343 }
5344
5345 if (allow_exec) {
5346 prot |= PROT_EXEC;
5347 }
5348
5349 if (addr != NULL) {
5350 flags |= MAP_FIXED;
5351 }
5352
5353 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5354 fd, file_offset);
5355 if (mapped_address == MAP_FAILED) {
5356 return NULL;
5357 }
5358 return mapped_address;
5359 }
5360
5361
5362 // Remap a block of memory.
5363 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5364 char *addr, size_t bytes, bool read_only,
5365 bool allow_exec) {
5366 // same as map_memory() on this OS
5367 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5368 allow_exec);
5369 }
5370
5371
5372 // Unmap a block of memory.
5373 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5374 return munmap(addr, bytes) == 0;
5375 }
5376
5377 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5378
5379 static clockid_t thread_cpu_clockid(Thread* thread) {
5380 pthread_t tid = thread->osthread()->pthread_id();
5381 clockid_t clockid;
5382
5383 // Get thread clockid
5384 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5385 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5386 return clockid;
5387 }
5388
5389 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5390 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5391 // of a thread.
5392 //
5393 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5394 // the fast estimate available on the platform.
5395
5396 jlong os::current_thread_cpu_time() {
5397 if (os::Linux::supports_fast_thread_cpu_time()) {
5398 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5399 } else {
5400 // return user + sys since the cost is the same
5401 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5402 }
5403 }
5404
5405 jlong os::thread_cpu_time(Thread* thread) {
5406 // consistent with what current_thread_cpu_time() returns
5407 if (os::Linux::supports_fast_thread_cpu_time()) {
5408 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5409 } else {
5410 return slow_thread_cpu_time(thread, true /* user + sys */);
5411 }
5412 }
5413
5414 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5415 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5416 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5417 } else {
5418 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5419 }
5420 }
5421
5422 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5423 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5424 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5425 } else {
5426 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5427 }
5428 }
5429
5430 //
5431 // -1 on error.
5432 //
5433
5434 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5435 static bool proc_task_unchecked = true;
5436 static const char *proc_stat_path = "/proc/%d/stat";
5437 pid_t tid = thread->osthread()->thread_id();
5438 char *s;
5439 char stat[2048];
5440 int statlen;
5441 char proc_name[64];
5442 int count;
5443 long sys_time, user_time;
5444 char cdummy;
5445 int idummy;
5446 long ldummy;
5447 FILE *fp;
5448
5449 // The /proc/<tid>/stat aggregates per-process usage on
5450 // new Linux kernels 2.6+ where NPTL is supported.
5451 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5452 // See bug 6328462.
5453 // There possibly can be cases where there is no directory
5454 // /proc/self/task, so we check its availability.
5455 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5456 // This is executed only once
5457 proc_task_unchecked = false;
5458 fp = fopen("/proc/self/task", "r");
5459 if (fp != NULL) {
5460 proc_stat_path = "/proc/self/task/%d/stat";
5461 fclose(fp);
5462 }
5463 }
5464
5465 sprintf(proc_name, proc_stat_path, tid);
5466 fp = fopen(proc_name, "r");
5467 if ( fp == NULL ) return -1;
5468 statlen = fread(stat, 1, 2047, fp);
5469 stat[statlen] = '\0';
5470 fclose(fp);
5471
5472 // Skip pid and the command string. Note that we could be dealing with
5473 // weird command names, e.g. user could decide to rename java launcher
5474 // to "java 1.4.2 :)", then the stat file would look like
5475 // 1234 (java 1.4.2 :)) R ... ...
5476 // We don't really need to know the command string, just find the last
5477 // occurrence of ")" and then start parsing from there. See bug 4726580.
