1 /*
2 * Copyright (c) 1999, 2011, 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 # define __STDC_FORMAT_MACROS
26
27 // no precompiled headers
28 #include "classfile/classLoader.hpp"
29 #include "classfile/systemDictionary.hpp"
30 #include "classfile/vmSymbols.hpp"
31 #include "code/icBuffer.hpp"
32 #include "code/vtableStubs.hpp"
33 #include "compiler/compileBroker.hpp"
34 #include "interpreter/interpreter.hpp"
35 #include "jvm_linux.h"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/filemap.hpp"
38 #include "mutex_linux.inline.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/interfaceSupport.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/threadCritical.hpp"
58 #include "runtime/timer.hpp"
59 #include "services/attachListener.hpp"
60 #include "services/runtimeService.hpp"
61 #include "thread_linux.inline.hpp"
62 #include "utilities/decoder.hpp"
63 #include "utilities/defaultStream.hpp"
64 #include "utilities/events.hpp"
65 #include "utilities/growableArray.hpp"
66 #include "utilities/vmError.hpp"
67 #ifdef TARGET_ARCH_x86
68 # include "assembler_x86.inline.hpp"
69 # include "nativeInst_x86.hpp"
70 #endif
71 #ifdef TARGET_ARCH_sparc
72 # include "assembler_sparc.inline.hpp"
73 # include "nativeInst_sparc.hpp"
74 #endif
75 #ifdef TARGET_ARCH_zero
76 # include "assembler_zero.inline.hpp"
77 # include "nativeInst_zero.hpp"
78 #endif
79 #ifdef TARGET_ARCH_arm
80 # include "assembler_arm.inline.hpp"
81 # include "nativeInst_arm.hpp"
82 #endif
83 #ifdef TARGET_ARCH_ppc
84 # include "assembler_ppc.inline.hpp"
85 # include "nativeInst_ppc.hpp"
86 #endif
87 #ifdef COMPILER1
88 #include "c1/c1_Runtime1.hpp"
89 #endif
90 #ifdef COMPILER2
91 #include "opto/runtime.hpp"
92 #endif
93
94 // put OS-includes here
95 # include <sys/types.h>
96 # include <sys/mman.h>
97 # include <sys/stat.h>
98 # include <sys/select.h>
99 # include <pthread.h>
100 # include <signal.h>
101 # include <errno.h>
102 # include <dlfcn.h>
103 # include <stdio.h>
104 # include <unistd.h>
105 # include <sys/resource.h>
106 # include <pthread.h>
107 # include <sys/stat.h>
108 # include <sys/time.h>
109 # include <sys/times.h>
110 # include <sys/utsname.h>
111 # include <sys/socket.h>
112 # include <sys/wait.h>
113 # include <pwd.h>
114 # include <poll.h>
115 # include <semaphore.h>
116 # include <fcntl.h>
117 # include <string.h>
118 # include <syscall.h>
119 # include <sys/sysinfo.h>
120 # include <gnu/libc-version.h>
121 # include <sys/ipc.h>
122 # include <sys/shm.h>
123 # include <link.h>
124 # include <stdint.h>
125 # include <inttypes.h>
126 # include <sys/ioctl.h>
127
128 #define MAX_PATH (2 * K)
129
130 // for timer info max values which include all bits
131 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
132 #define SEC_IN_NANOSECS 1000000000LL
133
134 #define LARGEPAGES_BIT (1 << 6)
135 ////////////////////////////////////////////////////////////////////////////////
136 // global variables
137 julong os::Linux::_physical_memory = 0;
138
139 address os::Linux::_initial_thread_stack_bottom = NULL;
140 uintptr_t os::Linux::_initial_thread_stack_size = 0;
141
142 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
143 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
144 Mutex* os::Linux::_createThread_lock = NULL;
145 pthread_t os::Linux::_main_thread;
146 int os::Linux::_page_size = -1;
147 bool os::Linux::_is_floating_stack = false;
148 bool os::Linux::_is_NPTL = false;
149 bool os::Linux::_supports_fast_thread_cpu_time = false;
150 const char * os::Linux::_glibc_version = NULL;
151 const char * os::Linux::_libpthread_version = NULL;
152
153 static jlong initial_time_count=0;
154
155 static int clock_tics_per_sec = 100;
156
157 // For diagnostics to print a message once. see run_periodic_checks
158 static sigset_t check_signal_done;
159 static bool check_signals = true;;
160
161 static pid_t _initial_pid = 0;
162
163 /* Signal number used to suspend/resume a thread */
164
165 /* do not use any signal number less than SIGSEGV, see 4355769 */
166 static int SR_signum = SIGUSR2;
167 sigset_t SR_sigset;
168
169 /* Used to protect dlsym() calls */
170 static pthread_mutex_t dl_mutex;
171
172 ////////////////////////////////////////////////////////////////////////////////
173 // utility functions
174
175 static int SR_initialize();
176 static int SR_finalize();
177
178 julong os::available_memory() {
179 return Linux::available_memory();
180 }
181
182 julong os::Linux::available_memory() {
183 // values in struct sysinfo are "unsigned long"
184 struct sysinfo si;
185 sysinfo(&si);
186
187 return (julong)si.freeram * si.mem_unit;
188 }
189
190 julong os::physical_memory() {
191 return Linux::physical_memory();
192 }
193
194 julong os::allocatable_physical_memory(julong size) {
195 #ifdef _LP64
196 return size;
197 #else
198 julong result = MIN2(size, (julong)3800*M);
199 if (!is_allocatable(result)) {
200 // See comments under solaris for alignment considerations
201 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
202 result = MIN2(size, reasonable_size);
203 }
204 return result;
205 #endif // _LP64
206 }
207
208 ////////////////////////////////////////////////////////////////////////////////
209 // environment support
210
211 bool os::getenv(const char* name, char* buf, int len) {
212 const char* val = ::getenv(name);
213 if (val != NULL && strlen(val) < (size_t)len) {
214 strcpy(buf, val);
215 return true;
216 }
217 if (len > 0) buf[0] = 0; // return a null string
218 return false;
219 }
220
221
222 // Return true if user is running as root.
223
224 bool os::have_special_privileges() {
225 static bool init = false;
226 static bool privileges = false;
227 if (!init) {
228 privileges = (getuid() != geteuid()) || (getgid() != getegid());
229 init = true;
230 }
231 return privileges;
232 }
233
234
235 #ifndef SYS_gettid
236 // i386: 224, ia64: 1105, amd64: 186, sparc 143
237 #ifdef __ia64__
238 #define SYS_gettid 1105
239 #elif __i386__
240 #define SYS_gettid 224
241 #elif __amd64__
242 #define SYS_gettid 186
243 #elif __sparc__
244 #define SYS_gettid 143
245 #else
246 #error define gettid for the arch
247 #endif
248 #endif
249
250 // Cpu architecture string
251 #if defined(ZERO)
252 static char cpu_arch[] = ZERO_LIBARCH;
253 #elif defined(IA64)
254 static char cpu_arch[] = "ia64";
255 #elif defined(IA32)
256 static char cpu_arch[] = "i386";
257 #elif defined(AMD64)
258 static char cpu_arch[] = "amd64";
259 #elif defined(ARM)
260 static char cpu_arch[] = "arm";
261 #elif defined(PPC)
262 static char cpu_arch[] = "ppc";
263 #elif defined(SPARC)
264 # ifdef _LP64
265 static char cpu_arch[] = "sparcv9";
266 # else
267 static char cpu_arch[] = "sparc";
268 # endif
269 #else
270 #error Add appropriate cpu_arch setting
271 #endif
272
273
274 // pid_t gettid()
275 //
276 // Returns the kernel thread id of the currently running thread. Kernel
277 // thread id is used to access /proc.
278 //
279 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
280 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
281 //
282 pid_t os::Linux::gettid() {
283 int rslt = syscall(SYS_gettid);
284 if (rslt == -1) {
285 // old kernel, no NPTL support
286 return getpid();
287 } else {
288 return (pid_t)rslt;
289 }
290 }
291
292 // Most versions of linux have a bug where the number of processors are
293 // determined by looking at the /proc file system. In a chroot environment,
294 // the system call returns 1. This causes the VM to act as if it is
295 // a single processor and elide locking (see is_MP() call).
296 static bool unsafe_chroot_detected = false;
297 static const char *unstable_chroot_error = "/proc file system not found.\n"
298 "Java may be unstable running multithreaded in a chroot "
299 "environment on Linux when /proc filesystem is not mounted.";
300
301 void os::Linux::initialize_system_info() {
302 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
303 if (processor_count() == 1) {
304 pid_t pid = os::Linux::gettid();
305 char fname[32];
306 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
307 FILE *fp = fopen(fname, "r");
308 if (fp == NULL) {
309 unsafe_chroot_detected = true;
310 } else {
311 fclose(fp);
312 }
313 }
314 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
315 assert(processor_count() > 0, "linux error");
316 }
317
318 void os::init_system_properties_values() {
319 // char arch[12];
320 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
321
322 // The next steps are taken in the product version:
323 //
324 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
325 // This library should be located at:
326 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
327 //
328 // If "/jre/lib/" appears at the right place in the path, then we
329 // assume libjvm[_g].so is installed in a JDK and we use this path.
330 //
331 // Otherwise exit with message: "Could not create the Java virtual machine."
332 //
333 // The following extra steps are taken in the debugging version:
334 //
335 // If "/jre/lib/" does NOT appear at the right place in the path
336 // instead of exit check for $JAVA_HOME environment variable.
337 //
338 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
339 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
340 // it looks like libjvm[_g].so is installed there
341 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
342 //
343 // Otherwise exit.
344 //
345 // Important note: if the location of libjvm.so changes this
346 // code needs to be changed accordingly.
347
348 // The next few definitions allow the code to be verbatim:
349 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n))
350 #define getenv(n) ::getenv(n)
351
352 /*
353 * See ld(1):
354 * The linker uses the following search paths to locate required
355 * shared libraries:
356 * 1: ...
357 * ...
358 * 7: The default directories, normally /lib and /usr/lib.
359 */
360 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
361 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
362 #else
363 #define DEFAULT_LIBPATH "/lib:/usr/lib"
364 #endif
365
366 #define EXTENSIONS_DIR "/lib/ext"
367 #define ENDORSED_DIR "/lib/endorsed"
368 #define REG_DIR "/usr/java/packages"
369
370 {
371 /* sysclasspath, java_home, dll_dir */
372 {
373 char *home_path;
374 char *dll_path;
375 char *pslash;
376 char buf[MAXPATHLEN];
377 os::jvm_path(buf, sizeof(buf));
378
379 // Found the full path to libjvm.so.
380 // Now cut the path to <java_home>/jre if we can.
381 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
382 pslash = strrchr(buf, '/');
383 if (pslash != NULL)
384 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
385 dll_path = malloc(strlen(buf) + 1);
386 if (dll_path == NULL)
387 return;
388 strcpy(dll_path, buf);
389 Arguments::set_dll_dir(dll_path);
390
391 if (pslash != NULL) {
392 pslash = strrchr(buf, '/');
393 if (pslash != NULL) {
394 *pslash = '\0'; /* get rid of /<arch> */
395 pslash = strrchr(buf, '/');
396 if (pslash != NULL)
397 *pslash = '\0'; /* get rid of /lib */
398 }
399 }
400
401 home_path = malloc(strlen(buf) + 1);
402 if (home_path == NULL)
403 return;
404 strcpy(home_path, buf);
405 Arguments::set_java_home(home_path);
406
407 if (!set_boot_path('/', ':'))
408 return;
409 }
410
411 /*
412 * Where to look for native libraries
413 *
414 * Note: Due to a legacy implementation, most of the library path
415 * is set in the launcher. This was to accomodate linking restrictions
416 * on legacy Linux implementations (which are no longer supported).
417 * Eventually, all the library path setting will be done here.
418 *
419 * However, to prevent the proliferation of improperly built native
420 * libraries, the new path component /usr/java/packages is added here.
421 * Eventually, all the library path setting will be done here.
422 */
423 {
424 char *ld_library_path;
425
426 /*
427 * Construct the invariant part of ld_library_path. Note that the
428 * space for the colon and the trailing null are provided by the
429 * nulls included by the sizeof operator (so actually we allocate
430 * a byte more than necessary).
431 */
432 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
433 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
434 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
435
436 /*
437 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
438 * should always exist (until the legacy problem cited above is
439 * addressed).
440 */
441 char *v = getenv("LD_LIBRARY_PATH");
442 if (v != NULL) {
443 char *t = ld_library_path;
444 /* That's +1 for the colon and +1 for the trailing '\0' */
445 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
446 sprintf(ld_library_path, "%s:%s", v, t);
447 }
448 Arguments::set_library_path(ld_library_path);
449 }
450
451 /*
452 * Extensions directories.
453 *
454 * Note that the space for the colon and the trailing null are provided
455 * by the nulls included by the sizeof operator (so actually one byte more
456 * than necessary is allocated).
457 */
458 {
459 char *buf = malloc(strlen(Arguments::get_java_home()) +
460 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
461 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
462 Arguments::get_java_home());
463 Arguments::set_ext_dirs(buf);
464 }
465
466 /* Endorsed standards default directory. */
467 {
468 char * buf;
469 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
470 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
471 Arguments::set_endorsed_dirs(buf);
472 }
473 }
474
475 #undef malloc
476 #undef getenv
477 #undef EXTENSIONS_DIR
478 #undef ENDORSED_DIR
479
480 // Done
481 return;
482 }
483
484 ////////////////////////////////////////////////////////////////////////////////
485 // breakpoint support
486
487 void os::breakpoint() {
488 BREAKPOINT;
489 }
490
491 extern "C" void breakpoint() {
492 // use debugger to set breakpoint here
493 }
494
495 ////////////////////////////////////////////////////////////////////////////////
496 // signal support
497
498 debug_only(static bool signal_sets_initialized = false);
499 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
500
501 bool os::Linux::is_sig_ignored(int sig) {
502 struct sigaction oact;
503 sigaction(sig, (struct sigaction*)NULL, &oact);
504 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
505 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
506 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
507 return true;
508 else
509 return false;
510 }
511
512 void os::Linux::signal_sets_init() {
513 // Should also have an assertion stating we are still single-threaded.
514 assert(!signal_sets_initialized, "Already initialized");
515 // Fill in signals that are necessarily unblocked for all threads in
516 // the VM. Currently, we unblock the following signals:
517 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
518 // by -Xrs (=ReduceSignalUsage));
519 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
520 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
521 // the dispositions or masks wrt these signals.
522 // Programs embedding the VM that want to use the above signals for their
523 // own purposes must, at this time, use the "-Xrs" option to prevent
524 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
525 // (See bug 4345157, and other related bugs).
526 // In reality, though, unblocking these signals is really a nop, since
527 // these signals are not blocked by default.
528 sigemptyset(&unblocked_sigs);
529 sigemptyset(&allowdebug_blocked_sigs);
530 sigaddset(&unblocked_sigs, SIGILL);
531 sigaddset(&unblocked_sigs, SIGSEGV);
532 sigaddset(&unblocked_sigs, SIGBUS);
533 sigaddset(&unblocked_sigs, SIGFPE);
534 sigaddset(&unblocked_sigs, SR_signum);
535
536 if (!ReduceSignalUsage) {
537 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
538 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
539 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
540 }
541 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
542 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
543 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
544 }
545 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
546 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
547 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
548 }
549 }
550 // Fill in signals that are blocked by all but the VM thread.
551 sigemptyset(&vm_sigs);
552 if (!ReduceSignalUsage)
553 sigaddset(&vm_sigs, BREAK_SIGNAL);
554 debug_only(signal_sets_initialized = true);
555
556 }
557
558 // These are signals that are unblocked while a thread is running Java.
559 // (For some reason, they get blocked by default.)
560 sigset_t* os::Linux::unblocked_signals() {
561 assert(signal_sets_initialized, "Not initialized");
562 return &unblocked_sigs;
563 }
564
565 // These are the signals that are blocked while a (non-VM) thread is
566 // running Java. Only the VM thread handles these signals.
567 sigset_t* os::Linux::vm_signals() {
568 assert(signal_sets_initialized, "Not initialized");
569 return &vm_sigs;
570 }
571
572 // These are signals that are blocked during cond_wait to allow debugger in
573 sigset_t* os::Linux::allowdebug_blocked_signals() {
574 assert(signal_sets_initialized, "Not initialized");
575 return &allowdebug_blocked_sigs;
576 }
577
578 void os::Linux::hotspot_sigmask(Thread* thread) {
579
580 //Save caller's signal mask before setting VM signal mask
581 sigset_t caller_sigmask;
582 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
583
584 OSThread* osthread = thread->osthread();
585 osthread->set_caller_sigmask(caller_sigmask);
586
587 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
588
589 if (!ReduceSignalUsage) {
590 if (thread->is_VM_thread()) {
591 // Only the VM thread handles BREAK_SIGNAL ...
592 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
593 } else {
594 // ... all other threads block BREAK_SIGNAL
595 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
596 }
597 }
598 }
599
600 //////////////////////////////////////////////////////////////////////////////
601 // detecting pthread library
602
603 void os::Linux::libpthread_init() {
604 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
605 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
606 // generic name for earlier versions.
607 // Define macros here so we can build HotSpot on old systems.
608 # ifndef _CS_GNU_LIBC_VERSION
609 # define _CS_GNU_LIBC_VERSION 2
610 # endif
611 # ifndef _CS_GNU_LIBPTHREAD_VERSION
612 # define _CS_GNU_LIBPTHREAD_VERSION 3
613 # endif
614
615 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
616 if (n > 0) {
617 char *str = (char *)malloc(n);
618 confstr(_CS_GNU_LIBC_VERSION, str, n);
619 os::Linux::set_glibc_version(str);
620 } else {
621 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
622 static char _gnu_libc_version[32];
623 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
624 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
625 os::Linux::set_glibc_version(_gnu_libc_version);
626 }
627
628 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
629 if (n > 0) {
630 char *str = (char *)malloc(n);
631 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
632 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
633 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
634 // is the case. LinuxThreads has a hard limit on max number of threads.
635 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
636 // On the other hand, NPTL does not have such a limit, sysconf()
637 // will return -1 and errno is not changed. Check if it is really NPTL.
638 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
639 strstr(str, "NPTL") &&
640 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
641 free(str);
642 os::Linux::set_libpthread_version("linuxthreads");
643 } else {
644 os::Linux::set_libpthread_version(str);
645 }
646 } else {
647 // glibc before 2.3.2 only has LinuxThreads.
648 os::Linux::set_libpthread_version("linuxthreads");
649 }
650
651 if (strstr(libpthread_version(), "NPTL")) {
652 os::Linux::set_is_NPTL();
653 } else {
654 os::Linux::set_is_LinuxThreads();
655 }
656
657 // LinuxThreads have two flavors: floating-stack mode, which allows variable
658 // stack size; and fixed-stack mode. NPTL is always floating-stack.