5478 s = strrchr(stat, ')');
5479 if (s == NULL ) return -1;
5480
5481 // Skip blank chars
5482 do s++; while (isspace(*s));
5483
5484 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5485 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5486 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5487 &user_time, &sys_time);
5488 if ( count != 13 ) return -1;
5489 if (user_sys_cpu_time) {
5490 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5491 } else {
5492 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5493 }
5494 }
5495
5496 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5497 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5498 info_ptr->may_skip_backward = false; // elapsed time not wall time
5499 info_ptr->may_skip_forward = false; // elapsed time not wall time
5500 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5501 }
5502
5503 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5504 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5505 info_ptr->may_skip_backward = false; // elapsed time not wall time
5506 info_ptr->may_skip_forward = false; // elapsed time not wall time
5507 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5508 }
5509
5510 bool os::is_thread_cpu_time_supported() {
5511 return true;
5512 }
5513
5514 // System loadavg support. Returns -1 if load average cannot be obtained.
5515 // Linux doesn't yet have a (official) notion of processor sets,
5516 // so just return the system wide load average.
5517 int os::loadavg(double loadavg[], int nelem) {
5518 return ::getloadavg(loadavg, nelem);
5519 }
5520
5521 void os::pause() {
5522 char filename[MAX_PATH];
5523 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5524 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5525 } else {
5526 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5527 }
5528
5529 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5530 if (fd != -1) {
5531 struct stat buf;
5532 ::close(fd);
5533 while (::stat(filename, &buf) == 0) {
5534 (void)::poll(NULL, 0, 100);
5535 }
5536 } else {
5537 jio_fprintf(stderr,
5538 "Could not open pause file '%s', continuing immediately.\n", filename);
5539 }
5540 }
5541
5542
5543 // Refer to the comments in os_solaris.cpp park-unpark.
5544 //
5545 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5546 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5547 // For specifics regarding the bug see GLIBC BUGID 261237 :
5548 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5549 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5550 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5551 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5552 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5553 // and monitorenter when we're using 1-0 locking. All those operations may result in
5554 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5555 // of libpthread avoids the problem, but isn't practical.
5556 //
5557 // Possible remedies:
5558 //
5559 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5560 // This is palliative and probabilistic, however. If the thread is preempted
5561 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5562 // than the minimum period may have passed, and the abstime may be stale (in the
5563 // past) resultin in a hang. Using this technique reduces the odds of a hang
5564 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5565 //
5566 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5567 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5568 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5569 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5570 // thread.
5571 //
5572 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5573 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5574 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5575 // This also works well. In fact it avoids kernel-level scalability impediments
5576 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5577 // timers in a graceful fashion.
5578 //
5579 // 4. When the abstime value is in the past it appears that control returns
5580 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5581 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5582 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5583 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5584 // It may be possible to avoid reinitialization by checking the return
5585 // value from pthread_cond_timedwait(). In addition to reinitializing the
5586 // condvar we must establish the invariant that cond_signal() is only called
5587 // within critical sections protected by the adjunct mutex. This prevents
5588 // cond_signal() from "seeing" a condvar that's in the midst of being
5589 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5590 // desirable signal-after-unlock optimization that avoids futile context switching.
5591 //
5592 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5593 // structure when a condvar is used or initialized. cond_destroy() would
5594 // release the helper structure. Our reinitialize-after-timedwait fix
5595 // put excessive stress on malloc/free and locks protecting the c-heap.
5596 //
5597 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5598 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5599 // and only enabling the work-around for vulnerable environments.
5600
5601 // utility to compute the abstime argument to timedwait:
5602 // millis is the relative timeout time
5603 // abstime will be the absolute timeout time
5604 // TODO: replace compute_abstime() with unpackTime()
5605
5606 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5607 if (millis < 0) millis = 0;
5608
5609 jlong seconds = millis / 1000;
5610 millis %= 1000;
5611 if (seconds > 50000000) { // see man cond_timedwait(3T)
5612 seconds = 50000000;
5613 }
5614
5615 if (os::Linux::supports_monotonic_clock()) {
5616 struct timespec now;
5617 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5618 assert_status(status == 0, status, "clock_gettime");
5619 abstime->tv_sec = now.tv_sec + seconds;
5620 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5621 if (nanos >= NANOSECS_PER_SEC) {
5622 abstime->tv_sec += 1;
5623 nanos -= NANOSECS_PER_SEC;
5624 }
5625 abstime->tv_nsec = nanos;
5626 } else {
5627 struct timeval now;
5628 int status = gettimeofday(&now, NULL);
5629 assert(status == 0, "gettimeofday");
5630 abstime->tv_sec = now.tv_sec + seconds;
5631 long usec = now.tv_usec + millis * 1000;
5632 if (usec >= 1000000) {
5633 abstime->tv_sec += 1;
5634 usec -= 1000000;
5635 }
5636 abstime->tv_nsec = usec * 1000;
5637 }
5638 return abstime;
5639 }
5640
5641
5642 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5643 // Conceptually TryPark() should be equivalent to park(0).
5644
5645 int os::PlatformEvent::TryPark() {
5646 for (;;) {
5647 const int v = _Event ;
5648 guarantee ((v == 0) || (v == 1), "invariant") ;
5649 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5650 }
5651 }
5652
5653 void os::PlatformEvent::park() { // AKA "down()"
5654 // Invariant: Only the thread associated with the Event/PlatformEvent
5655 // may call park().
5656 // TODO: assert that _Assoc != NULL or _Assoc == Self
5657 int v ;
5658 for (;;) {
5659 v = _Event ;
5660 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5661 }
5662 guarantee (v >= 0, "invariant") ;
5663 if (v == 0) {
5664 // Do this the hard way by blocking ...
5665 int status = pthread_mutex_lock(_mutex);
5666 assert_status(status == 0, status, "mutex_lock");
5667 guarantee (_nParked == 0, "invariant") ;
5668 ++ _nParked ;
5669 while (_Event < 0) {
5670 status = pthread_cond_wait(_cond, _mutex);
5671 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5672 // Treat this the same as if the wait was interrupted
5673 if (status == ETIME) { status = EINTR; }
5674 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5675 }
5676 -- _nParked ;
5677
5678 _Event = 0 ;
5679 status = pthread_mutex_unlock(_mutex);
5680 assert_status(status == 0, status, "mutex_unlock");
5681 // Paranoia to ensure our locked and lock-free paths interact
5682 // correctly with each other.
5683 OrderAccess::fence();
5684 }
5685 guarantee (_Event >= 0, "invariant") ;
5686 }
5687
5688 int os::PlatformEvent::park(jlong millis) {
5689 guarantee (_nParked == 0, "invariant") ;
5690
5691 int v ;
5692 for (;;) {
5693 v = _Event ;
5694 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5695 }
5696 guarantee (v >= 0, "invariant") ;
5697 if (v != 0) return OS_OK ;
5698
5699 // We do this the hard way, by blocking the thread.
5700 // Consider enforcing a minimum timeout value.
5701 struct timespec abst;
5702 compute_abstime(&abst, millis);
5703
5704 int ret = OS_TIMEOUT;
5705 int status = pthread_mutex_lock(_mutex);
5706 assert_status(status == 0, status, "mutex_lock");
5707 guarantee (_nParked == 0, "invariant") ;
5708 ++_nParked ;
5709
5710 // Object.wait(timo) will return because of
5711 // (a) notification
5712 // (b) timeout
5713 // (c) thread.interrupt
5714 //
5715 // Thread.interrupt and object.notify{All} both call Event::set.
5716 // That is, we treat thread.interrupt as a special case of notification.
5717 // The underlying Solaris implementation, cond_timedwait, admits
5718 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5719 // JVM from making those visible to Java code. As such, we must
5720 // filter out spurious wakeups. We assume all ETIME returns are valid.
5721 //
5722 // TODO: properly differentiate simultaneous notify+interrupt.
5723 // In that case, we should propagate the notify to another waiter.