659 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
660 os::Linux::set_is_floating_stack();
661 }
662 }
663
664 /////////////////////////////////////////////////////////////////////////////
665 // thread stack
666
667 // Force Linux kernel to expand current thread stack. If "bottom" is close
668 // to the stack guard, caller should block all signals.
669 //
670 // MAP_GROWSDOWN:
671 // A special mmap() flag that is used to implement thread stacks. It tells
672 // kernel that the memory region should extend downwards when needed. This
673 // allows early versions of LinuxThreads to only mmap the first few pages
674 // when creating a new thread. Linux kernel will automatically expand thread
675 // stack as needed (on page faults).
676 //
677 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
678 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
679 // region, it's hard to tell if the fault is due to a legitimate stack
680 // access or because of reading/writing non-exist memory (e.g. buffer
681 // overrun). As a rule, if the fault happens below current stack pointer,
682 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
683 // application (see Linux kernel fault.c).
684 //
685 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
686 // stack overflow detection.
687 //
688 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
689 // not use this flag. However, the stack of initial thread is not created
690 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
691 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
692 // and then attach the thread to JVM.
693 //
694 // To get around the problem and allow stack banging on Linux, we need to
695 // manually expand thread stack after receiving the SIGSEGV.
696 //
697 // There are two ways to expand thread stack to address "bottom", we used
698 // both of them in JVM before 1.5:
699 // 1. adjust stack pointer first so that it is below "bottom", and then
700 // touch "bottom"
701 // 2. mmap() the page in question
702 //
703 // Now alternate signal stack is gone, it's harder to use 2. For instance,
704 // if current sp is already near the lower end of page 101, and we need to
705 // call mmap() to map page 100, it is possible that part of the mmap() frame
706 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
707 // That will destroy the mmap() frame and cause VM to crash.
708 //
709 // The following code works by adjusting sp first, then accessing the "bottom"
710 // page to force a page fault. Linux kernel will then automatically expand the
711 // stack mapping.
712 //
713 // _expand_stack_to() assumes its frame size is less than page size, which
714 // should always be true if the function is not inlined.
715
716 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
717 #define NOINLINE
718 #else
719 #define NOINLINE __attribute__ ((noinline))
720 #endif
721
722 static void _expand_stack_to(address bottom) NOINLINE;
723
724 static void _expand_stack_to(address bottom) {
725 address sp;
726 size_t size;
727 volatile char *p;
728
729 // Adjust bottom to point to the largest address within the same page, it
730 // gives us a one-page buffer if alloca() allocates slightly more memory.
731 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
732 bottom += os::Linux::page_size() - 1;
733
734 // sp might be slightly above current stack pointer; if that's the case, we
735 // will alloca() a little more space than necessary, which is OK. Don't use
736 // os::current_stack_pointer(), as its result can be slightly below current
737 // stack pointer, causing us to not alloca enough to reach "bottom".
738 sp = (address)&sp;
739
740 if (sp > bottom) {
741 size = sp - bottom;
742 p = (volatile char *)alloca(size);
743 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
744 p[0] = '\0';
745 }
746 }
747
748 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
749 assert(t!=NULL, "just checking");
750 assert(t->osthread()->expanding_stack(), "expand should be set");
751 assert(t->stack_base() != NULL, "stack_base was not initialized");
752
753 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
754 sigset_t mask_all, old_sigset;
755 sigfillset(&mask_all);
756 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
757 _expand_stack_to(addr);
758 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
759 return true;
760 }
761 return false;
762 }
763
764 //////////////////////////////////////////////////////////////////////////////
765 // create new thread
766
767 static address highest_vm_reserved_address();
768
769 // check if it's safe to start a new thread
770 static bool _thread_safety_check(Thread* thread) {
771 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
772 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
773 // Heap is mmap'ed at lower end of memory space. Thread stacks are
774 // allocated (MAP_FIXED) from high address space. Every thread stack
775 // occupies a fixed size slot (usually 2Mbytes, but user can change
776 // it to other values if they rebuild LinuxThreads).
777 //
778 // Problem with MAP_FIXED is that mmap() can still succeed even part of
779 // the memory region has already been mmap'ed. That means if we have too
780 // many threads and/or very large heap, eventually thread stack will
781 // collide with heap.
782 //
783 // Here we try to prevent heap/stack collision by comparing current
784 // stack bottom with the highest address that has been mmap'ed by JVM
785 // plus a safety margin for memory maps created by native code.
786 //
787 // This feature can be disabled by setting ThreadSafetyMargin to 0
788 //
789 if (ThreadSafetyMargin > 0) {
790 address stack_bottom = os::current_stack_base() - os::current_stack_size();
791
792 // not safe if our stack extends below the safety margin
793 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
794 } else {
795 return true;
796 }
797 } else {
798 // Floating stack LinuxThreads or NPTL:
799 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
800 // there's not enough space left, pthread_create() will fail. If we come
801 // here, that means enough space has been reserved for stack.
802 return true;
803 }
804 }
805
806 // Thread start routine for all newly created threads
807 static void *java_start(Thread *thread) {
808 // Try to randomize the cache line index of hot stack frames.
809 // This helps when threads of the same stack traces evict each other's
810 // cache lines. The threads can be either from the same JVM instance, or
811 // from different JVM instances. The benefit is especially true for
812 // processors with hyperthreading technology.
813 static int counter = 0;
814 int pid = os::current_process_id();
815 alloca(((pid ^ counter++) & 7) * 128);
816
817 ThreadLocalStorage::set_thread(thread);
818
819 OSThread* osthread = thread->osthread();
820 Monitor* sync = osthread->startThread_lock();
821
822 // non floating stack LinuxThreads needs extra check, see above
823 if (!_thread_safety_check(thread)) {
824 // notify parent thread
825 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
826 osthread->set_state(ZOMBIE);
827 sync->notify_all();
828 return NULL;
829 }
830
831 // thread_id is kernel thread id (similar to Solaris LWP id)
832 osthread->set_thread_id(os::Linux::gettid());
833
834 if (UseNUMA) {
835 int lgrp_id = os::numa_get_group_id();
836 if (lgrp_id != -1) {
837 thread->set_lgrp_id(lgrp_id);
838 }
839 }
840 // initialize signal mask for this thread
841 os::Linux::hotspot_sigmask(thread);
842
843 // initialize floating point control register
844 os::Linux::init_thread_fpu_state();
845
846 // handshaking with parent thread
847 {
848 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
849
850 // notify parent thread
851 osthread->set_state(INITIALIZED);
852 sync->notify_all();
853
854 // wait until os::start_thread()
855 while (osthread->get_state() == INITIALIZED) {
856 sync->wait(Mutex::_no_safepoint_check_flag);
857 }
858 }
859
860 // call one more level start routine
861 thread->run();
862
863 return 0;
864 }
865
866 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
867 assert(thread->osthread() == NULL, "caller responsible");
868
869 // Allocate the OSThread object
870 OSThread* osthread = new OSThread(NULL, NULL);
871 if (osthread == NULL) {
872 return false;
873 }
874
875 // set the correct thread state
876 osthread->set_thread_type(thr_type);
877
878 // Initial state is ALLOCATED but not INITIALIZED
879 osthread->set_state(ALLOCATED);
880
881 thread->set_osthread(osthread);
882
883 // init thread attributes
884 pthread_attr_t attr;
885 pthread_attr_init(&attr);
886 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
887
888 // stack size
889 if (os::Linux::supports_variable_stack_size()) {
890 // calculate stack size if it's not specified by caller
891 if (stack_size == 0) {
892 stack_size = os::Linux::default_stack_size(thr_type);
893
894 switch (thr_type) {
895 case os::java_thread:
896 // Java threads use ThreadStackSize which default value can be
897 // changed with the flag -Xss
898 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
899 stack_size = JavaThread::stack_size_at_create();
900 break;
901 case os::compiler_thread:
902 if (CompilerThreadStackSize > 0) {
903 stack_size = (size_t)(CompilerThreadStackSize * K);
904 break;
905 } // else fall through:
906 // use VMThreadStackSize if CompilerThreadStackSize is not defined
907 case os::vm_thread:
908 case os::pgc_thread:
909 case os::cgc_thread:
910 case os::watcher_thread:
911 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
912 break;
913 }
914 }
915
916 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
917 pthread_attr_setstacksize(&attr, stack_size);
918 } else {
919 // let pthread_create() pick the default value.
920 }
921
922 // glibc guard page
923 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
924
925 ThreadState state;
926
927 {
928 // Serialize thread creation if we are running with fixed stack LinuxThreads
929 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
930 if (lock) {
931 os::Linux::createThread_lock()->lock_without_safepoint_check();
932 }
933
934 pthread_t tid;
935 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
936
937 pthread_attr_destroy(&attr);
938
939 if (ret != 0) {
940 if (PrintMiscellaneous && (Verbose || WizardMode)) {
941 perror("pthread_create()");
942 }
943 // Need to clean up stuff we've allocated so far
944 thread->set_osthread(NULL);
945 delete osthread;
946 if (lock) os::Linux::createThread_lock()->unlock();
947 return false;
948 }
949
950 // Store pthread info into the OSThread
951 osthread->set_pthread_id(tid);
952
953 // Wait until child thread is either initialized or aborted
954 {
955 Monitor* sync_with_child = osthread->startThread_lock();
956 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
957 while ((state = osthread->get_state()) == ALLOCATED) {
958 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
959 }
960 }
961
962 if (lock) {
963 os::Linux::createThread_lock()->unlock();
964 }
965 }
966
967 // Aborted due to thread limit being reached
968 if (state == ZOMBIE) {
969 thread->set_osthread(NULL);
970 delete osthread;
971 return false;
972 }
973
974 // The thread is returned suspended (in state INITIALIZED),
975 // and is started higher up in the call chain
976 assert(state == INITIALIZED, "race condition");
977 return true;
978 }
979
980 /////////////////////////////////////////////////////////////////////////////
981 // attach existing thread
982
983 // bootstrap the main thread
984 bool os::create_main_thread(JavaThread* thread) {
985 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
986 return create_attached_thread(thread);
987 }
988
989 bool os::create_attached_thread(JavaThread* thread) {
990 #ifdef ASSERT
991 thread->verify_not_published();
992 #endif
993
994 // Allocate the OSThread object
995 OSThread* osthread = new OSThread(NULL, NULL);
996
997 if (osthread == NULL) {
998 return false;
999 }
1000
1001 // Store pthread info into the OSThread
1002 osthread->set_thread_id(os::Linux::gettid());
1003 osthread->set_pthread_id(::pthread_self());
1004
1005 // initialize floating point control register
1006 os::Linux::init_thread_fpu_state();
1007
1008 // Initial thread state is RUNNABLE
1009 osthread->set_state(RUNNABLE);
1010
1011 thread->set_osthread(osthread);
1012
1013 if (UseNUMA) {
1014 int lgrp_id = os::numa_get_group_id();
1015 if (lgrp_id != -1) {
1016 thread->set_lgrp_id(lgrp_id);
1017 }
1018 }
1019
1020 if (os::Linux::is_initial_thread()) {
1021 // If current thread is initial thread, its stack is mapped on demand,
1022 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1023 // the entire stack region to avoid SEGV in stack banging.
1024 // It is also useful to get around the heap-stack-gap problem on SuSE
1025 // kernel (see 4821821 for details). We first expand stack to the top
1026 // of yellow zone, then enable stack yellow zone (order is significant,
1027 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1028 // is no gap between the last two virtual memory regions.
1029
1030 JavaThread *jt = (JavaThread *)thread;
1031 address addr = jt->stack_yellow_zone_base();
1032 assert(addr != NULL, "initialization problem?");
1033 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1034
1035 osthread->set_expanding_stack();
1036 os::Linux::manually_expand_stack(jt, addr);
1037 osthread->clear_expanding_stack();
1038 }
1039
1040 // initialize signal mask for this thread
1041 // and save the caller's signal mask
1042 os::Linux::hotspot_sigmask(thread);
1043
1044 return true;
1045 }
1046
1047 void os::pd_start_thread(Thread* thread) {
1048 OSThread * osthread = thread->osthread();
1049 assert(osthread->get_state() != INITIALIZED, "just checking");
1050 Monitor* sync_with_child = osthread->startThread_lock();
1051 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1052 sync_with_child->notify();
1053 }
1054
1055 // Free Linux resources related to the OSThread
1056 void os::free_thread(OSThread* osthread) {
1057 assert(osthread != NULL, "osthread not set");
1058
1059 if (Thread::current()->osthread() == osthread) {
1060 // Restore caller's signal mask
1061 sigset_t sigmask = osthread->caller_sigmask();
1062 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1063 }
1064
1065 delete osthread;
1066 }
1067
1068 //////////////////////////////////////////////////////////////////////////////
1069 // thread local storage
1070
1071 int os::allocate_thread_local_storage() {
1072 pthread_key_t key;
1073 int rslt = pthread_key_create(&key, NULL);
1074 assert(rslt == 0, "cannot allocate thread local storage");
1075 return (int)key;
1076 }
1077
1078 // Note: This is currently not used by VM, as we don't destroy TLS key
1079 // on VM exit.
1080 void os::free_thread_local_storage(int index) {
1081 int rslt = pthread_key_delete((pthread_key_t)index);
1082 assert(rslt == 0, "invalid index");
1083 }
1084
1085 void os::thread_local_storage_at_put(int index, void* value) {
1086 int rslt = pthread_setspecific((pthread_key_t)index, value);
1087 assert(rslt == 0, "pthread_setspecific failed");
1088 }
1089
1090 extern "C" Thread* get_thread() {
1091 return ThreadLocalStorage::thread();
1092 }
1093
1094 //////////////////////////////////////////////////////////////////////////////
1095 // initial thread
1096
1097 // Check if current thread is the initial thread, similar to Solaris thr_main.
1098 bool os::Linux::is_initial_thread(void) {
1099 char dummy;
1100 // If called before init complete, thread stack bottom will be null.
1101 // Can be called if fatal error occurs before initialization.
1102 if (initial_thread_stack_bottom() == NULL) return false;
1103 assert(initial_thread_stack_bottom() != NULL &&
1104 initial_thread_stack_size() != 0,
1105 "os::init did not locate initial thread's stack region");
1106 if ((address)&dummy >= initial_thread_stack_bottom() &&
1107 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1108 return true;
1109 else return false;
1110 }
1111
1112 // Find the virtual memory area that contains addr
1113 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1114 FILE *fp = fopen("/proc/self/maps", "r");
1115 if (fp) {
1116 address low, high;
1117 while (!feof(fp)) {
1118 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1119 if (low <= addr && addr < high) {
1120 if (vma_low) *vma_low = low;
1121 if (vma_high) *vma_high = high;
1122 fclose (fp);
1123 return true;
1124 }
1125 }
1126 for (;;) {
1127 int ch = fgetc(fp);
1128 if (ch == EOF || ch == (int)'\n') break;
1129 }
1130 }
1131 fclose(fp);
1132 }
1133 return false;
1134 }
1135
1136 // Locate initial thread stack. This special handling of initial thread stack
1137 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1138 // bogus value for initial thread.
1139 void os::Linux::capture_initial_stack(size_t max_size) {
1140 // stack size is the easy part, get it from RLIMIT_STACK
1141 size_t stack_size;
1142 struct rlimit rlim;
1143 getrlimit(RLIMIT_STACK, &rlim);
1144 stack_size = rlim.rlim_cur;
1145
1146 // 6308388: a bug in ld.so will relocate its own .data section to the
1147 // lower end of primordial stack; reduce ulimit -s value a little bit
1148 // so we won't install guard page on ld.so's data section.
1149 stack_size -= 2 * page_size();
1150
1151 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1152 // 7.1, in both cases we will get 2G in return value.
1153 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1154 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1155 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1156 // in case other parts in glibc still assumes 2M max stack size.
1157 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1158 #ifndef IA64
1159 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1160 #else
1161 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1162 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1163 #endif
1164
1165 // Try to figure out where the stack base (top) is. This is harder.
1166 //
1167 // When an application is started, glibc saves the initial stack pointer in
1168 // a global variable "__libc_stack_end", which is then used by system
1169 // libraries. __libc_stack_end should be pretty close to stack top. The
1170 // variable is available since the very early days. However, because it is
1171 // a private interface, it could disappear in the future.
1172 //
1173 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1174 // to __libc_stack_end, it is very close to stack top, but isn't the real
1175 // stack top. Note that /proc may not exist if VM is running as a chroot
1176 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1177 // /proc/<pid>/stat could change in the future (though unlikely).
1178 //
1179 // We try __libc_stack_end first. If that doesn't work, look for
1180 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1181 // as a hint, which should work well in most cases.
1182
1183 uintptr_t stack_start;
1184
1185 // try __libc_stack_end first
1186 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1187 if (p && *p) {
1188 stack_start = *p;
1189 } else {
1190 // see if we can get the start_stack field from /proc/self/stat
1191 FILE *fp;
1192 int pid;
1193 char state;
1194 int ppid;
1195 int pgrp;
1196 int session;
1197 int nr;
1198 int tpgrp;
1199 unsigned long flags;
1200 unsigned long minflt;
1201 unsigned long cminflt;
1202 unsigned long majflt;
1203 unsigned long cmajflt;
1204 unsigned long utime;
1205 unsigned long stime;
1206 long cutime;
1207 long cstime;
1208 long prio;
1209 long nice;
1210 long junk;
1211 long it_real;
1212 uintptr_t start;
1213 uintptr_t vsize;
1214 intptr_t rss;
1215 uintptr_t rsslim;
1216 uintptr_t scodes;
1217 uintptr_t ecode;
1218 int i;
1219
1220 // Figure what the primordial thread stack base is. Code is inspired
1221 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1222 // followed by command name surrounded by parentheses, state, etc.
1223 char stat[2048];
1224 int statlen;
1225
1226 fp = fopen("/proc/self/stat", "r");
1227 if (fp) {
1228 statlen = fread(stat, 1, 2047, fp);
1229 stat[statlen] = '\0';
1230 fclose(fp);
1231
1232 // Skip pid and the command string. Note that we could be dealing with
1233 // weird command names, e.g. user could decide to rename java launcher
1234 // to "java 1.4.2 :)", then the stat file would look like
1235 // 1234 (java 1.4.2 :)) R ... ...
1236 // We don't really need to know the command string, just find the last
1237 // occurrence of ")" and then start parsing from there. See bug 4726580.
1238 char * s = strrchr(stat, ')');
1239
1240 i = 0;
1241 if (s) {
1242 // Skip blank chars
1243 do s++; while (isspace(*s));
1244
1245 #define _UFM UINTX_FORMAT
1246 #define _DFM INTX_FORMAT
1247
1248 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1249 /* 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 */
1250 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,
1251 &state, /* 3 %c */
1252 &ppid, /* 4 %d */
1253 &pgrp, /* 5 %d */
1254 &session, /* 6 %d */
1255 &nr, /* 7 %d */
1256 &tpgrp, /* 8 %d */
1257 &flags, /* 9 %lu */
1258 &minflt, /* 10 %lu */
1259 &cminflt, /* 11 %lu */
1260 &majflt, /* 12 %lu */
1261 &cmajflt, /* 13 %lu */
1262 &utime, /* 14 %lu */
1263 &stime, /* 15 %lu */
1264 &cutime, /* 16 %ld */
1265 &cstime, /* 17 %ld */
1266 &prio, /* 18 %ld */
1267 &nice, /* 19 %ld */
1268 &junk, /* 20 %ld */
1269 &it_real, /* 21 %ld */
1270 &start, /* 22 UINTX_FORMAT */
1271 &vsize, /* 23 UINTX_FORMAT */
1272 &rss, /* 24 INTX_FORMAT */
1273 &rsslim, /* 25 UINTX_FORMAT */
1274 &scodes, /* 26 UINTX_FORMAT */
1275 &ecode, /* 27 UINTX_FORMAT */
1276 &stack_start); /* 28 UINTX_FORMAT */
1277 }
1278
1279 #undef _UFM
1280 #undef _DFM
1281
1282 if (i != 28 - 2) {
1283 assert(false, "Bad conversion from /proc/self/stat");
1284 // product mode - assume we are the initial thread, good luck in the
1285 // embedded case.
1286 warning("Can't detect initial thread stack location - bad conversion");
1287 stack_start = (uintptr_t) &rlim;
1288 }
1289 } else {
1290 // For some reason we can't open /proc/self/stat (for example, running on
1291 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1292 // most cases, so don't abort:
1293 warning("Can't detect initial thread stack location - no /proc/self/stat");
1294 stack_start = (uintptr_t) &rlim;
1295 }
1296 }
1297
1298 // Now we have a pointer (stack_start) very close to the stack top, the
1299 // next thing to do is to figure out the exact location of stack top. We
1300 // can find out the virtual memory area that contains stack_start by
1301 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1302 // and its upper limit is the real stack top. (again, this would fail if
1303 // running inside chroot, because /proc may not exist.)
1304
1305 uintptr_t stack_top;
1306 address low, high;
1307 if (find_vma((address)stack_start, &low, &high)) {
1308 // success, "high" is the true stack top. (ignore "low", because initial
1309 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1310 stack_top = (uintptr_t)high;
1311 } else {
1312 // failed, likely because /proc/self/maps does not exist
1313 warning("Can't detect initial thread stack location - find_vma failed");
1314 // best effort: stack_start is normally within a few pages below the real
1315 // stack top, use it as stack top, and reduce stack size so we won't put
1316 // guard page outside stack.
1317 stack_top = stack_start;
1318 stack_size -= 16 * page_size();
1319 }
1320
1321 // stack_top could be partially down the page so align it
1322 stack_top = align_size_up(stack_top, page_size());
1323
1324 if (max_size && stack_size > max_size) {
1325 _initial_thread_stack_size = max_size;
1326 } else {
1327 _initial_thread_stack_size = stack_size;
1328 }
1329
1330 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1331 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1332 }
1333
1334 ////////////////////////////////////////////////////////////////////////////////
1335 // time support
1336
1337 // Time since start-up in seconds to a fine granularity.
1338 // Used by VMSelfDestructTimer and the MemProfiler.
1339 double os::elapsedTime() {
1340
1341 return (double)(os::elapsed_counter()) * 0.000001;
1342 }
1343
1344 jlong os::elapsed_counter() {
1345 timeval time;
1346 int status = gettimeofday(&time, NULL);
1347 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1348 }
1349
1350 jlong os::elapsed_frequency() {
1351 return (1000 * 1000);
1352 }
1353
1354 // For now, we say that linux does not support vtime. I have no idea
1355 // whether it can actually be made to (DLD, 9/13/05).
1356
1357 bool os::supports_vtime() { return false; }
1358 bool os::enable_vtime() { return false; }
1359 bool os::vtime_enabled() { return false; }
1360 double os::elapsedVTime() {
1361 // better than nothing, but not much
1362 return elapsedTime();
1363 }
1364
1365 jlong os::javaTimeMillis() {
1366 timeval time;
1367 int status = gettimeofday(&time, NULL);
1368 assert(status != -1, "linux error");
1369 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1370 }
1371
1372 #ifndef CLOCK_MONOTONIC
1373 #define CLOCK_MONOTONIC (1)
1374 #endif
1375
1376 void os::Linux::clock_init() {
1377 // we do dlopen's in this particular order due to bug in linux
1378 // dynamical loader (see 6348968) leading to crash on exit
1379 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1380 if (handle == NULL) {
1381 handle = dlopen("librt.so", RTLD_LAZY);
1382 }
1383
1384 if (handle) {
1385 int (*clock_getres_func)(clockid_t, struct timespec*) =
1386 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1387 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1388 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1389 if (clock_getres_func && clock_gettime_func) {
1390 // See if monotonic clock is supported by the kernel. Note that some
1391 // early implementations simply return kernel jiffies (updated every
1392 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1393 // for nano time (though the monotonic property is still nice to have).
1394 // It's fixed in newer kernels, however clock_getres() still returns
1395 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1396 // resolution for now. Hopefully as people move to new kernels, this
1397 // won't be a problem.
1398 struct timespec res;
1399 struct timespec tp;
1400 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1401 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1402 // yes, monotonic clock is supported
1403 _clock_gettime = clock_gettime_func;
1404 } else {
1405 // close librt if there is no monotonic clock
1406 dlclose(handle);
1407 }
1408 }
1409 }
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 void os::dll_build_name(char* buffer, size_t buflen,
1635 const char* pname, const char* fname) {
1636 // Copied from libhpi
1637 const size_t pnamelen = pname ? strlen(pname) : 0;
1638
1639 // Quietly truncate on buffer overflow. Should be an error.
1640 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1641 *buffer = '\0';
1642 return;
1643 }
1644
1645 if (pnamelen == 0) {
1646 snprintf(buffer, buflen, "lib%s.so", fname);
1647 } else if (strchr(pname, *os::path_separator()) != NULL) {
1648 int n;
1649 char** pelements = split_path(pname, &n);
1650 for (int i = 0 ; i < n ; i++) {
1651 // Really shouldn't be NULL, but check can't hurt
1652 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1653 continue; // skip the empty path values
1654 }
1655 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1656 if (file_exists(buffer)) {
1657 break;
1658 }
1659 }
1660 // release the storage
1661 for (int i = 0 ; i < n ; i++) {
1662 if (pelements[i] != NULL) {
1663 FREE_C_HEAP_ARRAY(char, pelements[i]);
1664 }
1665 }
1666 if (pelements != NULL) {
1667 FREE_C_HEAP_ARRAY(char*, pelements);
1668 }
1669 } else {
1670 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1671 }
1672 }
1673
1674 const char* os::get_current_directory(char *buf, int buflen) {
1675 return getcwd(buf, buflen);
1676 }
1677
1678 // check if addr is inside libjvm[_g].so
1679 bool os::address_is_in_vm(address addr) {
1680 static address libjvm_base_addr;
1681 Dl_info dlinfo;
1682
1683 if (libjvm_base_addr == NULL) {
1684 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1685 libjvm_base_addr = (address)dlinfo.dli_fbase;
1686 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1687 }
1688
1689 if (dladdr((void *)addr, &dlinfo)) {
1690 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1691 }
1692
1693 return false;
1694 }
1695
1696 bool os::dll_address_to_function_name(address addr, char *buf,
1697 int buflen, int *offset) {
1698 Dl_info dlinfo;
1699
1700 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1701 if (buf != NULL) {
1702 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1703 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1704 }
1705 }
1706 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1707 return true;
1708 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1709 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1710 dlinfo.dli_fname, buf, buflen, offset) == Decoder::no_error) {
1711 return true;
1712 }
1713 }
1714
1715 if (buf != NULL) buf[0] = '\0';
1716 if (offset != NULL) *offset = -1;
1717 return false;
1718 }
1719
1720 struct _address_to_library_name {
1721 address addr; // input : memory address
1722 size_t buflen; // size of fname
1723 char* fname; // output: library name
1724 address base; // library base addr
1725 };
1726
1727 static int address_to_library_name_callback(struct dl_phdr_info *info,
1728 size_t size, void *data) {
1729 int i;
1730 bool found = false;
1731 address libbase = NULL;
1732 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1733
1734 // iterate through all loadable segments
1735 for (i = 0; i < info->dlpi_phnum; i++) {
1736 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1737 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1738 // base address of a library is the lowest address of its loaded
1739 // segments.
1740 if (libbase == NULL || libbase > segbase) {
1741 libbase = segbase;
1742 }
1743 // see if 'addr' is within current segment
1744 if (segbase <= d->addr &&
1745 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1746 found = true;
1747 }
1748 }
1749 }
1750
1751 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1752 // so dll_address_to_library_name() can fall through to use dladdr() which
1753 // can figure out executable name from argv[0].
1754 if (found && info->dlpi_name && info->dlpi_name[0]) {
1755 d->base = libbase;
1756 if (d->fname) {
1757 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1758 }
1759 return 1;
1760 }
1761 return 0;
1762 }
1763
1764 bool os::dll_address_to_library_name(address addr, char* buf,
1765 int buflen, int* offset) {
1766 Dl_info dlinfo;
1767 struct _address_to_library_name data;
1768
1769 // There is a bug in old glibc dladdr() implementation that it could resolve
1770 // to wrong library name if the .so file has a base address != NULL. Here
1771 // we iterate through the program headers of all loaded libraries to find
1772 // out which library 'addr' really belongs to. This workaround can be
1773 // removed once the minimum requirement for glibc is moved to 2.3.x.
1774 data.addr = addr;
1775 data.fname = buf;
1776 data.buflen = buflen;
1777 data.base = NULL;
1778 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1779
1780 if (rslt) {
1781 // buf already contains library name
1782 if (offset) *offset = addr - data.base;
1783 return true;
1784 } else if (dladdr((void*)addr, &dlinfo)){
1785 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1786 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1787 return true;
1788 } else {
1789 if (buf) buf[0] = '\0';
1790 if (offset) *offset = -1;
1791 return false;
1792 }
1793 }
1794
1795 // Loads .dll/.so and
1796 // in case of error it checks if .dll/.so was built for the
1797 // same architecture as Hotspot is running on
1798
1799 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1800 {
1801 void * result= ::dlopen(filename, RTLD_LAZY);
1802 if (result != NULL) {
1803 // Successful loading
1804 return result;
1805 }
1806
1807 Elf32_Ehdr elf_head;
1808
1809 // Read system error message into ebuf
1810 // It may or may not be overwritten below
1811 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1812 ebuf[ebuflen-1]='\0';
1813 int diag_msg_max_length=ebuflen-strlen(ebuf);
1814 char* diag_msg_buf=ebuf+strlen(ebuf);
1815
1816 if (diag_msg_max_length==0) {
1817 // No more space in ebuf for additional diagnostics message
1818 return NULL;
1819 }
1820
1821
1822 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1823
1824 if (file_descriptor < 0) {
1825 // Can't open library, report dlerror() message
1826 return NULL;
1827 }
1828
1829 bool failed_to_read_elf_head=
1830 (sizeof(elf_head)!=
1831 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1832
1833 ::close(file_descriptor);
1834 if (failed_to_read_elf_head) {
1835 // file i/o error - report dlerror() msg
1836 return NULL;
1837 }
1838
1839 typedef struct {
1840 Elf32_Half code; // Actual value as defined in elf.h
1841 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1842 char elf_class; // 32 or 64 bit
1843 char endianess; // MSB or LSB
1844 char* name; // String representation
1845 } arch_t;
1846
1847 #ifndef EM_486
1848 #define EM_486 6 /* Intel 80486 */
1849 #endif
1850
1851 static const arch_t arch_array[]={
1852 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1853 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1854 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1855 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1856 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1857 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1858 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1859 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1860 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1861 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1862 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1863 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1864 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1865 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1866 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1867 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1868 };
1869
1870 #if (defined IA32)
1871 static Elf32_Half running_arch_code=EM_386;
1872 #elif (defined AMD64)
1873 static Elf32_Half running_arch_code=EM_X86_64;
1874 #elif (defined IA64)
1875 static Elf32_Half running_arch_code=EM_IA_64;
1876 #elif (defined __sparc) && (defined _LP64)
1877 static Elf32_Half running_arch_code=EM_SPARCV9;
1878 #elif (defined __sparc) && (!defined _LP64)
1879 static Elf32_Half running_arch_code=EM_SPARC;
1880 #elif (defined __powerpc64__)
1881 static Elf32_Half running_arch_code=EM_PPC64;
1882 #elif (defined __powerpc__)
1883 static Elf32_Half running_arch_code=EM_PPC;
1884 #elif (defined ARM)
1885 static Elf32_Half running_arch_code=EM_ARM;
1886 #elif (defined S390)
1887 static Elf32_Half running_arch_code=EM_S390;
1888 #elif (defined ALPHA)
1889 static Elf32_Half running_arch_code=EM_ALPHA;
1890 #elif (defined MIPSEL)
1891 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1892 #elif (defined PARISC)
1893 static Elf32_Half running_arch_code=EM_PARISC;
1894 #elif (defined MIPS)
1895 static Elf32_Half running_arch_code=EM_MIPS;
1896 #elif (defined M68K)
1897 static Elf32_Half running_arch_code=EM_68K;
1898 #else
1899 #error Method os::dll_load requires that one of following is defined:\
1900 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1901 #endif
1902
1903 // Identify compatability class for VM's architecture and library's architecture
1904 // Obtain string descriptions for architectures
1905
1906 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1907 int running_arch_index=-1;
1908
1909 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1910 if (running_arch_code == arch_array[i].code) {
1911 running_arch_index = i;
1912 }
1913 if (lib_arch.code == arch_array[i].code) {
1914 lib_arch.compat_class = arch_array[i].compat_class;
1915 lib_arch.name = arch_array[i].name;
1916 }
1917 }
1918
1919 assert(running_arch_index != -1,
1920 "Didn't find running architecture code (running_arch_code) in arch_array");
1921 if (running_arch_index == -1) {
1922 // Even though running architecture detection failed
1923 // we may still continue with reporting dlerror() message
1924 return NULL;
1925 }
1926
1927 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1928 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1929 return NULL;
1930 }
1931
1932 #ifndef S390
1933 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1934 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1935 return NULL;
1936 }
1937 #endif // !S390
1938
1939 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1940 if ( lib_arch.name!=NULL ) {
1941 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1942 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1943 lib_arch.name, arch_array[running_arch_index].name);
1944 } else {
1945 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1946 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1947 lib_arch.code,
1948 arch_array[running_arch_index].name);
1949 }
1950 }
1951
1952 return NULL;
1953 }
1954
1955 /*
1956 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1957 * chances are you might want to run the generated bits against glibc-2.0
1958 * libdl.so, so always use locking for any version of glibc.
1959 */
1960 void* os::dll_lookup(void* handle, const char* name) {
1961 pthread_mutex_lock(&dl_mutex);
1962 void* res = dlsym(handle, name);
1963 pthread_mutex_unlock(&dl_mutex);
1964 return res;
1965 }
1966
1967
1968 static bool _print_ascii_file(const char* filename, outputStream* st) {
1969 int fd = ::open(filename, O_RDONLY);
1970 if (fd == -1) {
1971 return false;
1972 }
1973
1974 char buf[32];
1975 int bytes;
1976 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1977 st->print_raw(buf, bytes);
1978 }
1979
1980 ::close(fd);
1981
1982 return true;
1983 }
1984
1985 void os::print_dll_info(outputStream *st) {
1986 st->print_cr("Dynamic libraries:");
1987
1988 char fname[32];
1989 pid_t pid = os::Linux::gettid();
1990
1991 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1992
1993 if (!_print_ascii_file(fname, st)) {
1994 st->print("Can not get library information for pid = %d\n", pid);
1995 }
1996 }
1997
1998
1999 void os::print_os_info(outputStream* st) {
2000 st->print("OS:");
2001
2002 // Try to identify popular distros.
2003 // Most Linux distributions have /etc/XXX-release file, which contains
2004 // the OS version string. Some have more than one /etc/XXX-release file
2005 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2006 // so the order is important.
2007 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2008 !_print_ascii_file("/etc/sun-release", st) &&
2009 !_print_ascii_file("/etc/redhat-release", st) &&
2010 !_print_ascii_file("/etc/SuSE-release", st) &&
2011 !_print_ascii_file("/etc/turbolinux-release", st) &&
2012 !_print_ascii_file("/etc/gentoo-release", st) &&
2013 !_print_ascii_file("/etc/debian_version", st) &&
2014 !_print_ascii_file("/etc/ltib-release", st) &&
2015 !_print_ascii_file("/etc/angstrom-version", st)) {
2016 st->print("Linux");
2017 }
2018 st->cr();
2019
2020 // kernel
2021 st->print("uname:");
2022 struct utsname name;
2023 uname(&name);
2024 st->print(name.sysname); st->print(" ");
2025 st->print(name.release); st->print(" ");
2026 st->print(name.version); st->print(" ");
2027 st->print(name.machine);
2028 st->cr();
2029
2030 // Print warning if unsafe chroot environment detected
2031 if (unsafe_chroot_detected) {
2032 st->print("WARNING!! ");
2033 st->print_cr(unstable_chroot_error);
2034 }
2035
2036 // libc, pthread
2037 st->print("libc:");
2038 st->print(os::Linux::glibc_version()); st->print(" ");
2039 st->print(os::Linux::libpthread_version()); st->print(" ");
2040 if (os::Linux::is_LinuxThreads()) {
2041 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2042 }
2043 st->cr();
2044
2045 // rlimit
2046 st->print("rlimit:");
2047 struct rlimit rlim;
2048
2049 st->print(" STACK ");
2050 getrlimit(RLIMIT_STACK, &rlim);
2051 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2052 else st->print("%uk", rlim.rlim_cur >> 10);
2053
2054 st->print(", CORE ");
2055 getrlimit(RLIMIT_CORE, &rlim);
2056 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2057 else st->print("%uk", rlim.rlim_cur >> 10);
2058
2059 st->print(", NPROC ");
2060 getrlimit(RLIMIT_NPROC, &rlim);
2061 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2062 else st->print("%d", rlim.rlim_cur);
2063
2064 st->print(", NOFILE ");
2065 getrlimit(RLIMIT_NOFILE, &rlim);
2066 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2067 else st->print("%d", rlim.rlim_cur);
2068
2069 st->print(", AS ");
2070 getrlimit(RLIMIT_AS, &rlim);
2071 if (rlim.rlim_cur == RLIM_INFINITY) st->print("infinity");
2072 else st->print("%uk", rlim.rlim_cur >> 10);
2073 st->cr();
2074
2075 // load average
2076 st->print("load average:");
2077 double loadavg[3];
2078 os::loadavg(loadavg, 3);
2079 st->print("%0.02f %0.02f %0.02f", loadavg[0], loadavg[1], loadavg[2]);
2080 st->cr();
2081
2082 // meminfo
2083 st->print("\n/proc/meminfo:\n");
2084 _print_ascii_file("/proc/meminfo", st);
2085 st->cr();
2086 }
2087
2088 void os::print_memory_info(outputStream* st) {
2089
2090 st->print("Memory:");
2091 st->print(" %dk page", os::vm_page_size()>>10);
2092
2093 // values in struct sysinfo are "unsigned long"
2094 struct sysinfo si;
2095 sysinfo(&si);
2096
2097 st->print(", physical " UINT64_FORMAT "k",
2098 os::physical_memory() >> 10);
2099 st->print("(" UINT64_FORMAT "k free)",
2100 os::available_memory() >> 10);
2101 st->print(", swap " UINT64_FORMAT "k",
2102 ((jlong)si.totalswap * si.mem_unit) >> 10);
2103 st->print("(" UINT64_FORMAT "k free)",
2104 ((jlong)si.freeswap * si.mem_unit) >> 10);
2105 st->cr();
2106 }
2107
2108 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2109 // but they're the same for all the linux arch that we support
2110 // and they're the same for solaris but there's no common place to put this.