5724
5725 while (_Event < 0) {
5726 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5727 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5728 pthread_cond_destroy (_cond);
5729 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5730 }
5731 assert_status(status == 0 || status == EINTR ||
5732 status == ETIME || status == ETIMEDOUT,
5733 status, "cond_timedwait");
5734 if (!FilterSpuriousWakeups) break ; // previous semantics
5735 if (status == ETIME || status == ETIMEDOUT) break ;
5736 // We consume and ignore EINTR and spurious wakeups.
5737 }
5738 --_nParked ;
5739 if (_Event >= 0) {
5740 ret = OS_OK;
5741 }
5742 _Event = 0 ;
5743 status = pthread_mutex_unlock(_mutex);
5744 assert_status(status == 0, status, "mutex_unlock");
5745 assert (_nParked == 0, "invariant") ;
5746 // Paranoia to ensure our locked and lock-free paths interact
5747 // correctly with each other.
5748 OrderAccess::fence();
5749 return ret;
5750 }
5751
5752 void os::PlatformEvent::unpark() {
5753 // Transitions for _Event:
5754 // 0 :=> 1
5755 // 1 :=> 1
5756 // -1 :=> either 0 or 1; must signal target thread
5757 // That is, we can safely transition _Event from -1 to either
5758 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5759 // unpark() calls.
5760 // See also: "Semaphores in Plan 9" by Mullender & Cox
5761 //
5762 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5763 // that it will take two back-to-back park() calls for the owning
5764 // thread to block. This has the benefit of forcing a spurious return
5765 // from the first park() call after an unpark() call which will help
5766 // shake out uses of park() and unpark() without condition variables.
5767
5768 if (Atomic::xchg(1, &_Event) >= 0) return;
5769
5770 // Wait for the thread associated with the event to vacate
5771 int status = pthread_mutex_lock(_mutex);
5772 assert_status(status == 0, status, "mutex_lock");
5773 int AnyWaiters = _nParked;
5774 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5775 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5776 AnyWaiters = 0;
5777 pthread_cond_signal(_cond);
5778 }
5779 status = pthread_mutex_unlock(_mutex);
5780 assert_status(status == 0, status, "mutex_unlock");
5781 if (AnyWaiters != 0) {
5782 status = pthread_cond_signal(_cond);
5783 assert_status(status == 0, status, "cond_signal");
5784 }
5785
5786 // Note that we signal() _after dropping the lock for "immortal" Events.
5787 // This is safe and avoids a common class of futile wakeups. In rare
5788 // circumstances this can cause a thread to return prematurely from
5789 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5790 // simply re-test the condition and re-park itself.
5791 }
5792
5793
5794 // JSR166
5795 // -------------------------------------------------------
5796
5797 /*
5798 * The solaris and linux implementations of park/unpark are fairly
5799 * conservative for now, but can be improved. They currently use a
5800 * mutex/condvar pair, plus a a count.
5801 * Park decrements count if > 0, else does a condvar wait. Unpark
5802 * sets count to 1 and signals condvar. Only one thread ever waits
5803 * on the condvar. Contention seen when trying to park implies that someone
5804 * is unparking you, so don't wait. And spurious returns are fine, so there
5805 * is no need to track notifications.
5806 */
5807
5808 #define MAX_SECS 100000000
5809 /*
5810 * This code is common to linux and solaris and will be moved to a
5811 * common place in dolphin.
5812 *
5813 * The passed in time value is either a relative time in nanoseconds
5814 * or an absolute time in milliseconds. Either way it has to be unpacked
5815 * into suitable seconds and nanoseconds components and stored in the
5816 * given timespec structure.
5817 * Given time is a 64-bit value and the time_t used in the timespec is only
5818 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5819 * overflow if times way in the future are given. Further on Solaris versions
5820 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5821 * number of seconds, in abstime, is less than current_time + 100,000,000.
5822 * As it will be 28 years before "now + 100000000" will overflow we can
5823 * ignore overflow and just impose a hard-limit on seconds using the value
5824 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5825 * years from "now".