2111 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2112 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2113 "ILL_COPROC", "ILL_BADSTK" };
2114
2115 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2116 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2117 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2118
2119 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2120
2121 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2122
2123 void os::print_siginfo(outputStream* st, void* siginfo) {
2124 st->print("siginfo:");
2125
2126 const int buflen = 100;
2127 char buf[buflen];
2128 siginfo_t *si = (siginfo_t*)siginfo;
2129 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2130 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2131 st->print("si_errno=%s", buf);
2132 } else {
2133 st->print("si_errno=%d", si->si_errno);
2134 }
2135 const int c = si->si_code;
2136 assert(c > 0, "unexpected si_code");
2137 switch (si->si_signo) {
2138 case SIGILL:
2139 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2140 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2141 break;
2142 case SIGFPE:
2143 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2144 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2145 break;
2146 case SIGSEGV:
2147 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2148 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2149 break;
2150 case SIGBUS:
2151 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2152 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2153 break;
2154 default:
2155 st->print(", si_code=%d", si->si_code);
2156 // no si_addr
2157 }
2158
2159 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2160 UseSharedSpaces) {
2161 FileMapInfo* mapinfo = FileMapInfo::current_info();
2162 if (mapinfo->is_in_shared_space(si->si_addr)) {
2163 st->print("\n\nError accessing class data sharing archive." \
2164 " Mapped file inaccessible during execution, " \
2165 " possible disk/network problem.");
2166 }
2167 }
2168 st->cr();
2169 }
2170
2171
2172 static void print_signal_handler(outputStream* st, int sig,
2173 char* buf, size_t buflen);
2174
2175 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2176 st->print_cr("Signal Handlers:");
2177 print_signal_handler(st, SIGSEGV, buf, buflen);
2178 print_signal_handler(st, SIGBUS , buf, buflen);
2179 print_signal_handler(st, SIGFPE , buf, buflen);
2180 print_signal_handler(st, SIGPIPE, buf, buflen);
2181 print_signal_handler(st, SIGXFSZ, buf, buflen);
2182 print_signal_handler(st, SIGILL , buf, buflen);
2183 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2184 print_signal_handler(st, SR_signum, buf, buflen);
2185 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2186 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2187 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2188 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2189 }
2190
2191 static char saved_jvm_path[MAXPATHLEN] = {0};
2192
2193 // Find the full path to the current module, libjvm.so or libjvm_g.so
2194 void os::jvm_path(char *buf, jint buflen) {
2195 // Error checking.
2196 if (buflen < MAXPATHLEN) {
2197 assert(false, "must use a large-enough buffer");
2198 buf[0] = '\0';
2199 return;
2200 }
2201 // Lazy resolve the path to current module.
2202 if (saved_jvm_path[0] != 0) {
2203 strcpy(buf, saved_jvm_path);
2204 return;
2205 }
2206
2207 char dli_fname[MAXPATHLEN];
2208 bool ret = dll_address_to_library_name(
2209 CAST_FROM_FN_PTR(address, os::jvm_path),
2210 dli_fname, sizeof(dli_fname), NULL);
2211 assert(ret != 0, "cannot locate libjvm");
2212 char *rp = realpath(dli_fname, buf);
2213 if (rp == NULL)
2214 return;
2215
2216 if (Arguments::created_by_gamma_launcher()) {
2217 // Support for the gamma launcher. Typical value for buf is
2218 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2219 // the right place in the string, then assume we are installed in a JDK and
2220 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2221 // up the path so it looks like libjvm.so is installed there (append a
2222 // fake suffix hotspot/libjvm.so).
2223 const char *p = buf + strlen(buf) - 1;
2224 for (int count = 0; p > buf && count < 5; ++count) {
2225 for (--p; p > buf && *p != '/'; --p)
2226 /* empty */ ;
2227 }
2228
2229 if (strncmp(p, "/jre/lib/", 9) != 0) {
2230 // Look for JAVA_HOME in the environment.
2231 char* java_home_var = ::getenv("JAVA_HOME");
2232 if (java_home_var != NULL && java_home_var[0] != 0) {
2233 char* jrelib_p;
2234 int len;
2235
2236 // Check the current module name "libjvm.so" or "libjvm_g.so".
2237 p = strrchr(buf, '/');
2238 assert(strstr(p, "/libjvm") == p, "invalid library name");
2239 p = strstr(p, "_g") ? "_g" : "";
2240
2241 rp = realpath(java_home_var, buf);
2242 if (rp == NULL)
2243 return;
2244
2245 // determine if this is a legacy image or modules image
2246 // modules image doesn't have "jre" subdirectory
2247 len = strlen(buf);
2248 jrelib_p = buf + len;
2249 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2250 if (0 != access(buf, F_OK)) {
2251 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2252 }
2253
2254 if (0 == access(buf, F_OK)) {
2255 // Use current module name "libjvm[_g].so" instead of
2256 // "libjvm"debug_only("_g")".so" since for fastdebug version
2257 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2258 len = strlen(buf);
2259 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2260 } else {
2261 // Go back to path of .so
2262 rp = realpath(dli_fname, buf);
2263 if (rp == NULL)
2264 return;
2265 }
2266 }
2267 }
2268 }
2269
2270 strcpy(saved_jvm_path, buf);
2271 }
2272
2273 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2274 // no prefix required, not even "_"
2275 }
2276
2277 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2278 // no suffix required
2279 }
2280
2281 ////////////////////////////////////////////////////////////////////////////////
2282 // sun.misc.Signal support
2283
2284 static volatile jint sigint_count = 0;
2285
2286 static void
2287 UserHandler(int sig, void *siginfo, void *context) {
2288 // 4511530 - sem_post is serialized and handled by the manager thread. When
2289 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2290 // don't want to flood the manager thread with sem_post requests.
2291 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2292 return;
2293
2294 // Ctrl-C is pressed during error reporting, likely because the error
2295 // handler fails to abort. Let VM die immediately.
2296 if (sig == SIGINT && is_error_reported()) {
2297 os::die();
2298 }
2299
2300 os::signal_notify(sig);
2301 }
2302
2303 void* os::user_handler() {
2304 return CAST_FROM_FN_PTR(void*, UserHandler);
2305 }
2306
2307 extern "C" {
2308 typedef void (*sa_handler_t)(int);
2309 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2310 }
2311
2312 void* os::signal(int signal_number, void* handler) {
2313 struct sigaction sigAct, oldSigAct;
2314
2315 sigfillset(&(sigAct.sa_mask));
2316 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2317 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2318
2319 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2320 // -1 means registration failed
2321 return (void *)-1;
2322 }
2323
2324 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2325 }
2326
2327 void os::signal_raise(int signal_number) {
2328 ::raise(signal_number);
2329 }
2330
2331 /*
2332 * The following code is moved from os.cpp for making this
2333 * code platform specific, which it is by its very nature.
2334 */
2335
2336 // Will be modified when max signal is changed to be dynamic
2337 int os::sigexitnum_pd() {
2338 return NSIG;
2339 }
2340
2341 // a counter for each possible signal value
2342 static volatile jint pending_signals[NSIG+1] = { 0 };
2343
2344 // Linux(POSIX) specific hand shaking semaphore.
2345 static sem_t sig_sem;
2346
2347 void os::signal_init_pd() {
2348 // Initialize signal structures
2349 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2350
2351 // Initialize signal semaphore
2352 ::sem_init(&sig_sem, 0, 0);
2353 }
2354
2355 void os::signal_notify(int sig) {
2356 Atomic::inc(&pending_signals[sig]);
2357 ::sem_post(&sig_sem);
2358 }
2359
2360 static int check_pending_signals(bool wait) {
2361 Atomic::store(0, &sigint_count);
2362 for (;;) {
2363 for (int i = 0; i < NSIG + 1; i++) {
2364 jint n = pending_signals[i];
2365 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2366 return i;
2367 }
2368 }
2369 if (!wait) {
2370 return -1;
2371 }
2372 JavaThread *thread = JavaThread::current();
2373 ThreadBlockInVM tbivm(thread);
2374
2375 bool threadIsSuspended;
2376 do {
2377 thread->set_suspend_equivalent();
2378 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2379 ::sem_wait(&sig_sem);
2380
2381 // were we externally suspended while we were waiting?
2382 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2383 if (threadIsSuspended) {
2384 //
2385 // The semaphore has been incremented, but while we were waiting
2386 // another thread suspended us. We don't want to continue running
2387 // while suspended because that would surprise the thread that
2388 // suspended us.
2389 //
2390 ::sem_post(&sig_sem);
2391
2392 thread->java_suspend_self();
2393 }
2394 } while (threadIsSuspended);
2395 }
2396 }
2397
2398 int os::signal_lookup() {
2399 return check_pending_signals(false);
2400 }
2401
2402 int os::signal_wait() {
2403 return check_pending_signals(true);
2404 }
2405
2406 ////////////////////////////////////////////////////////////////////////////////
2407 // Virtual Memory
2408
2409 int os::vm_page_size() {
2410 // Seems redundant as all get out
2411 assert(os::Linux::page_size() != -1, "must call os::init");
2412 return os::Linux::page_size();
2413 }
2414
2415 // Solaris allocates memory by pages.
2416 int os::vm_allocation_granularity() {
2417 assert(os::Linux::page_size() != -1, "must call os::init");
2418 return os::Linux::page_size();
2419 }
2420
2421 // Rationale behind this function:
2422 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2423 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2424 // samples for JITted code. Here we create private executable mapping over the code cache
2425 // and then we can use standard (well, almost, as mapping can change) way to provide
2426 // info for the reporting script by storing timestamp and location of symbol
2427 void linux_wrap_code(char* base, size_t size) {
2428 static volatile jint cnt = 0;
2429
2430 if (!UseOprofile) {
2431 return;
2432 }
2433
2434 char buf[PATH_MAX+1];
2435 int num = Atomic::add(1, &cnt);
2436
2437 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2438 os::get_temp_directory(), os::current_process_id(), num);
2439 unlink(buf);
2440
2441 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2442
2443 if (fd != -1) {
2444 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2445 if (rv != (off_t)-1) {
2446 if (::write(fd, "", 1) == 1) {
2447 mmap(base, size,
2448 PROT_READ|PROT_WRITE|PROT_EXEC,
2449 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2450 }
2451 }
2452 ::close(fd);
2453 unlink(buf);
2454 }
2455 }
2456
2457 // NOTE: Linux kernel does not really reserve the pages for us.
2458 // All it does is to check if there are enough free pages
2459 // left at the time of mmap(). This could be a potential
2460 // problem.
2461 bool os::commit_memory(char* addr, size_t size, bool exec) {
2462 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2463 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2464 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2465 return res != (uintptr_t) MAP_FAILED;
2466 }
2467
2468 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2469 #ifndef MAP_HUGETLB
2470 #define MAP_HUGETLB 0x40000
2471 #endif
2472
2473 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2474 #ifndef MADV_HUGEPAGE
2475 #define MADV_HUGEPAGE 14
2476 #endif
2477
2478 bool os::commit_memory(char* addr, size_t size, size_t alignment_hint,
2479 bool exec) {
2480 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2481 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2482 uintptr_t res =
2483 (uintptr_t) ::mmap(addr, size, prot,
2484 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2485 -1, 0);
2486 return res != (uintptr_t) MAP_FAILED;
2487 }
2488
2489 return commit_memory(addr, size, exec);
2490 }
2491
2492 void os::realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2493 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2494 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2495 // be supported or the memory may already be backed by huge pages.
2496 ::madvise(addr, bytes, MADV_HUGEPAGE);
2497 }
2498 }
2499
2500 void os::free_memory(char *addr, size_t bytes) {
2501 ::madvise(addr, bytes, MADV_DONTNEED);
2502 }
2503
2504 void os::numa_make_global(char *addr, size_t bytes) {
2505 Linux::numa_interleave_memory(addr, bytes);
2506 }
2507
2508 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2509 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2510 }
2511
2512 bool os::numa_topology_changed() { return false; }
2513
2514 size_t os::numa_get_groups_num() {
2515 int max_node = Linux::numa_max_node();
2516 return max_node > 0 ? max_node + 1 : 1;
2517 }
2518
2519 int os::numa_get_group_id() {
2520 int cpu_id = Linux::sched_getcpu();
2521 if (cpu_id != -1) {
2522 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2523 if (lgrp_id != -1) {
2524 return lgrp_id;
2525 }
2526 }
2527 return 0;
2528 }
2529
2530 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2531 for (size_t i = 0; i < size; i++) {
2532 ids[i] = i;
2533 }
2534 return size;
2535 }
2536
2537 bool os::get_page_info(char *start, page_info* info) {
2538 return false;
2539 }
2540
2541 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2542 return end;
2543 }
2544
2545 // Something to do with the numa-aware allocator needs these symbols
2546 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2547 extern "C" JNIEXPORT void numa_error(char *where) { }
2548 extern "C" JNIEXPORT int fork1() { return fork(); }
2549
2550
2551 // If we are running with libnuma version > 2, then we should
2552 // be trying to use symbols with versions 1.1
2553 // If we are running with earlier version, which did not have symbol versions,
2554 // we should use the base version.
2555 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2556 void *f = dlvsym(handle, name, "libnuma_1.1");
2557 if (f == NULL) {
2558 f = dlsym(handle, name);
2559 }
2560 return f;
2561 }
2562
2563 bool os::Linux::libnuma_init() {
2564 // sched_getcpu() should be in libc.
2565 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2566 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2567
2568 if (sched_getcpu() != -1) { // Does it work?
2569 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2570 if (handle != NULL) {
2571 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2572 libnuma_dlsym(handle, "numa_node_to_cpus")));
2573 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2574 libnuma_dlsym(handle, "numa_max_node")));
2575 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2576 libnuma_dlsym(handle, "numa_available")));
2577 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2578 libnuma_dlsym(handle, "numa_tonode_memory")));
2579 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2580 libnuma_dlsym(handle, "numa_interleave_memory")));
2581
2582
2583 if (numa_available() != -1) {
2584 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2585 // Create a cpu -> node mapping
2586 _cpu_to_node = new (ResourceObj::C_HEAP) GrowableArray<int>(0, true);
2587 rebuild_cpu_to_node_map();
2588 return true;
2589 }
2590 }
2591 }
2592 return false;
2593 }
2594
2595 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2596 // The table is later used in get_node_by_cpu().
2597 void os::Linux::rebuild_cpu_to_node_map() {
2598 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2599 // in libnuma (possible values are starting from 16,
2600 // and continuing up with every other power of 2, but less
2601 // than the maximum number of CPUs supported by kernel), and
2602 // is a subject to change (in libnuma version 2 the requirements
2603 // are more reasonable) we'll just hardcode the number they use
2604 // in the library.
2605 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2606
2607 size_t cpu_num = os::active_processor_count();
2608 size_t cpu_map_size = NCPUS / BitsPerCLong;
2609 size_t cpu_map_valid_size =
2610 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2611
2612 cpu_to_node()->clear();
2613 cpu_to_node()->at_grow(cpu_num - 1);
2614 size_t node_num = numa_get_groups_num();
2615
2616 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size);
2617 for (size_t i = 0; i < node_num; i++) {
2618 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2619 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2620 if (cpu_map[j] != 0) {
2621 for (size_t k = 0; k < BitsPerCLong; k++) {
2622 if (cpu_map[j] & (1UL << k)) {
2623 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2624 }
2625 }
2626 }
2627 }
2628 }
2629 }
2630 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2631 }
2632
2633 int os::Linux::get_node_by_cpu(int cpu_id) {
2634 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2635 return cpu_to_node()->at(cpu_id);
2636 }
2637 return -1;
2638 }
2639
2640 GrowableArray<int>* os::Linux::_cpu_to_node;
2641 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2642 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2643 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2644 os::Linux::numa_available_func_t os::Linux::_numa_available;
2645 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2646 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2647 unsigned long* os::Linux::_numa_all_nodes;
2648
2649 bool os::uncommit_memory(char* addr, size_t size) {
2650 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2651 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2652 return res != (uintptr_t) MAP_FAILED;
2653 }
2654
2655 // Linux uses a growable mapping for the stack, and if the mapping for
2656 // the stack guard pages is not removed when we detach a thread the
2657 // stack cannot grow beyond the pages where the stack guard was
2658 // mapped. If at some point later in the process the stack expands to
2659 // that point, the Linux kernel cannot expand the stack any further
2660 // because the guard pages are in the way, and a segfault occurs.
2661 //
2662 // However, it's essential not to split the stack region by unmapping
2663 // a region (leaving a hole) that's already part of the stack mapping,
2664 // so if the stack mapping has already grown beyond the guard pages at
2665 // the time we create them, we have to truncate the stack mapping.
2666 // So, we need to know the extent of the stack mapping when
2667 // create_stack_guard_pages() is called.
2668
2669 // Find the bounds of the stack mapping. Return true for success.
2670 //
2671 // We only need this for stacks that are growable: at the time of
2672 // writing thread stacks don't use growable mappings (i.e. those
2673 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2674 // only applies to the main thread.