5826 */
5827
5828 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5829 assert (time > 0, "convertTime");
5830 time_t max_secs = 0;
5831
5832 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5833 struct timeval now;
5834 int status = gettimeofday(&now, NULL);
5835 assert(status == 0, "gettimeofday");
5836
5837 max_secs = now.tv_sec + MAX_SECS;
5838
5839 if (isAbsolute) {
5840 jlong secs = time / 1000;
5841 if (secs > max_secs) {
5842 absTime->tv_sec = max_secs;
5843 } else {
5844 absTime->tv_sec = secs;
5845 }
5846 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5847 } else {
5848 jlong secs = time / NANOSECS_PER_SEC;
5849 if (secs >= MAX_SECS) {
5850 absTime->tv_sec = max_secs;
5851 absTime->tv_nsec = 0;
5852 } else {
5853 absTime->tv_sec = now.tv_sec + secs;
5854 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5855 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5856 absTime->tv_nsec -= NANOSECS_PER_SEC;
5857 ++absTime->tv_sec; // note: this must be <= max_secs
5858 }
5859 }
5860 }
5861 } else {
5862 // must be relative using monotonic clock
5863 struct timespec now;
5864 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5865 assert_status(status == 0, status, "clock_gettime");
5866 max_secs = now.tv_sec + MAX_SECS;
5867 jlong secs = time / NANOSECS_PER_SEC;
5868 if (secs >= MAX_SECS) {
5869 absTime->tv_sec = max_secs;
5870 absTime->tv_nsec = 0;
5871 } else {
5872 absTime->tv_sec = now.tv_sec + secs;
5873 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5874 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5875 absTime->tv_nsec -= NANOSECS_PER_SEC;
5876 ++absTime->tv_sec; // note: this must be <= max_secs
5877 }
5878 }
5879 }
5880 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5881 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5882 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5883 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5884 }
5885
5886 void Parker::park(bool isAbsolute, jlong time) {
5887 // Ideally we'd do something useful while spinning, such
5888 // as calling unpackTime().
5889
5890 // Optional fast-path check:
5891 // Return immediately if a permit is available.
5892 // We depend on Atomic::xchg() having full barrier semantics
5893 // since we are doing a lock-free update to _counter.
5894 if (Atomic::xchg(0, &_counter) > 0) return;
5895
5896 Thread* thread = Thread::current();
5897 assert(thread->is_Java_thread(), "Must be JavaThread");
5898 JavaThread *jt = (JavaThread *)thread;
5899
5900 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5901 // Check interrupt before trying to wait
5902 if (Thread::is_interrupted(thread, false)) {
5903 return;
5904 }
5905
5906 // Next, demultiplex/decode time arguments
5907 timespec absTime;
5908 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5909 return;
5910 }
5911 if (time > 0) {
5912 unpackTime(&absTime, isAbsolute, time);
5913 }
5914
5915
5916 // Enter safepoint region
5917 // Beware of deadlocks such as 6317397.
5918 // The per-thread Parker:: mutex is a classic leaf-lock.
5919 // In particular a thread must never block on the Threads_lock while
5920 // holding the Parker:: mutex. If safepoints are pending both the
5921 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5922 ThreadBlockInVM tbivm(jt);
5923
5924 // Don't wait if cannot get lock since interference arises from
5925 // unblocking. Also. check interrupt before trying wait
5926 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5927 return;
5928 }
5929
5930 int status ;
5931 if (_counter > 0) { // no wait needed
5932 _counter = 0;
5933 status = pthread_mutex_unlock(_mutex);
5934 assert (status == 0, "invariant") ;
5935 // Paranoia to ensure our locked and lock-free paths interact
5936 // correctly with each other and Java-level accesses.
5937 OrderAccess::fence();
5938 return;
5939 }
5940
5941 #ifdef ASSERT
5942 // Don't catch signals while blocked; let the running threads have the signals.
5943 // (This allows a debugger to break into the running thread.)
5944 sigset_t oldsigs;
5945 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5946 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5947 #endif
5948
5949 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5950 jt->set_suspend_equivalent();
5951 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5952
5953 assert(_cur_index == -1, "invariant");
5954 if (time == 0) {
5955 _cur_index = REL_INDEX; // arbitrary choice when not timed
5956 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5957 } else {
5958 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5959 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5960 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5961 pthread_cond_destroy (&_cond[_cur_index]) ;
5962 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5963 }
5964 }
5965 _cur_index = -1;
5966 assert_status(status == 0 || status == EINTR ||
5967 status == ETIME || status == ETIMEDOUT,
5968 status, "cond_timedwait");
5969
5970 #ifdef ASSERT
5971 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5972 #endif
5973
5974 _counter = 0 ;
5975 status = pthread_mutex_unlock(_mutex) ;
5976 assert_status(status == 0, status, "invariant") ;
5977 // Paranoia to ensure our locked and lock-free paths interact
5978 // correctly with each other and Java-level accesses.
5979 OrderAccess::fence();
5980
5981 // If externally suspended while waiting, re-suspend
5982 if (jt->handle_special_suspend_equivalent_condition()) {
5983 jt->java_suspend_self();
5984 }
5985 }
5986
5987 void Parker::unpark() {
5988 int s, status ;
5989 status = pthread_mutex_lock(_mutex);
5990 assert (status == 0, "invariant") ;
5991 s = _counter;
5992 _counter = 1;
5993 if (s < 1) {
5994 // thread might be parked
5995 if (_cur_index != -1) {
5996 // thread is definitely parked
5997 if (WorkAroundNPTLTimedWaitHang) {
5998 status = pthread_cond_signal (&_cond[_cur_index]);
5999 assert (status == 0, "invariant");
6000 status = pthread_mutex_unlock(_mutex);
6001 assert (status == 0, "invariant");
6002 } else {
6003 status = pthread_mutex_unlock(_mutex);
6004 assert (status == 0, "invariant");
6005 status = pthread_cond_signal (&_cond[_cur_index]);
6006 assert (status == 0, "invariant");
6007 }
6008 } else {
6009 pthread_mutex_unlock(_mutex);
6010 assert (status == 0, "invariant") ;
6011 }
6012 } else {
6013 pthread_mutex_unlock(_mutex);
6014 assert (status == 0, "invariant") ;
6015 }
6016 }
6017
6018
6019 extern char** environ;
6020
6021 #ifndef __NR_fork
6022 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
6023 #endif
6024
6025 #ifndef __NR_execve
6026 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
6027 #endif
6028
6029 // Run the specified command in a separate process. Return its exit value,
6030 // or -1 on failure (e.g. can't fork a new process).
6031 // Unlike system(), this function can be called from signal handler. It
6032 // doesn't block SIGINT et al.
6033 int os::fork_and_exec(char* cmd) {
6034 const char * argv[4] = {"sh", "-c", cmd, NULL};
6035
6036 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
6037 // pthread_atfork handlers and reset pthread library. All we need is a
6038 // separate process to execve. Make a direct syscall to fork process.
6039 // On IA64 there's no fork syscall, we have to use fork() and hope for
6040 // the best...
6041 pid_t pid = NOT_IA64(syscall(__NR_fork);)
6042 IA64_ONLY(fork();)
6043
6044 if (pid < 0) {
6045 // fork failed
6046 return -1;
6047
6048 } else if (pid == 0) {
6049 // child process
6050
6051 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6052 // first to kill every thread on the thread list. Because this list is
6053 // not reset by fork() (see notes above), execve() will instead kill
6054 // every thread in the parent process. We know this is the only thread
6055 // in the new process, so make a system call directly.
6056 // IA64 should use normal execve() from glibc to match the glibc fork()
6057 // above.
6058 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6059 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6060
6061 // execve failed
6062 _exit(-1);
6063
6064 } else {
6065 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6066 // care about the actual exit code, for now.