2675
2676 static
2677 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2678
2679 char buf[128];
2680 int fd, sz;
2681
2682 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2683 return false;
2684 }
2685
2686 const char kw[] = "[stack]";
2687 const int kwlen = sizeof(kw)-1;
2688
2689 // Address part of /proc/self/maps couldn't be more than 128 bytes
2690 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2691 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2692 // Extract addresses
2693 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2694 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2695 if (sp >= *bottom && sp <= *top) {
2696 ::close(fd);
2697 return true;
2698 }
2699 }
2700 }
2701 }
2702
2703 ::close(fd);
2704 return false;
2705 }
2706
2707
2708 // If the (growable) stack mapping already extends beyond the point
2709 // where we're going to put our guard pages, truncate the mapping at
2710 // that point by munmap()ping it. This ensures that when we later
2711 // munmap() the guard pages we don't leave a hole in the stack
2712 // mapping. This only affects the main/initial thread, but guard
2713 // against future OS changes
2714 bool os::create_stack_guard_pages(char* addr, size_t size) {
2715 uintptr_t stack_extent, stack_base;
2716 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2717 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2718 assert(os::Linux::is_initial_thread(),
2719 "growable stack in non-initial thread");
2720 if (stack_extent < (uintptr_t)addr)
2721 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2722 }
2723
2724 return os::commit_memory(addr, size);
2725 }
2726
2727 // If this is a growable mapping, remove the guard pages entirely by
2728 // munmap()ping them. If not, just call uncommit_memory(). This only
2729 // affects the main/initial thread, but guard against future OS changes
2730 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2731 uintptr_t stack_extent, stack_base;
2732 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2733 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2734 assert(os::Linux::is_initial_thread(),
2735 "growable stack in non-initial thread");
2736
2737 return ::munmap(addr, size) == 0;
2738 }
2739
2740 return os::uncommit_memory(addr, size);
2741 }
2742
2743 static address _highest_vm_reserved_address = NULL;
2744
2745 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2746 // at 'requested_addr'. If there are existing memory mappings at the same
2747 // location, however, they will be overwritten. If 'fixed' is false,
2748 // 'requested_addr' is only treated as a hint, the return value may or
2749 // may not start from the requested address. Unlike Linux mmap(), this
2750 // function returns NULL to indicate failure.
2751 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2752 char * addr;
2753 int flags;
2754
2755 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2756 if (fixed) {
2757 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2758 flags |= MAP_FIXED;
2759 }
2760
2761 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2762 // to PROT_EXEC if executable when we commit the page.
2763 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2764 flags, -1, 0);
2765
2766 if (addr != MAP_FAILED) {
2767 // anon_mmap() should only get called during VM initialization,
2768 // don't need lock (actually we can skip locking even it can be called
2769 // from multiple threads, because _highest_vm_reserved_address is just a
2770 // hint about the upper limit of non-stack memory regions.)
2771 if ((address)addr + bytes > _highest_vm_reserved_address) {
2772 _highest_vm_reserved_address = (address)addr + bytes;
2773 }
2774 }
2775
2776 return addr == MAP_FAILED ? NULL : addr;
2777 }
2778
2779 // Don't update _highest_vm_reserved_address, because there might be memory
2780 // regions above addr + size. If so, releasing a memory region only creates
2781 // a hole in the address space, it doesn't help prevent heap-stack collision.
2782 //
2783 static int anon_munmap(char * addr, size_t size) {
2784 return ::munmap(addr, size) == 0;
2785 }
2786
2787 char* os::reserve_memory(size_t bytes, char* requested_addr,
2788 size_t alignment_hint) {
2789 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2790 }
2791
2792 bool os::release_memory(char* addr, size_t size) {
2793 return anon_munmap(addr, size);
2794 }
2795
2796 static address highest_vm_reserved_address() {
2797 return _highest_vm_reserved_address;
2798 }
2799
2800 static bool linux_mprotect(char* addr, size_t size, int prot) {
2801 // Linux wants the mprotect address argument to be page aligned.
2802 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2803
2804 // According to SUSv3, mprotect() should only be used with mappings
2805 // established by mmap(), and mmap() always maps whole pages. Unaligned
2806 // 'addr' likely indicates problem in the VM (e.g. trying to change
2807 // protection of malloc'ed or statically allocated memory). Check the
2808 // caller if you hit this assert.
2809 assert(addr == bottom, "sanity check");
2810
2811 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2812 return ::mprotect(bottom, size, prot) == 0;
2813 }
2814
2815 // Set protections specified
2816 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2817 bool is_committed) {
2818 unsigned int p = 0;
2819 switch (prot) {
2820 case MEM_PROT_NONE: p = PROT_NONE; break;
2821 case MEM_PROT_READ: p = PROT_READ; break;
2822 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2823 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2824 default:
2825 ShouldNotReachHere();
2826 }
2827 // is_committed is unused.
2828 return linux_mprotect(addr, bytes, p);
2829 }
2830
2831 bool os::guard_memory(char* addr, size_t size) {
2832 return linux_mprotect(addr, size, PROT_NONE);
2833 }
2834
2835 bool os::unguard_memory(char* addr, size_t size) {
2836 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2837 }
2838
2839 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2840 bool result = false;
2841 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2842 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2843 -1, 0);
2844
2845 if (p != (void *) -1) {
2846 // We don't know if this really is a huge page or not.
2847 FILE *fp = fopen("/proc/self/maps", "r");
2848 if (fp) {
2849 while (!feof(fp)) {
2850 char chars[257];
2851 long x = 0;
2852 if (fgets(chars, sizeof(chars), fp)) {
2853 if (sscanf(chars, "%lx-%*lx", &x) == 1
2854 && x == (long)p) {
2855 if (strstr (chars, "hugepage")) {
2856 result = true;
2857 break;
2858 }
2859 }
2860 }
2861 }
2862 fclose(fp);
2863 }
2864 munmap (p, page_size);
2865 if (result)
2866 return true;
2867 }
2868
2869 if (warn) {
2870 warning("HugeTLBFS is not supported by the operating system.");
2871 }
2872
2873 return result;
2874 }
2875
2876 /*
2877 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2878 *
2879 * From the coredump_filter documentation:
2880 *
2881 * - (bit 0) anonymous private memory
2882 * - (bit 1) anonymous shared memory
2883 * - (bit 2) file-backed private memory
2884 * - (bit 3) file-backed shared memory
2885 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2886 * effective only if the bit 2 is cleared)
2887 * - (bit 5) hugetlb private memory
2888 * - (bit 6) hugetlb shared memory
2889 */
2890 static void set_coredump_filter(void) {
2891 FILE *f;
2892 long cdm;
2893
2894 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2895 return;
2896 }
2897
2898 if (fscanf(f, "%lx", &cdm) != 1) {
2899 fclose(f);
2900 return;
2901 }
2902
2903 rewind(f);
2904
2905 if ((cdm & LARGEPAGES_BIT) == 0) {
2906 cdm |= LARGEPAGES_BIT;
2907 fprintf(f, "%#lx", cdm);
2908 }
2909
2910 fclose(f);
2911 }
2912
2913 // Large page support
2914
2915 static size_t _large_page_size = 0;
2916
2917 void os::large_page_init() {
2918 if (!UseLargePages) {
2919 UseHugeTLBFS = false;
2920 UseSHM = false;
2921 return;
2922 }
2923
2924 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
2925 // If UseLargePages is specified on the command line try both methods,
2926 // if it's default, then try only HugeTLBFS.
2927 if (FLAG_IS_DEFAULT(UseLargePages)) {
2928 UseHugeTLBFS = true;
2929 } else {
2930 UseHugeTLBFS = UseSHM = true;
2931 }
2932 }
2933
2934 if (LargePageSizeInBytes) {
2935 _large_page_size = LargePageSizeInBytes;
2936 } else {
2937 // large_page_size on Linux is used to round up heap size. x86 uses either
2938 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2939 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2940 // page as large as 256M.
2941 //
2942 // Here we try to figure out page size by parsing /proc/meminfo and looking
2943 // for a line with the following format:
2944 // Hugepagesize: 2048 kB
2945 //
2946 // If we can't determine the value (e.g. /proc is not mounted, or the text
2947 // format has been changed), we'll use the largest page size supported by
2948 // the processor.
2949
2950 #ifndef ZERO
2951 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
2952 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
2953 #endif // ZERO
2954
2955 FILE *fp = fopen("/proc/meminfo", "r");
2956 if (fp) {
2957 while (!feof(fp)) {
2958 int x = 0;
2959 char buf[16];
2960 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
2961 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
2962 _large_page_size = x * K;
2963 break;
2964 }
2965 } else {
2966 // skip to next line
2967 for (;;) {
2968 int ch = fgetc(fp);
2969 if (ch == EOF || ch == (int)'\n') break;
2970 }
2971 }
2972 }
2973 fclose(fp);
2974 }
2975 }
2976
2977 // print a warning if any large page related flag is specified on command line
2978 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
2979
2980 const size_t default_page_size = (size_t)Linux::page_size();
2981 if (_large_page_size > default_page_size) {
2982 _page_sizes[0] = _large_page_size;
2983 _page_sizes[1] = default_page_size;
2984 _page_sizes[2] = 0;
2985 }
2986 UseHugeTLBFS = UseHugeTLBFS &&
2987 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
2988
2989 if (UseHugeTLBFS)
2990 UseSHM = false;
2991
2992 UseLargePages = UseHugeTLBFS || UseSHM;
2993
2994 set_coredump_filter();
2995 }
2996
2997 #ifndef SHM_HUGETLB
2998 #define SHM_HUGETLB 04000
2999 #endif
3000
3001 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3002 // "exec" is passed in but not used. Creating the shared image for
3003 // the code cache doesn't have an SHM_X executable permission to check.
3004 assert(UseLargePages && UseSHM, "only for SHM large pages");
3005
3006 key_t key = IPC_PRIVATE;
3007 char *addr;
3008
3009 bool warn_on_failure = UseLargePages &&
3010 (!FLAG_IS_DEFAULT(UseLargePages) ||
3011 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3012 );
3013 char msg[128];
3014
3015 // Create a large shared memory region to attach to based on size.
3016 // Currently, size is the total size of the heap
3017 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3018 if (shmid == -1) {
3019 // Possible reasons for shmget failure:
3020 // 1. shmmax is too small for Java heap.
3021 // > check shmmax value: cat /proc/sys/kernel/shmmax
3022 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3023 // 2. not enough large page memory.
3024 // > check available large pages: cat /proc/meminfo
3025 // > increase amount of large pages:
3026 // echo new_value > /proc/sys/vm/nr_hugepages
3027 // Note 1: different Linux may use different name for this property,
3028 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3029 // Note 2: it's possible there's enough physical memory available but
3030 // they are so fragmented after a long run that they can't
3031 // coalesce into large pages. Try to reserve large pages when
3032 // the system is still "fresh".
3033 if (warn_on_failure) {
3034 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3035 warning(msg);
3036 }
3037 return NULL;
3038 }
3039
3040 // attach to the region
3041 addr = (char*)shmat(shmid, req_addr, 0);
3042 int err = errno;
3043
3044 // Remove shmid. If shmat() is successful, the actual shared memory segment
3045 // will be deleted when it's detached by shmdt() or when the process
3046 // terminates. If shmat() is not successful this will remove the shared
3047 // segment immediately.
3048 shmctl(shmid, IPC_RMID, NULL);
3049
3050 if ((intptr_t)addr == -1) {
3051 if (warn_on_failure) {
3052 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3053 warning(msg);
3054 }
3055 return NULL;
3056 }
3057
3058 return addr;
3059 }
3060
3061 bool os::release_memory_special(char* base, size_t bytes) {
3062 // detaching the SHM segment will also delete it, see reserve_memory_special()
3063 int rslt = shmdt(base);
3064 return rslt == 0;
3065 }
3066
3067 size_t os::large_page_size() {
3068 return _large_page_size;
3069 }
3070
3071 // HugeTLBFS allows application to commit large page memory on demand;
3072 // with SysV SHM the entire memory region must be allocated as shared
3073 // memory.
3074 bool os::can_commit_large_page_memory() {
3075 return UseHugeTLBFS;
3076 }
3077
3078 bool os::can_execute_large_page_memory() {
3079 return UseHugeTLBFS;
3080 }
3081
3082 // Reserve memory at an arbitrary address, only if that area is
3083 // available (and not reserved for something else).
3084
3085 char* os::attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3086 const int max_tries = 10;
3087 char* base[max_tries];
3088 size_t size[max_tries];
3089 const size_t gap = 0x000000;
3090
3091 // Assert only that the size is a multiple of the page size, since
3092 // that's all that mmap requires, and since that's all we really know
3093 // about at this low abstraction level. If we need higher alignment,
3094 // we can either pass an alignment to this method or verify alignment
3095 // in one of the methods further up the call chain. See bug 5044738.
3096 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3097
3098 // Repeatedly allocate blocks until the block is allocated at the
3099 // right spot. Give up after max_tries. Note that reserve_memory() will
3100 // automatically update _highest_vm_reserved_address if the call is
3101 // successful. The variable tracks the highest memory address every reserved
3102 // by JVM. It is used to detect heap-stack collision if running with
3103 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3104 // space than needed, it could confuse the collision detecting code. To
3105 // solve the problem, save current _highest_vm_reserved_address and
3106 // calculate the correct value before return.
3107 address old_highest = _highest_vm_reserved_address;
3108
3109 // Linux mmap allows caller to pass an address as hint; give it a try first,
3110 // if kernel honors the hint then we can return immediately.
3111 char * addr = anon_mmap(requested_addr, bytes, false);
3112 if (addr == requested_addr) {
3113 return requested_addr;
3114 }
3115
3116 if (addr != NULL) {
3117 // mmap() is successful but it fails to reserve at the requested address
3118 anon_munmap(addr, bytes);
3119 }
3120
3121 int i;
3122 for (i = 0; i < max_tries; ++i) {
3123 base[i] = reserve_memory(bytes);
3124
3125 if (base[i] != NULL) {
3126 // Is this the block we wanted?
3127 if (base[i] == requested_addr) {
3128 size[i] = bytes;
3129 break;
3130 }
3131
3132 // Does this overlap the block we wanted? Give back the overlapped
3133 // parts and try again.
3134
3135 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3136 if (top_overlap >= 0 && top_overlap < bytes) {
3137 unmap_memory(base[i], top_overlap);
3138 base[i] += top_overlap;
3139 size[i] = bytes - top_overlap;
3140 } else {
3141 size_t bottom_overlap = base[i] + bytes - requested_addr;
3142 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3143 unmap_memory(requested_addr, bottom_overlap);
3144 size[i] = bytes - bottom_overlap;
3145 } else {
3146 size[i] = bytes;
3147 }
3148 }
3149 }
3150 }
3151
3152 // Give back the unused reserved pieces.
3153
3154 for (int j = 0; j < i; ++j) {
3155 if (base[j] != NULL) {
3156 unmap_memory(base[j], size[j]);
3157 }
3158 }
3159
3160 if (i < max_tries) {
3161 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3162 return requested_addr;
3163 } else {
3164 _highest_vm_reserved_address = old_highest;
3165 return NULL;
3166 }
3167 }
3168
3169 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3170 return ::read(fd, buf, nBytes);
3171 }
3172
3173 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3174 // Solaris uses poll(), linux uses park().
3175 // Poll() is likely a better choice, assuming that Thread.interrupt()
3176 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3177 // SIGSEGV, see 4355769.
3178
3179 const int NANOSECS_PER_MILLISECS = 1000000;
3180
3181 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3182 assert(thread == Thread::current(), "thread consistency check");
3183
3184 ParkEvent * const slp = thread->_SleepEvent ;
3185 slp->reset() ;
3186 OrderAccess::fence() ;
3187
3188 if (interruptible) {
3189 jlong prevtime = javaTimeNanos();
3190
3191 for (;;) {
3192 if (os::is_interrupted(thread, true)) {
3193 return OS_INTRPT;
3194 }
3195
3196 jlong newtime = javaTimeNanos();
3197
3198 if (newtime - prevtime < 0) {
3199 // time moving backwards, should only happen if no monotonic clock
3200 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3201 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3202 } else {
3203 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3204 }
3205
3206 if(millis <= 0) {
3207 return OS_OK;
3208 }
3209
3210 prevtime = newtime;
3211
3212 {
3213 assert(thread->is_Java_thread(), "sanity check");
3214 JavaThread *jt = (JavaThread *) thread;
3215 ThreadBlockInVM tbivm(jt);
3216 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3217
3218 jt->set_suspend_equivalent();
3219 // cleared by handle_special_suspend_equivalent_condition() or
3220 // java_suspend_self() via check_and_wait_while_suspended()
3221
3222 slp->park(millis);
3223
3224 // were we externally suspended while we were waiting?
3225 jt->check_and_wait_while_suspended();
3226 }
3227 }
3228 } else {
3229 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3230 jlong prevtime = javaTimeNanos();
3231
3232 for (;;) {
3233 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3234 // the 1st iteration ...
3235 jlong newtime = javaTimeNanos();
3236
3237 if (newtime - prevtime < 0) {
3238 // time moving backwards, should only happen if no monotonic clock
3239 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3240 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3241 } else {
3242 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISECS;
3243 }
3244
3245 if(millis <= 0) break ;
3246
3247 prevtime = newtime;
3248 slp->park(millis);
3249 }
3250 return OS_OK ;
3251 }
3252 }
3253
3254 int os::naked_sleep() {
3255 // %% make the sleep time an integer flag. for now use 1 millisec.
3256 return os::sleep(Thread::current(), 1, false);
3257 }
3258
3259 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3260 void os::infinite_sleep() {
3261 while (true) { // sleep forever ...
3262 ::sleep(100); // ... 100 seconds at a time
3263 }
3264 }
3265
3266 // Used to convert frequent JVM_Yield() to nops
3267 bool os::dont_yield() {
3268 return DontYieldALot;
3269 }
3270
3271 void os::yield() {
3272 sched_yield();
3273 }
3274
3275 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3276
3277 void os::yield_all(int attempts) {
3278 // Yields to all threads, including threads with lower priorities
3279 // Threads on Linux are all with same priority. The Solaris style
3280 // os::yield_all() with nanosleep(1ms) is not necessary.
3281 sched_yield();
3282 }
3283
3284 // Called from the tight loops to possibly influence time-sharing heuristics
3285 void os::loop_breaker(int attempts) {
3286 os::yield_all(attempts);
3287 }
3288
3289 ////////////////////////////////////////////////////////////////////////////////
3290 // thread priority support
3291
3292 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3293 // only supports dynamic priority, static priority must be zero. For real-time
3294 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3295 // However, for large multi-threaded applications, SCHED_RR is not only slower
3296 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3297 // of 5 runs - Sep 2005).
3298 //
3299 // The following code actually changes the niceness of kernel-thread/LWP. It
3300 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3301 // not the entire user process, and user level threads are 1:1 mapped to kernel
3302 // threads. It has always been the case, but could change in the future. For
3303 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3304 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3305
3306 int os::java_to_os_priority[MaxPriority + 1] = {
3307 19, // 0 Entry should never be used
3308
3309 4, // 1 MinPriority
3310 3, // 2
3311 2, // 3
3312
3313 1, // 4
3314 0, // 5 NormPriority
3315 -1, // 6
3316
3317 -2, // 7
3318 -3, // 8
3319 -4, // 9 NearMaxPriority
3320
3321 -5 // 10 MaxPriority
3322 };
3323
3324 static int prio_init() {
3325 if (ThreadPriorityPolicy == 1) {
3326 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3327 // if effective uid is not root. Perhaps, a more elegant way of doing
3328 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3329 if (geteuid() != 0) {
3330 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3331 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3332 }
3333 ThreadPriorityPolicy = 0;
3334 }
3335 }
3336 return 0;
3337 }
3338
3339 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3340 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3341
3342 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3343 return (ret == 0) ? OS_OK : OS_ERR;
3344 }
3345
3346 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3347 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3348 *priority_ptr = java_to_os_priority[NormPriority];
3349 return OS_OK;
3350 }
3351
3352 errno = 0;
3353 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3354 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3355 }
3356
3357 // Hint to the underlying OS that a task switch would not be good.