6067
6068 int status;
6069
6070 // Wait for the child process to exit. This returns immediately if
6071 // the child has already exited. */
6072 while (waitpid(pid, &status, 0) < 0) {
6073 switch (errno) {
6074 case ECHILD: return 0;
6075 case EINTR: break;
6076 default: return -1;
6077 }
6078 }
6079
6080 if (WIFEXITED(status)) {
6081 // The child exited normally; get its exit code.
6082 return WEXITSTATUS(status);
6083 } else if (WIFSIGNALED(status)) {
6084 // The child exited because of a signal
6085 // The best value to return is 0x80 + signal number,
6086 // because that is what all Unix shells do, and because
6087 // it allows callers to distinguish between process exit and
6088 // process death by signal.
6089 return 0x80 + WTERMSIG(status);
6090 } else {
6091 // Unknown exit code; pass it through
6092 return status;
6093 }
6094 }
6095 }
6096
6097 // is_headless_jre()
6098 //
6099 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6100 // in order to report if we are running in a headless jre
6101 //
6102 // Since JDK8 xawt/libmawt.so was moved into the same directory
6103 // as libawt.so, and renamed libawt_xawt.so
6104 //
6105 bool os::is_headless_jre() {
6106 struct stat statbuf;
6107 char buf[MAXPATHLEN];
6108 char libmawtpath[MAXPATHLEN];
6109 const char *xawtstr = "/xawt/libmawt.so";
6110 const char *new_xawtstr = "/libawt_xawt.so";
6111 char *p;
6112
6113 // Get path to libjvm.so
6114 os::jvm_path(buf, sizeof(buf));
6115
6116 // Get rid of libjvm.so
6117 p = strrchr(buf, '/');
6118 if (p == NULL) return false;
6119 else *p = '\0';
6120
6121 // Get rid of client or server
6122 p = strrchr(buf, '/');
6123 if (p == NULL) return false;
6124 else *p = '\0';
6125
6126 // check xawt/libmawt.so
6127 strcpy(libmawtpath, buf);
6128 strcat(libmawtpath, xawtstr);
6129 if (::stat(libmawtpath, &statbuf) == 0) return false;
6130
6131 // check libawt_xawt.so
6132 strcpy(libmawtpath, buf);
6133 strcat(libmawtpath, new_xawtstr);
6134 if (::stat(libmawtpath, &statbuf) == 0) return false;
6135
6136 return true;
6137 }
6138
6139 // Get the default path to the core file
6140 // Returns the length of the string
6141 int os::get_core_path(char* buffer, size_t bufferSize) {
6142 const char* p = get_current_directory(buffer, bufferSize);
6143
6144 if (p == NULL) {
6145 assert(p != NULL, "failed to get current directory");
6146 return 0;
6147 }
6148
6149 return strlen(buffer);
6150 }
6151
6152 #ifdef JAVASE_EMBEDDED
6153 //
6154 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6155 //
6156 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6157
6158 // ctor
6159 //
6160 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6161 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6162 _fd = fd;
6163
6164 if (os::create_thread(this, os::os_thread)) {
6165 _memnotify_thread = this;
6166 os::set_priority(this, NearMaxPriority);
6167 os::start_thread(this);
6168 }
6169 }
6170
6171 // Where all the work gets done
6172 //
6173 void MemNotifyThread::run() {
6174 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6175
6176 // Set up the select arguments
6177 fd_set rfds;
6178 if (_fd != -1) {
6179 FD_ZERO(&rfds);
6180 FD_SET(_fd, &rfds);
6181 }
6182
6183 // Now wait for the mem_notify device to wake up
6184 while (1) {
6185 // Wait for the mem_notify device to signal us..
6186 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6187 if (rc == -1) {
6188 perror("select!\n");
6189 break;
6190 } else if (rc) {
6191 //ssize_t free_before = os::available_memory();
6192 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6193
6194 // The kernel is telling us there is not much memory left...
6195 // try to do something about that
6196
6197 // If we are not already in a GC, try one.
6198 if (!Universe::heap()->is_gc_active()) {
6199 Universe::heap()->collect(GCCause::_allocation_failure);
6200
6201 //ssize_t free_after = os::available_memory();
6202 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6203 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6204 }
6205 // We might want to do something like the following if we find the GC's are not helping...