3358 // Void return because it's a hint and can fail.
3359 void os::hint_no_preempt() {}
3360
3361 ////////////////////////////////////////////////////////////////////////////////
3362 // suspend/resume support
3363
3364 // the low-level signal-based suspend/resume support is a remnant from the
3365 // old VM-suspension that used to be for java-suspension, safepoints etc,
3366 // within hotspot. Now there is a single use-case for this:
3367 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3368 // that runs in the watcher thread.
3369 // The remaining code is greatly simplified from the more general suspension
3370 // code that used to be used.
3371 //
3372 // The protocol is quite simple:
3373 // - suspend:
3374 // - sends a signal to the target thread
3375 // - polls the suspend state of the osthread using a yield loop
3376 // - target thread signal handler (SR_handler) sets suspend state
3377 // and blocks in sigsuspend until continued
3378 // - resume:
3379 // - sets target osthread state to continue
3380 // - sends signal to end the sigsuspend loop in the SR_handler
3381 //
3382 // Note that the SR_lock plays no role in this suspend/resume protocol.
3383 //
3384
3385 static void resume_clear_context(OSThread *osthread) {
3386 osthread->set_ucontext(NULL);
3387 osthread->set_siginfo(NULL);
3388
3389 // notify the suspend action is completed, we have now resumed
3390 osthread->sr.clear_suspended();
3391 }
3392
3393 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3394 osthread->set_ucontext(context);
3395 osthread->set_siginfo(siginfo);
3396 }
3397
3398 //
3399 // Handler function invoked when a thread's execution is suspended or
3400 // resumed. We have to be careful that only async-safe functions are
3401 // called here (Note: most pthread functions are not async safe and
3402 // should be avoided.)
3403 //
3404 // Note: sigwait() is a more natural fit than sigsuspend() from an
3405 // interface point of view, but sigwait() prevents the signal hander
3406 // from being run. libpthread would get very confused by not having
3407 // its signal handlers run and prevents sigwait()'s use with the
3408 // mutex granting granting signal.
3409 //
3410 // Currently only ever called on the VMThread
3411 //
3412 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3413 // Save and restore errno to avoid confusing native code with EINTR
3414 // after sigsuspend.
3415 int old_errno = errno;
3416
3417 Thread* thread = Thread::current();
3418 OSThread* osthread = thread->osthread();
3419 assert(thread->is_VM_thread(), "Must be VMThread");
3420 // read current suspend action
3421 int action = osthread->sr.suspend_action();
3422 if (action == SR_SUSPEND) {
3423 suspend_save_context(osthread, siginfo, context);
3424
3425 // Notify the suspend action is about to be completed. do_suspend()
3426 // waits until SR_SUSPENDED is set and then returns. We will wait
3427 // here for a resume signal and that completes the suspend-other
3428 // action. do_suspend/do_resume is always called as a pair from
3429 // the same thread - so there are no races
3430
3431 // notify the caller
3432 osthread->sr.set_suspended();
3433
3434 sigset_t suspend_set; // signals for sigsuspend()
3435
3436 // get current set of blocked signals and unblock resume signal
3437 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3438 sigdelset(&suspend_set, SR_signum);
3439
3440 // wait here until we are resumed
3441 do {
3442 sigsuspend(&suspend_set);
3443 // ignore all returns until we get a resume signal
3444 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3445
3446 resume_clear_context(osthread);
3447
3448 } else {
3449 assert(action == SR_CONTINUE, "unexpected sr action");
3450 // nothing special to do - just leave the handler
3451 }
3452
3453 errno = old_errno;
3454 }
3455
3456
3457 static int SR_initialize() {
3458 struct sigaction act;
3459 char *s;
3460 /* Get signal number to use for suspend/resume */
3461 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3462 int sig = ::strtol(s, 0, 10);
3463 if (sig > 0 || sig < _NSIG) {
3464 SR_signum = sig;
3465 }
3466 }
3467
3468 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3469 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3470
3471 sigemptyset(&SR_sigset);
3472 sigaddset(&SR_sigset, SR_signum);
3473
3474 /* Set up signal handler for suspend/resume */
3475 act.sa_flags = SA_RESTART|SA_SIGINFO;
3476 act.sa_handler = (void (*)(int)) SR_handler;
3477
3478 // SR_signum is blocked by default.
3479 // 4528190 - We also need to block pthread restart signal (32 on all
3480 // supported Linux platforms). Note that LinuxThreads need to block
3481 // this signal for all threads to work properly. So we don't have
3482 // to use hard-coded signal number when setting up the mask.
3483 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3484
3485 if (sigaction(SR_signum, &act, 0) == -1) {
3486 return -1;
3487 }
3488
3489 // Save signal flag
3490 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3491 return 0;
3492 }
3493
3494 static int SR_finalize() {
3495 return 0;
3496 }
3497
3498
3499 // returns true on success and false on error - really an error is fatal
3500 // but this seems the normal response to library errors
3501 static bool do_suspend(OSThread* osthread) {
3502 // mark as suspended and send signal
3503 osthread->sr.set_suspend_action(SR_SUSPEND);
3504 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3505 assert_status(status == 0, status, "pthread_kill");
3506
3507 // check status and wait until notified of suspension
3508 if (status == 0) {
3509 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3510 os::yield_all(i);
3511 }
3512 osthread->sr.set_suspend_action(SR_NONE);
3513 return true;
3514 }
3515 else {
3516 osthread->sr.set_suspend_action(SR_NONE);
3517 return false;
3518 }
3519 }
3520
3521 static void do_resume(OSThread* osthread) {
3522 assert(osthread->sr.is_suspended(), "thread should be suspended");
3523 osthread->sr.set_suspend_action(SR_CONTINUE);
3524
3525 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3526 assert_status(status == 0, status, "pthread_kill");
3527 // check status and wait unit notified of resumption
3528 if (status == 0) {
3529 for (int i = 0; osthread->sr.is_suspended(); i++) {
3530 os::yield_all(i);
3531 }
3532 }
3533 osthread->sr.set_suspend_action(SR_NONE);
3534 }
3535
3536 ////////////////////////////////////////////////////////////////////////////////
3537 // interrupt support
3538
3539 void os::interrupt(Thread* thread) {
3540 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3541 "possibility of dangling Thread pointer");
3542
3543 OSThread* osthread = thread->osthread();
3544
3545 if (!osthread->interrupted()) {
3546 osthread->set_interrupted(true);
3547 // More than one thread can get here with the same value of osthread,
3548 // resulting in multiple notifications. We do, however, want the store
3549 // to interrupted() to be visible to other threads before we execute unpark().
3550 OrderAccess::fence();
3551 ParkEvent * const slp = thread->_SleepEvent ;
3552 if (slp != NULL) slp->unpark() ;
3553 }
3554
3555 // For JSR166. Unpark even if interrupt status already was set
3556 if (thread->is_Java_thread())
3557 ((JavaThread*)thread)->parker()->unpark();
3558
3559 ParkEvent * ev = thread->_ParkEvent ;
3560 if (ev != NULL) ev->unpark() ;
3561
3562 }
3563
3564 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3565 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3566 "possibility of dangling Thread pointer");
3567
3568 OSThread* osthread = thread->osthread();
3569
3570 bool interrupted = osthread->interrupted();
3571
3572 if (interrupted && clear_interrupted) {
3573 osthread->set_interrupted(false);
3574 // consider thread->_SleepEvent->reset() ... optional optimization
3575 }
3576
3577 return interrupted;
3578 }
3579
3580 ///////////////////////////////////////////////////////////////////////////////////
3581 // signal handling (except suspend/resume)
3582
3583 // This routine may be used by user applications as a "hook" to catch signals.
3584 // The user-defined signal handler must pass unrecognized signals to this
3585 // routine, and if it returns true (non-zero), then the signal handler must
3586 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3587 // routine will never retun false (zero), but instead will execute a VM panic
3588 // routine kill the process.
3589 //
3590 // If this routine returns false, it is OK to call it again. This allows
3591 // the user-defined signal handler to perform checks either before or after
3592 // the VM performs its own checks. Naturally, the user code would be making
3593 // a serious error if it tried to handle an exception (such as a null check
3594 // or breakpoint) that the VM was generating for its own correct operation.
3595 //
3596 // This routine may recognize any of the following kinds of signals:
3597 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3598 // It should be consulted by handlers for any of those signals.
3599 //
3600 // The caller of this routine must pass in the three arguments supplied
3601 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3602 // field of the structure passed to sigaction(). This routine assumes that
3603 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3604 //
3605 // Note that the VM will print warnings if it detects conflicting signal
3606 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3607 //
3608 extern "C" JNIEXPORT int
3609 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3610 void* ucontext, int abort_if_unrecognized);
3611
3612 void signalHandler(int sig, siginfo_t* info, void* uc) {
3613 assert(info != NULL && uc != NULL, "it must be old kernel");
3614 JVM_handle_linux_signal(sig, info, uc, true);
3615 }
3616
3617
3618 // This boolean allows users to forward their own non-matching signals
3619 // to JVM_handle_linux_signal, harmlessly.
3620 bool os::Linux::signal_handlers_are_installed = false;
3621
3622 // For signal-chaining
3623 struct sigaction os::Linux::sigact[MAXSIGNUM];
3624 unsigned int os::Linux::sigs = 0;
3625 bool os::Linux::libjsig_is_loaded = false;
3626 typedef struct sigaction *(*get_signal_t)(int);
3627 get_signal_t os::Linux::get_signal_action = NULL;
3628
3629 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3630 struct sigaction *actp = NULL;
3631
3632 if (libjsig_is_loaded) {
3633 // Retrieve the old signal handler from libjsig
3634 actp = (*get_signal_action)(sig);
3635 }
3636 if (actp == NULL) {
3637 // Retrieve the preinstalled signal handler from jvm
3638 actp = get_preinstalled_handler(sig);
3639 }
3640
3641 return actp;
3642 }
3643
3644 static bool call_chained_handler(struct sigaction *actp, int sig,
3645 siginfo_t *siginfo, void *context) {
3646 // Call the old signal handler
3647 if (actp->sa_handler == SIG_DFL) {
3648 // It's more reasonable to let jvm treat it as an unexpected exception
3649 // instead of taking the default action.
3650 return false;
3651 } else if (actp->sa_handler != SIG_IGN) {
3652 if ((actp->sa_flags & SA_NODEFER) == 0) {
3653 // automaticlly block the signal
3654 sigaddset(&(actp->sa_mask), sig);
3655 }
3656
3657 sa_handler_t hand;
3658 sa_sigaction_t sa;
3659 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3660 // retrieve the chained handler
3661 if (siginfo_flag_set) {
3662 sa = actp->sa_sigaction;
3663 } else {
3664 hand = actp->sa_handler;
3665 }
3666
3667 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3668 actp->sa_handler = SIG_DFL;
3669 }
3670
3671 // try to honor the signal mask
3672 sigset_t oset;
3673 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3674
3675 // call into the chained handler
3676 if (siginfo_flag_set) {
3677 (*sa)(sig, siginfo, context);
3678 } else {
3679 (*hand)(sig);
3680 }
3681
3682 // restore the signal mask
3683 pthread_sigmask(SIG_SETMASK, &oset, 0);
3684 }
3685 // Tell jvm's signal handler the signal is taken care of.
3686 return true;
3687 }
3688
3689 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3690 bool chained = false;
3691 // signal-chaining
3692 if (UseSignalChaining) {
3693 struct sigaction *actp = get_chained_signal_action(sig);
3694 if (actp != NULL) {
3695 chained = call_chained_handler(actp, sig, siginfo, context);
3696 }
3697 }
3698 return chained;
3699 }
3700
3701 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3702 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3703 return &sigact[sig];
3704 }
3705 return NULL;
3706 }
3707
3708 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3709 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3710 sigact[sig] = oldAct;
3711 sigs |= (unsigned int)1 << sig;
3712 }
3713
3714 // for diagnostic
3715 int os::Linux::sigflags[MAXSIGNUM];
3716
3717 int os::Linux::get_our_sigflags(int sig) {
3718 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3719 return sigflags[sig];
3720 }
3721
3722 void os::Linux::set_our_sigflags(int sig, int flags) {
3723 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3724 sigflags[sig] = flags;
3725 }
3726
3727 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3728 // Check for overwrite.
3729 struct sigaction oldAct;
3730 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3731
3732 void* oldhand = oldAct.sa_sigaction
3733 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3734 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3735 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3736 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3737 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3738 if (AllowUserSignalHandlers || !set_installed) {
3739 // Do not overwrite; user takes responsibility to forward to us.
3740 return;
3741 } else if (UseSignalChaining) {
3742 // save the old handler in jvm
3743 save_preinstalled_handler(sig, oldAct);
3744 // libjsig also interposes the sigaction() call below and saves the
3745 // old sigaction on it own.
3746 } else {
3747 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3748 "%#lx for signal %d.", (long)oldhand, sig));
3749 }
3750 }
3751
3752 struct sigaction sigAct;
3753 sigfillset(&(sigAct.sa_mask));
3754 sigAct.sa_handler = SIG_DFL;
3755 if (!set_installed) {
3756 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3757 } else {
3758 sigAct.sa_sigaction = signalHandler;
3759 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3760 }
3761 // Save flags, which are set by ours
3762 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3763 sigflags[sig] = sigAct.sa_flags;
3764
3765 int ret = sigaction(sig, &sigAct, &oldAct);
3766 assert(ret == 0, "check");
3767
3768 void* oldhand2 = oldAct.sa_sigaction
3769 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3770 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3771 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3772 }
3773
3774 // install signal handlers for signals that HotSpot needs to
3775 // handle in order to support Java-level exception handling.
3776
3777 void os::Linux::install_signal_handlers() {
3778 if (!signal_handlers_are_installed) {
3779 signal_handlers_are_installed = true;
3780
3781 // signal-chaining
3782 typedef void (*signal_setting_t)();
3783 signal_setting_t begin_signal_setting = NULL;
3784 signal_setting_t end_signal_setting = NULL;
3785 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3786 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3787 if (begin_signal_setting != NULL) {
3788 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3789 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3790 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3791 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3792 libjsig_is_loaded = true;
3793 assert(UseSignalChaining, "should enable signal-chaining");
3794 }
3795 if (libjsig_is_loaded) {
3796 // Tell libjsig jvm is setting signal handlers
3797 (*begin_signal_setting)();
3798 }
3799
3800 set_signal_handler(SIGSEGV, true);
3801 set_signal_handler(SIGPIPE, true);
3802 set_signal_handler(SIGBUS, true);
3803 set_signal_handler(SIGILL, true);
3804 set_signal_handler(SIGFPE, true);
3805 set_signal_handler(SIGXFSZ, true);
3806
3807 if (libjsig_is_loaded) {
3808 // Tell libjsig jvm finishes setting signal handlers
3809 (*end_signal_setting)();
3810 }
3811
3812 // We don't activate signal checker if libjsig is in place, we trust ourselves
3813 // and if UserSignalHandler is installed all bets are off
3814 if (CheckJNICalls) {
3815 if (libjsig_is_loaded) {
3816 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3817 check_signals = false;
3818 }
3819 if (AllowUserSignalHandlers) {
3820 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3821 check_signals = false;
3822 }
3823 }
3824 }
3825 }
3826
3827 // This is the fastest way to get thread cpu time on Linux.
3828 // Returns cpu time (user+sys) for any thread, not only for current.
3829 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3830 // It might work on 2.6.10+ with a special kernel/glibc patch.
3831 // For reference, please, see IEEE Std 1003.1-2004:
3832 // http://www.unix.org/single_unix_specification
3833
3834 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3835 struct timespec tp;
3836 int rc = os::Linux::clock_gettime(clockid, &tp);
3837 assert(rc == 0, "clock_gettime is expected to return 0 code");
3838
3839 return (tp.tv_sec * SEC_IN_NANOSECS) + tp.tv_nsec;
3840 }
3841
3842 /////
3843 // glibc on Linux platform uses non-documented flag
3844 // to indicate, that some special sort of signal
3845 // trampoline is used.
3846 // We will never set this flag, and we should
3847 // ignore this flag in our diagnostic
3848 #ifdef SIGNIFICANT_SIGNAL_MASK
3849 #undef SIGNIFICANT_SIGNAL_MASK
3850 #endif
3851 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3852
3853 static const char* get_signal_handler_name(address handler,
3854 char* buf, int buflen) {
3855 int offset;
3856 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3857 if (found) {
3858 // skip directory names
3859 const char *p1, *p2;
3860 p1 = buf;
3861 size_t len = strlen(os::file_separator());
3862 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3863 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3864 } else {
3865 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3866 }
3867 return buf;
3868 }
3869
3870 static void print_signal_handler(outputStream* st, int sig,
3871 char* buf, size_t buflen) {
3872 struct sigaction sa;
3873
3874 sigaction(sig, NULL, &sa);
3875
3876 // See comment for SIGNIFICANT_SIGNAL_MASK define
3877 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3878
3879 st->print("%s: ", os::exception_name(sig, buf, buflen));
3880
3881 address handler = (sa.sa_flags & SA_SIGINFO)
3882 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3883 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3884
3885 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3886 st->print("SIG_DFL");
3887 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3888 st->print("SIG_IGN");
3889 } else {
3890 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3891 }
3892
3893 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3894
3895 address rh = VMError::get_resetted_sighandler(sig);
3896 // May be, handler was resetted by VMError?
3897 if(rh != NULL) {
3898 handler = rh;
3899 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3900 }
3901
3902 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3903
3904 // Check: is it our handler?
3905 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3906 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3907 // It is our signal handler
3908 // check for flags, reset system-used one!
3909 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3910 st->print(
3911 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3912 os::Linux::get_our_sigflags(sig));
3913 }
3914 }
3915 st->cr();
3916 }
3917
3918
3919 #define DO_SIGNAL_CHECK(sig) \
3920 if (!sigismember(&check_signal_done, sig)) \
3921 os::Linux::check_signal_handler(sig)
3922
3923 // This method is a periodic task to check for misbehaving JNI applications
3924 // under CheckJNI, we can add any periodic checks here
3925
3926 void os::run_periodic_checks() {
3927
3928 if (check_signals == false) return;
3929
3930 // SEGV and BUS if overridden could potentially prevent
3931 // generation of hs*.log in the event of a crash, debugging
3932 // such a case can be very challenging, so we absolutely
3933 // check the following for a good measure:
3934 DO_SIGNAL_CHECK(SIGSEGV);
3935 DO_SIGNAL_CHECK(SIGILL);
3936 DO_SIGNAL_CHECK(SIGFPE);
3937 DO_SIGNAL_CHECK(SIGBUS);
3938 DO_SIGNAL_CHECK(SIGPIPE);
3939 DO_SIGNAL_CHECK(SIGXFSZ);
3940
3941
3942 // ReduceSignalUsage allows the user to override these handlers
3943 // see comments at the very top and jvm_solaris.h
3944 if (!ReduceSignalUsage) {
3945 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
3946 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
3947 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
3948 DO_SIGNAL_CHECK(BREAK_SIGNAL);
3949 }
3950
3951 DO_SIGNAL_CHECK(SR_signum);
3952 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
3953 }
3954
3955 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
3956
3957 static os_sigaction_t os_sigaction = NULL;
3958
3959 void os::Linux::check_signal_handler(int sig) {
3960 char buf[O_BUFLEN];
3961 address jvmHandler = NULL;
3962
3963
3964 struct sigaction act;
3965 if (os_sigaction == NULL) {
3966 // only trust the default sigaction, in case it has been interposed
3967 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
3968 if (os_sigaction == NULL) return;
3969 }
3970
3971 os_sigaction(sig, (struct sigaction*)NULL, &act);
3972
3973
3974 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3975
3976 address thisHandler = (act.sa_flags & SA_SIGINFO)
3977 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
3978 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
3979
3980
3981 switch(sig) {
3982 case SIGSEGV:
3983 case SIGBUS:
3984 case SIGFPE:
3985 case SIGPIPE:
3986 case SIGILL:
3987 case SIGXFSZ:
3988 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
3989 break;
3990
3991 case SHUTDOWN1_SIGNAL:
3992 case SHUTDOWN2_SIGNAL:
3993 case SHUTDOWN3_SIGNAL:
3994 case BREAK_SIGNAL:
3995 jvmHandler = (address)user_handler();
3996 break;
3997
3998 case INTERRUPT_SIGNAL:
3999 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4000 break;
4001
4002 default:
4003 if (sig == SR_signum) {
4004 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4005 } else {
4006 return;
4007 }
4008 break;
4009 }
4010
4011 if (thisHandler != jvmHandler) {
4012 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4013 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4014 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4015 // No need to check this sig any longer
4016 sigaddset(&check_signal_done, sig);
4017 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4018 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4019 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4020 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4021 // No need to check this sig any longer
4022 sigaddset(&check_signal_done, sig);
4023 }
4024
4025 // Dump all the signal
4026 if (sigismember(&check_signal_done, sig)) {
4027 print_signal_handlers(tty, buf, O_BUFLEN);
4028 }
4029 }
4030
4031 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4032
4033 extern bool signal_name(int signo, char* buf, size_t len);
4034
4035 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4036 if (0 < exception_code && exception_code <= SIGRTMAX) {
4037 // signal
4038 if (!signal_name(exception_code, buf, size)) {
4039 jio_snprintf(buf, size, "SIG%d", exception_code);
4040 }
4041 return buf;
4042 } else {
4043 return NULL;
4044 }
4045 }
4046
4047 // this is called _before_ the most of global arguments have been parsed
4048 void os::init(void) {
4049 char dummy; /* used to get a guess on initial stack address */
4050 // first_hrtime = gethrtime();
4051
4052 // With LinuxThreads the JavaMain thread pid (primordial thread)
4053 // is different than the pid of the java launcher thread.
4054 // So, on Linux, the launcher thread pid is passed to the VM
4055 // via the sun.java.launcher.pid property.
4056 // Use this property instead of getpid() if it was correctly passed.
4057 // See bug 6351349.
4058 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4059
4060 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4061
4062 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4063
4064 init_random(1234567);
4065
4066 ThreadCritical::initialize();
4067
4068 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4069 if (Linux::page_size() == -1) {
4070 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4071 strerror(errno)));
4072 }
4073 init_page_sizes((size_t) Linux::page_size());
4074
4075 Linux::initialize_system_info();
4076
4077 // main_thread points to the aboriginal thread
4078 Linux::_main_thread = pthread_self();
4079
4080 Linux::clock_init();
4081 initial_time_count = os::elapsed_counter();
4082 pthread_mutex_init(&dl_mutex, NULL);
4083 }
4084
4085 // To install functions for atexit system call
4086 extern "C" {
4087 static void perfMemory_exit_helper() {
4088 perfMemory_exit();
4089 }
4090 }
4091
4092 // this is called _after_ the global arguments have been parsed
4093 jint os::init_2(void)
4094 {
4095 Linux::fast_thread_clock_init();
4096
4097 // Allocate a single page and mark it as readable for safepoint polling
4098 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4099 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4100
4101 os::set_polling_page( polling_page );
4102
4103 #ifndef PRODUCT
4104 if(Verbose && PrintMiscellaneous)
4105 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4106 #endif
4107
4108 if (!UseMembar) {
4109 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4110 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4111 os::set_memory_serialize_page( mem_serialize_page );
4112
4113 #ifndef PRODUCT
4114 if(Verbose && PrintMiscellaneous)
4115 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4116 #endif
4117 }
4118
4119 os::large_page_init();
4120
4121 // initialize suspend/resume support - must do this before signal_sets_init()
4122 if (SR_initialize() != 0) {
4123 perror("SR_initialize failed");
4124 return JNI_ERR;
4125 }
4126
4127 Linux::signal_sets_init();
4128 Linux::install_signal_handlers();
4129
4130 // Check minimum allowable stack size for thread creation and to initialize
4131 // the java system classes, including StackOverflowError - depends on page
4132 // size. Add a page for compiler2 recursion in main thread.
4133 // Add in 2*BytesPerWord times page size to account for VM stack during
4134 // class initialization depending on 32 or 64 bit VM.
4135 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4136 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4137 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4138
4139 size_t threadStackSizeInBytes = ThreadStackSize * K;
4140 if (threadStackSizeInBytes != 0 &&
4141 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4142 tty->print_cr("\nThe stack size specified is too small, "
4143 "Specify at least %dk",
4144 os::Linux::min_stack_allowed/ K);
4145 return JNI_ERR;
4146 }
4147
4148 // Make the stack size a multiple of the page size so that
4149 // the yellow/red zones can be guarded.
4150 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4151 vm_page_size()));
4152
4153 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4154
4155 Linux::libpthread_init();
4156 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4157 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4158 Linux::glibc_version(), Linux::libpthread_version(),
4159 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4160 }
4161
4162 if (UseNUMA) {
4163 if (!Linux::libnuma_init()) {
4164 UseNUMA = false;
4165 } else {
4166 if ((Linux::numa_max_node() < 1)) {
4167 // There's only one node(they start from 0), disable NUMA.
4168 UseNUMA = false;
4169 }
4170 }
4171 // With SHM large pages we cannot uncommit a page, so there's not way
4172 // we can make the adaptive lgrp chunk resizing work. If the user specified
4173 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4174 // disable adaptive resizing.
4175 if (UseNUMA && UseLargePages && UseSHM) {
4176 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4177 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4178 UseLargePages = false;
4179 } else {
4180 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4181 UseAdaptiveSizePolicy = false;
4182 UseAdaptiveNUMAChunkSizing = false;
4183 }
4184 } else {
4185 UseNUMA = false;
4186 }
4187 }
4188 if (!UseNUMA && ForceNUMA) {
4189 UseNUMA = true;
4190 }
4191 }
4192
4193 if (MaxFDLimit) {
4194 // set the number of file descriptors to max. print out error
4195 // if getrlimit/setrlimit fails but continue regardless.
4196 struct rlimit nbr_files;
4197 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4198 if (status != 0) {
4199 if (PrintMiscellaneous && (Verbose || WizardMode))
4200 perror("os::init_2 getrlimit failed");
4201 } else {
4202 nbr_files.rlim_cur = nbr_files.rlim_max;
4203 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4204 if (status != 0) {
4205 if (PrintMiscellaneous && (Verbose || WizardMode))
4206 perror("os::init_2 setrlimit failed");
4207 }
4208 }
4209 }
4210
4211 // Initialize lock used to serialize thread creation (see os::create_thread)
4212 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4213
4214 // at-exit methods are called in the reverse order of their registration.
4215 // atexit functions are called on return from main or as a result of a
4216 // call to exit(3C). There can be only 32 of these functions registered
4217 // and atexit() does not set errno.
4218
4219 if (PerfAllowAtExitRegistration) {
4220 // only register atexit functions if PerfAllowAtExitRegistration is set.
4221 // atexit functions can be delayed until process exit time, which
4222 // can be problematic for embedded VM situations. Embedded VMs should
4223 // call DestroyJavaVM() to assure that VM resources are released.
4224
4225 // note: perfMemory_exit_helper atexit function may be removed in
4226 // the future if the appropriate cleanup code can be added to the
4227 // VM_Exit VMOperation's doit method.
4228 if (atexit(perfMemory_exit_helper) != 0) {
4229 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4230 }
4231 }
4232
4233 // initialize thread priority policy
4234 prio_init();
4235
4236 return JNI_OK;
4237 }
4238
4239 // this is called at the end of vm_initialization
4240 void os::init_3(void) { }
4241
4242 // Mark the polling page as unreadable
4243 void os::make_polling_page_unreadable(void) {
4244 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4245 fatal("Could not disable polling page");
4246 };
4247
4248 // Mark the polling page as readable
4249 void os::make_polling_page_readable(void) {
4250 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4251 fatal("Could not enable polling page");
4252 }
4253 };
4254
4255 int os::active_processor_count() {
4256 // Linux doesn't yet have a (official) notion of processor sets,
4257 // so just return the number of online processors.
4258 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4259 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4260 return online_cpus;
4261 }
4262
4263 bool os::distribute_processes(uint length, uint* distribution) {
4264 // Not yet implemented.
4265 return false;
4266 }
4267
4268 bool os::bind_to_processor(uint processor_id) {
4269 // Not yet implemented.
4270 return false;
4271 }
4272
4273 ///
4274
4275 // Suspends the target using the signal mechanism and then grabs the PC before
4276 // resuming the target. Used by the flat-profiler only
4277 ExtendedPC os::get_thread_pc(Thread* thread) {
4278 // Make sure that it is called by the watcher for the VMThread
4279 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4280 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4281
4282 ExtendedPC epc;
4283
4284 OSThread* osthread = thread->osthread();
4285 if (do_suspend(osthread)) {
4286 if (osthread->ucontext() != NULL) {
4287 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4288 } else {
4289 // NULL context is unexpected, double-check this is the VMThread
4290 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4291 }
4292 do_resume(osthread);
4293 }
4294 // failure means pthread_kill failed for some reason - arguably this is
4295 // a fatal problem, but such problems are ignored elsewhere
4296
4297 return epc;
4298 }
4299
4300 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4301 {
4302 if (is_NPTL()) {
4303 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4304 } else {
4305 #ifndef IA64
4306 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4307 // word back to default 64bit precision if condvar is signaled. Java
4308 // wants 53bit precision. Save and restore current value.
4309 int fpu = get_fpu_control_word();
4310 #endif // IA64
4311 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4312 #ifndef IA64
4313 set_fpu_control_word(fpu);
4314 #endif // IA64
4315 return status;
4316 }
4317 }
4318
4319 ////////////////////////////////////////////////////////////////////////////////
4320 // debug support
4321
4322 static address same_page(address x, address y) {
4323 int page_bits = -os::vm_page_size();
4324 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4325 return x;
4326 else if (x > y)
4327 return (address)(intptr_t(y) | ~page_bits) + 1;
4328 else
4329 return (address)(intptr_t(y) & page_bits);
4330 }
4331
4332 bool os::find(address addr, outputStream* st) {
4333 Dl_info dlinfo;
4334 memset(&dlinfo, 0, sizeof(dlinfo));
4335 if (dladdr(addr, &dlinfo)) {
4336 st->print(PTR_FORMAT ": ", addr);
4337 if (dlinfo.dli_sname != NULL) {
4338 st->print("%s+%#x", dlinfo.dli_sname,
4339 addr - (intptr_t)dlinfo.dli_saddr);
4340 } else if (dlinfo.dli_fname) {
4341 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4342 } else {
4343 st->print("<absolute address>");
4344 }
4345 if (dlinfo.dli_fname) {
4346 st->print(" in %s", dlinfo.dli_fname);
4347 }
4348 if (dlinfo.dli_fbase) {
4349 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4350 }
4351 st->cr();
4352
4353 if (Verbose) {
4354 // decode some bytes around the PC
4355 address begin = same_page(addr-40, addr);
4356 address end = same_page(addr+40, addr);
4357 address lowest = (address) dlinfo.dli_sname;
4358 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4359 if (begin < lowest) begin = lowest;
4360 Dl_info dlinfo2;
4361 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4362 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4363 end = (address) dlinfo2.dli_saddr;
4364 Disassembler::decode(begin, end, st);
4365 }
4366 return true;
4367 }
4368 return false;
4369 }
4370
4371 ////////////////////////////////////////////////////////////////////////////////
4372 // misc
4373
4374 // This does not do anything on Linux. This is basically a hook for being
4375 // able to use structured exception handling (thread-local exception filters)
4376 // on, e.g., Win32.
4377 void
4378 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4379 JavaCallArguments* args, Thread* thread) {
4380 f(value, method, args, thread);
4381 }
4382
4383 void os::print_statistics() {
4384 }
4385
4386 int os::message_box(const char* title, const char* message) {
4387 int i;
4388 fdStream err(defaultStream::error_fd());
4389 for (i = 0; i < 78; i++) err.print_raw("=");
4390 err.cr();
4391 err.print_raw_cr(title);
4392 for (i = 0; i < 78; i++) err.print_raw("-");
4393 err.cr();
4394 err.print_raw_cr(message);
4395 for (i = 0; i < 78; i++) err.print_raw("=");
4396 err.cr();
4397
4398 char buf[16];
4399 // Prevent process from exiting upon "read error" without consuming all CPU
4400 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4401
4402 return buf[0] == 'y' || buf[0] == 'Y';
4403 }
4404
4405 int os::stat(const char *path, struct stat *sbuf) {
4406 char pathbuf[MAX_PATH];
4407 if (strlen(path) > MAX_PATH - 1) {
4408 errno = ENAMETOOLONG;
4409 return -1;
4410 }
4411 os::native_path(strcpy(pathbuf, path));
4412 return ::stat(pathbuf, sbuf);
4413 }
4414
4415 bool os::check_heap(bool force) {
4416 return true;
4417 }
4418
4419 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4420 return ::vsnprintf(buf, count, format, args);
4421 }
4422
4423 // Is a (classpath) directory empty?
4424 bool os::dir_is_empty(const char* path) {
4425 DIR *dir = NULL;
4426 struct dirent *ptr;
4427
4428 dir = opendir(path);
4429 if (dir == NULL) return true;
4430
4431 /* Scan the directory */
4432 bool result = true;
4433 char buf[sizeof(struct dirent) + MAX_PATH];
4434 while (result && (ptr = ::readdir(dir)) != NULL) {
4435 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4436 result = false;
4437 }
4438 }
4439 closedir(dir);
4440 return result;
4441 }
4442
4443 // This code originates from JDK's sysOpen and open64_w
4444 // from src/solaris/hpi/src/system_md.c
4445
4446 #ifndef O_DELETE
4447 #define O_DELETE 0x10000
4448 #endif
4449
4450 // Open a file. Unlink the file immediately after open returns
4451 // if the specified oflag has the O_DELETE flag set.
4452 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4453
4454 int os::open(const char *path, int oflag, int mode) {
4455
4456 if (strlen(path) > MAX_PATH - 1) {
4457 errno = ENAMETOOLONG;
4458 return -1;
4459 }
4460 int fd;
4461 int o_delete = (oflag & O_DELETE);
4462 oflag = oflag & ~O_DELETE;
4463
4464 fd = ::open64(path, oflag, mode);
4465 if (fd == -1) return -1;
4466
4467 //If the open succeeded, the file might still be a directory
4468 {
4469 struct stat64 buf64;
4470 int ret = ::fstat64(fd, &buf64);
4471 int st_mode = buf64.st_mode;
4472
4473 if (ret != -1) {
4474 if ((st_mode & S_IFMT) == S_IFDIR) {
4475 errno = EISDIR;
4476 ::close(fd);
4477 return -1;
4478 }
4479 } else {
4480 ::close(fd);
4481 return -1;
4482 }
4483 }
4484
4485 /*
4486 * All file descriptors that are opened in the JVM and not
4487 * specifically destined for a subprocess should have the
4488 * close-on-exec flag set. If we don't set it, then careless 3rd
4489 * party native code might fork and exec without closing all
4490 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4491 * UNIXProcess.c), and this in turn might:
4492 *
4493 * - cause end-of-file to fail to be detected on some file
4494 * descriptors, resulting in mysterious hangs, or
4495 *
4496 * - might cause an fopen in the subprocess to fail on a system
4497 * suffering from bug 1085341.
4498 *
4499 * (Yes, the default setting of the close-on-exec flag is a Unix
4500 * design flaw)
4501 *
4502 * See:
4503 * 1085341: 32-bit stdio routines should support file descriptors >255
4504 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4505 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4506 */
4507 #ifdef FD_CLOEXEC
4508 {
4509 int flags = ::fcntl(fd, F_GETFD);
4510 if (flags != -1)
4511 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4512 }
4513 #endif
4514
4515 if (o_delete != 0) {
4516 ::unlink(path);
4517 }
4518 return fd;
4519 }
4520
4521
4522 // create binary file, rewriting existing file if required
4523 int os::create_binary_file(const char* path, bool rewrite_existing) {
4524 int oflags = O_WRONLY | O_CREAT;
4525 if (!rewrite_existing) {
4526 oflags |= O_EXCL;
4527 }
4528 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4529 }
4530
4531 // return current position of file pointer
4532 jlong os::current_file_offset(int fd) {
4533 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4534 }
4535
4536 // move file pointer to the specified offset
4537 jlong os::seek_to_file_offset(int fd, jlong offset) {
4538 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4539 }
4540
4541 // This code originates from JDK's sysAvailable
4542 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4543
4544 int os::available(int fd, jlong *bytes) {
4545 jlong cur, end;
4546 int mode;
4547 struct stat64 buf64;
4548
4549 if (::fstat64(fd, &buf64) >= 0) {
4550 mode = buf64.st_mode;
4551 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4552 /*
4553 * XXX: is the following call interruptible? If so, this might
4554 * need to go through the INTERRUPT_IO() wrapper as for other
4555 * blocking, interruptible calls in this file.
4556 */
4557 int n;
4558 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4559 *bytes = n;
4560 return 1;
4561 }
4562 }
4563 }
4564 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4565 return 0;
4566 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4567 return 0;
4568 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4569 return 0;
4570 }
4571 *bytes = end - cur;
4572 return 1;
4573 }
4574
4575 int os::socket_available(int fd, jint *pbytes) {
4576 // Linux doc says EINTR not returned, unlike Solaris
4577 int ret = ::ioctl(fd, FIONREAD, pbytes);
4578
4579 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4580 // is expected to return 0 on failure and 1 on success to the jdk.
4581 return (ret < 0) ? 0 : 1;
4582 }
4583
4584 // Map a block of memory.