6206 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6207 }
6208 }
6209 }
6210
6211 //
6212 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6213 //
6214 void MemNotifyThread::start() {
6215 int fd;
6216 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6217 if (fd < 0) {
6218 return;
6219 }
6220
6221 if (memnotify_thread() == NULL) {
6222 new MemNotifyThread(fd);
6223 }
6224 }
6225
6226 #endif // JAVASE_EMBEDDED
6227
6228
6229 /////////////// Unit tests ///////////////
6230
6231 #ifndef PRODUCT
6232
6233 #define test_log(...) \
6234 do {\
6235 if (VerboseInternalVMTests) { \
6236 tty->print_cr(__VA_ARGS__); \
6237 tty->flush(); \
6238 }\
6239 } while (false)
6240
6241 class TestReserveMemorySpecial : AllStatic {
6242 public:
6243 static void small_page_write(void* addr, size_t size) {
6244 size_t page_size = os::vm_page_size();
6245
6246 char* end = (char*)addr + size;
6247 for (char* p = (char*)addr; p < end; p += page_size) {
6248 *p = 1;
6249 }
6250 }
6251
6252 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6253 if (!UseHugeTLBFS) {
6254 return;
6255 }
6256
6257 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6258
6259 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6260
6261 if (addr != NULL) {
6262 small_page_write(addr, size);
6263
6264 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6265 }
6266 }
6267
6268 static void test_reserve_memory_special_huge_tlbfs_only() {
6269 if (!UseHugeTLBFS) {
6270 return;
6271 }
6272
6273 size_t lp = os::large_page_size();
6274
6275 for (size_t size = lp; size <= lp * 10; size += lp) {
6276 test_reserve_memory_special_huge_tlbfs_only(size);
6277 }
6278 }
6279
6280 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6281 if (!UseHugeTLBFS) {
6282 return;
6283 }
6284
6285 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6286 size, alignment);
6287
6288 assert(size >= os::large_page_size(), "Incorrect input to test");
6289
6290 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6291
6292 if (addr != NULL) {
6293 small_page_write(addr, size);
6294
6295 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6296 }
6297 }
6298
6299 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6300 size_t lp = os::large_page_size();
6301 size_t ag = os::vm_allocation_granularity();
6302
6303 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6304 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6305 }
6306 }
6307
6308 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6309 size_t lp = os::large_page_size();
6310 size_t ag = os::vm_allocation_granularity();
6311
6312 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6313 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6314 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6315 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6316 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6317 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6318 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6319 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6320 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6321 }
6322
6323 static void test_reserve_memory_special_huge_tlbfs() {
6324 if (!UseHugeTLBFS) {
6325 return;
6326 }
6327
6328 test_reserve_memory_special_huge_tlbfs_only();
6329 test_reserve_memory_special_huge_tlbfs_mixed();
6330 }
6331
6332 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6333 if (!UseSHM) {
6334 return;
6335 }
6336
6337 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6338
6339 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6340
6341 if (addr != NULL) {
6342 assert(is_ptr_aligned(addr, alignment), "Check");
6343 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6344
6345 small_page_write(addr, size);
6346
6347 os::Linux::release_memory_special_shm(addr, size);
6348 }
6349 }
6350
6351 static void test_reserve_memory_special_shm() {
6352 size_t lp = os::large_page_size();
6353 size_t ag = os::vm_allocation_granularity();
6354
6355 for (size_t size = ag; size < lp * 3; size += ag) {
6356 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6357 test_reserve_memory_special_shm(size, alignment);
6358 }
6359 }
6360 }
6361
6362 static void test() {
6363 test_reserve_memory_special_huge_tlbfs();
6364 test_reserve_memory_special_shm();
6365 }
6366 };
6367
6368 void TestReserveMemorySpecial_test() {
6369 TestReserveMemorySpecial::test();
6370 }
6371
6372 #endif