4585 char* os::map_memory(int fd, const char* file_name, size_t file_offset,
4586 char *addr, size_t bytes, bool read_only,
4587 bool allow_exec) {
4588 int prot;
4589 int flags;
4590
4591 if (read_only) {
4592 prot = PROT_READ;
4593 flags = MAP_SHARED;
4594 } else {
4595 prot = PROT_READ | PROT_WRITE;
4596 flags = MAP_PRIVATE;
4597 }
4598
4599 if (allow_exec) {
4600 prot |= PROT_EXEC;
4601 }
4602
4603 if (addr != NULL) {
4604 flags |= MAP_FIXED;
4605 }
4606
4607 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4608 fd, file_offset);
4609 if (mapped_address == MAP_FAILED) {
4610 return NULL;
4611 }
4612 return mapped_address;
4613 }
4614
4615
4616 // Remap a block of memory.
4617 char* os::remap_memory(int fd, const char* file_name, size_t file_offset,
4618 char *addr, size_t bytes, bool read_only,
4619 bool allow_exec) {
4620 // same as map_memory() on this OS
4621 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4622 allow_exec);
4623 }
4624
4625
4626 // Unmap a block of memory.
4627 bool os::unmap_memory(char* addr, size_t bytes) {
4628 return munmap(addr, bytes) == 0;
4629 }
4630
4631 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4632
4633 static clockid_t thread_cpu_clockid(Thread* thread) {
4634 pthread_t tid = thread->osthread()->pthread_id();
4635 clockid_t clockid;
4636
4637 // Get thread clockid
4638 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4639 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4640 return clockid;
4641 }
4642
4643 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4644 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4645 // of a thread.
4646 //
4647 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4648 // the fast estimate available on the platform.
4649
4650 jlong os::current_thread_cpu_time() {
4651 if (os::Linux::supports_fast_thread_cpu_time()) {
4652 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4653 } else {
4654 // return user + sys since the cost is the same
4655 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4656 }
4657 }
4658
4659 jlong os::thread_cpu_time(Thread* thread) {
4660 // consistent with what current_thread_cpu_time() returns
4661 if (os::Linux::supports_fast_thread_cpu_time()) {
4662 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4663 } else {
4664 return slow_thread_cpu_time(thread, true /* user + sys */);
4665 }
4666 }
4667
4668 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4669 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4670 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4671 } else {
4672 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4673 }
4674 }
4675
4676 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4677 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4678 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4679 } else {
4680 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4681 }
4682 }
4683
4684 //
4685 // -1 on error.
4686 //
4687
4688 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4689 static bool proc_pid_cpu_avail = true;
4690 static bool proc_task_unchecked = true;
4691 static const char *proc_stat_path = "/proc/%d/stat";
4692 pid_t tid = thread->osthread()->thread_id();
4693 int i;
4694 char *s;
4695 char stat[2048];
4696 int statlen;
4697 char proc_name[64];
4698 int count;
4699 long sys_time, user_time;
4700 char string[64];
4701 char cdummy;
4702 int idummy;
4703 long ldummy;
4704 FILE *fp;
4705
4706 // We first try accessing /proc/<pid>/cpu since this is faster to
4707 // process. If this file is not present (linux kernels 2.5 and above)
4708 // then we open /proc/<pid>/stat.
4709 if ( proc_pid_cpu_avail ) {
4710 sprintf(proc_name, "/proc/%d/cpu", tid);
4711 fp = fopen(proc_name, "r");
4712 if ( fp != NULL ) {
4713 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4714 fclose(fp);
4715 if ( count != 3 ) return -1;
4716
4717 if (user_sys_cpu_time) {
4718 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4719 } else {
4720 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4721 }
4722 }
4723 else proc_pid_cpu_avail = false;
4724 }
4725
4726 // The /proc/<tid>/stat aggregates per-process usage on
4727 // new Linux kernels 2.6+ where NPTL is supported.
4728 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4729 // See bug 6328462.
4730 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4731 // and possibly in some other cases, so we check its availability.
4732 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4733 // This is executed only once
4734 proc_task_unchecked = false;
4735 fp = fopen("/proc/self/task", "r");
4736 if (fp != NULL) {
4737 proc_stat_path = "/proc/self/task/%d/stat";
4738 fclose(fp);
4739 }
4740 }
4741
4742 sprintf(proc_name, proc_stat_path, tid);
4743 fp = fopen(proc_name, "r");
4744 if ( fp == NULL ) return -1;
4745 statlen = fread(stat, 1, 2047, fp);
4746 stat[statlen] = '\0';
4747 fclose(fp);
4748
4749 // Skip pid and the command string. Note that we could be dealing with
4750 // weird command names, e.g. user could decide to rename java launcher
4751 // to "java 1.4.2 :)", then the stat file would look like
4752 // 1234 (java 1.4.2 :)) R ... ...
4753 // We don't really need to know the command string, just find the last
4754 // occurrence of ")" and then start parsing from there. See bug 4726580.
4755 s = strrchr(stat, ')');
4756 i = 0;
4757 if (s == NULL ) return -1;
4758
4759 // Skip blank chars
4760 do s++; while (isspace(*s));
4761
4762 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4763 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4764 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4765 &user_time, &sys_time);
4766 if ( count != 13 ) return -1;
4767 if (user_sys_cpu_time) {
4768 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4769 } else {
4770 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4771 }
4772 }
4773
4774 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4775 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4776 info_ptr->may_skip_backward = false; // elapsed time not wall time
4777 info_ptr->may_skip_forward = false; // elapsed time not wall time
4778 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4779 }
4780
4781 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4782 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4783 info_ptr->may_skip_backward = false; // elapsed time not wall time
4784 info_ptr->may_skip_forward = false; // elapsed time not wall time
4785 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4786 }
4787
4788 bool os::is_thread_cpu_time_supported() {
4789 return true;
4790 }
4791
4792 // System loadavg support. Returns -1 if load average cannot be obtained.
4793 // Linux doesn't yet have a (official) notion of processor sets,
4794 // so just return the system wide load average.
4795 int os::loadavg(double loadavg[], int nelem) {
4796 return ::getloadavg(loadavg, nelem);
4797 }
4798
4799 void os::pause() {
4800 char filename[MAX_PATH];
4801 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4802 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4803 } else {
4804 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4805 }
4806
4807 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4808 if (fd != -1) {
4809 struct stat buf;
4810 ::close(fd);
4811 while (::stat(filename, &buf) == 0) {
4812 (void)::poll(NULL, 0, 100);
4813 }
4814 } else {
4815 jio_fprintf(stderr,
4816 "Could not open pause file '%s', continuing immediately.\n", filename);
4817 }
4818 }
4819
4820
4821 // Refer to the comments in os_solaris.cpp park-unpark.
4822 //
4823 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4824 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4825 // For specifics regarding the bug see GLIBC BUGID 261237 :
4826 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4827 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4828 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4829 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4830 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4831 // and monitorenter when we're using 1-0 locking. All those operations may result in
4832 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4833 // of libpthread avoids the problem, but isn't practical.
4834 //
4835 // Possible remedies:
4836 //
4837 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4838 // This is palliative and probabilistic, however. If the thread is preempted
4839 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4840 // than the minimum period may have passed, and the abstime may be stale (in the
4841 // past) resultin in a hang. Using this technique reduces the odds of a hang
4842 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4843 //
4844 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4845 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4846 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4847 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4848 // thread.
4849 //
4850 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4851 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4852 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4853 // This also works well. In fact it avoids kernel-level scalability impediments
4854 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4855 // timers in a graceful fashion.
4856 //
4857 // 4. When the abstime value is in the past it appears that control returns
4858 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4859 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4860 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4861 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4862 // It may be possible to avoid reinitialization by checking the return
4863 // value from pthread_cond_timedwait(). In addition to reinitializing the
4864 // condvar we must establish the invariant that cond_signal() is only called
4865 // within critical sections protected by the adjunct mutex. This prevents
4866 // cond_signal() from "seeing" a condvar that's in the midst of being
4867 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4868 // desirable signal-after-unlock optimization that avoids futile context switching.
4869 //
4870 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4871 // structure when a condvar is used or initialized. cond_destroy() would
4872 // release the helper structure. Our reinitialize-after-timedwait fix
4873 // put excessive stress on malloc/free and locks protecting the c-heap.
4874 //
4875 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4876 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4877 // and only enabling the work-around for vulnerable environments.
4878
4879 // utility to compute the abstime argument to timedwait:
4880 // millis is the relative timeout time
4881 // abstime will be the absolute timeout time
4882 // TODO: replace compute_abstime() with unpackTime()
4883
4884 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4885 if (millis < 0) millis = 0;
4886 struct timeval now;
4887 int status = gettimeofday(&now, NULL);
4888 assert(status == 0, "gettimeofday");
4889 jlong seconds = millis / 1000;
4890 millis %= 1000;
4891 if (seconds > 50000000) { // see man cond_timedwait(3T)
4892 seconds = 50000000;
4893 }
4894 abstime->tv_sec = now.tv_sec + seconds;
4895 long usec = now.tv_usec + millis * 1000;
4896 if (usec >= 1000000) {
4897 abstime->tv_sec += 1;
4898 usec -= 1000000;
4899 }
4900 abstime->tv_nsec = usec * 1000;
4901 return abstime;
4902 }
4903
4904
4905 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4906 // Conceptually TryPark() should be equivalent to park(0).
4907
4908 int os::PlatformEvent::TryPark() {
4909 for (;;) {
4910 const int v = _Event ;
4911 guarantee ((v == 0) || (v == 1), "invariant") ;
4912 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4913 }
4914 }
4915
4916 void os::PlatformEvent::park() { // AKA "down()"
4917 // Invariant: Only the thread associated with the Event/PlatformEvent
4918 // may call park().
4919 // TODO: assert that _Assoc != NULL or _Assoc == Self
4920 int v ;
4921 for (;;) {
4922 v = _Event ;
4923 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4924 }
4925 guarantee (v >= 0, "invariant") ;
4926 if (v == 0) {
4927 // Do this the hard way by blocking ...
4928 int status = pthread_mutex_lock(_mutex);
4929 assert_status(status == 0, status, "mutex_lock");
4930 guarantee (_nParked == 0, "invariant") ;
4931 ++ _nParked ;
4932 while (_Event < 0) {
4933 status = pthread_cond_wait(_cond, _mutex);
4934 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
4935 // Treat this the same as if the wait was interrupted
4936 if (status == ETIME) { status = EINTR; }
4937 assert_status(status == 0 || status == EINTR, status, "cond_wait");
4938 }
4939 -- _nParked ;
4940
4941 // In theory we could move the ST of 0 into _Event past the unlock(),
4942 // but then we'd need a MEMBAR after the ST.
4943 _Event = 0 ;
4944 status = pthread_mutex_unlock(_mutex);
4945 assert_status(status == 0, status, "mutex_unlock");
4946 }
4947 guarantee (_Event >= 0, "invariant") ;
4948 }
4949
4950 int os::PlatformEvent::park(jlong millis) {
4951 guarantee (_nParked == 0, "invariant") ;
4952
4953 int v ;
4954 for (;;) {
4955 v = _Event ;
4956 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4957 }
4958 guarantee (v >= 0, "invariant") ;
4959 if (v != 0) return OS_OK ;
4960
4961 // We do this the hard way, by blocking the thread.
4962 // Consider enforcing a minimum timeout value.
4963 struct timespec abst;
4964 compute_abstime(&abst, millis);
4965
4966 int ret = OS_TIMEOUT;
4967 int status = pthread_mutex_lock(_mutex);
4968 assert_status(status == 0, status, "mutex_lock");
4969 guarantee (_nParked == 0, "invariant") ;
4970 ++_nParked ;
4971
4972 // Object.wait(timo) will return because of
4973 // (a) notification
4974 // (b) timeout
4975 // (c) thread.interrupt
4976 //
4977 // Thread.interrupt and object.notify{All} both call Event::set.
4978 // That is, we treat thread.interrupt as a special case of notification.
4979 // The underlying Solaris implementation, cond_timedwait, admits
4980 // spurious/premature wakeups, but the JLS/JVM spec prevents the
4981 // JVM from making those visible to Java code. As such, we must
4982 // filter out spurious wakeups. We assume all ETIME returns are valid.
4983 //
4984 // TODO: properly differentiate simultaneous notify+interrupt.
4985 // In that case, we should propagate the notify to another waiter.
4986
4987 while (_Event < 0) {
4988 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
4989 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
4990 pthread_cond_destroy (_cond);
4991 pthread_cond_init (_cond, NULL) ;
4992 }
4993 assert_status(status == 0 || status == EINTR ||
4994 status == ETIME || status == ETIMEDOUT,
4995 status, "cond_timedwait");
4996 if (!FilterSpuriousWakeups) break ; // previous semantics
4997 if (status == ETIME || status == ETIMEDOUT) break ;
4998 // We consume and ignore EINTR and spurious wakeups.
4999 }
5000 --_nParked ;
5001 if (_Event >= 0) {
5002 ret = OS_OK;
5003 }
5004 _Event = 0 ;
5005 status = pthread_mutex_unlock(_mutex);
5006 assert_status(status == 0, status, "mutex_unlock");
5007 assert (_nParked == 0, "invariant") ;
5008 return ret;
5009 }
5010
5011 void os::PlatformEvent::unpark() {
5012 int v, AnyWaiters ;
5013 for (;;) {
5014 v = _Event ;
5015 if (v > 0) {
5016 // The LD of _Event could have reordered or be satisfied
5017 // by a read-aside from this processor's write buffer.
5018 // To avoid problems execute a barrier and then
5019 // ratify the value.
5020 OrderAccess::fence() ;
5021 if (_Event == v) return ;
5022 continue ;
5023 }
5024 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5025 }
5026 if (v < 0) {
5027 // Wait for the thread associated with the event to vacate
5028 int status = pthread_mutex_lock(_mutex);
5029 assert_status(status == 0, status, "mutex_lock");
5030 AnyWaiters = _nParked ;
5031 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5032 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5033 AnyWaiters = 0 ;
5034 pthread_cond_signal (_cond);
5035 }
5036 status = pthread_mutex_unlock(_mutex);
5037 assert_status(status == 0, status, "mutex_unlock");
5038 if (AnyWaiters != 0) {
5039 status = pthread_cond_signal(_cond);
5040 assert_status(status == 0, status, "cond_signal");
5041 }
5042 }
5043
5044 // Note that we signal() _after dropping the lock for "immortal" Events.
5045 // This is safe and avoids a common class of futile wakeups. In rare
5046 // circumstances this can cause a thread to return prematurely from
5047 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5048 // simply re-test the condition and re-park itself.
5049 }
5050
5051
5052 // JSR166
5053 // -------------------------------------------------------
5054
5055 /*
5056 * The solaris and linux implementations of park/unpark are fairly
5057 * conservative for now, but can be improved. They currently use a
5058 * mutex/condvar pair, plus a a count.
5059 * Park decrements count if > 0, else does a condvar wait. Unpark
5060 * sets count to 1 and signals condvar. Only one thread ever waits
5061 * on the condvar. Contention seen when trying to park implies that someone
5062 * is unparking you, so don't wait. And spurious returns are fine, so there
5063 * is no need to track notifications.
5064 */
5065
5066
5067 #define NANOSECS_PER_SEC 1000000000
5068 #define NANOSECS_PER_MILLISEC 1000000
5069 #define MAX_SECS 100000000
5070 /*
5071 * This code is common to linux and solaris and will be moved to a
5072 * common place in dolphin.
5073 *
5074 * The passed in time value is either a relative time in nanoseconds
5075 * or an absolute time in milliseconds. Either way it has to be unpacked
5076 * into suitable seconds and nanoseconds components and stored in the
5077 * given timespec structure.
5078 * Given time is a 64-bit value and the time_t used in the timespec is only
5079 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5080 * overflow if times way in the future are given. Further on Solaris versions
5081 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5082 * number of seconds, in abstime, is less than current_time + 100,000,000.
5083 * As it will be 28 years before "now + 100000000" will overflow we can
5084 * ignore overflow and just impose a hard-limit on seconds using the value
5085 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5086 * years from "now".
5087 */
5088
5089 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5090 assert (time > 0, "convertTime");
5091
5092 struct timeval now;
5093 int status = gettimeofday(&now, NULL);
5094 assert(status == 0, "gettimeofday");
5095
5096 time_t max_secs = now.tv_sec + MAX_SECS;
5097
5098 if (isAbsolute) {
5099 jlong secs = time / 1000;
5100 if (secs > max_secs) {
5101 absTime->tv_sec = max_secs;
5102 }
5103 else {
5104 absTime->tv_sec = secs;
5105 }
5106 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5107 }
5108 else {
5109 jlong secs = time / NANOSECS_PER_SEC;
5110 if (secs >= MAX_SECS) {
5111 absTime->tv_sec = max_secs;
5112 absTime->tv_nsec = 0;
5113 }
5114 else {
5115 absTime->tv_sec = now.tv_sec + secs;
5116 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5117 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5118 absTime->tv_nsec -= NANOSECS_PER_SEC;
5119 ++absTime->tv_sec; // note: this must be <= max_secs
5120 }
5121 }
5122 }
5123 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5124 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5125 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5126 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5127 }
5128
5129 void Parker::park(bool isAbsolute, jlong time) {
5130 // Optional fast-path check:
5131 // Return immediately if a permit is available.
5132 if (_counter > 0) {
5133 _counter = 0 ;
5134 OrderAccess::fence();
5135 return ;
5136 }
5137
5138 Thread* thread = Thread::current();
5139 assert(thread->is_Java_thread(), "Must be JavaThread");
5140 JavaThread *jt = (JavaThread *)thread;
5141
5142 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5143 // Check interrupt before trying to wait
5144 if (Thread::is_interrupted(thread, false)) {
5145 return;
5146 }
5147
5148 // Next, demultiplex/decode time arguments
5149 timespec absTime;
5150 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5151 return;
5152 }
5153 if (time > 0) {
5154 unpackTime(&absTime, isAbsolute, time);
5155 }
5156
5157
5158 // Enter safepoint region
5159 // Beware of deadlocks such as 6317397.
5160 // The per-thread Parker:: mutex is a classic leaf-lock.
5161 // In particular a thread must never block on the Threads_lock while
5162 // holding the Parker:: mutex. If safepoints are pending both the
5163 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5164 ThreadBlockInVM tbivm(jt);
5165
5166 // Don't wait if cannot get lock since interference arises from
5167 // unblocking. Also. check interrupt before trying wait
5168 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5169 return;
5170 }
5171
5172 int status ;
5173 if (_counter > 0) { // no wait needed
5174 _counter = 0;
5175 status = pthread_mutex_unlock(_mutex);
5176 assert (status == 0, "invariant") ;
5177 OrderAccess::fence();
5178 return;
5179 }
5180
5181 #ifdef ASSERT
5182 // Don't catch signals while blocked; let the running threads have the signals.
5183 // (This allows a debugger to break into the running thread.)
5184 sigset_t oldsigs;
5185 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5186 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5187 #endif
5188
5189 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5190 jt->set_suspend_equivalent();
5191 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5192
5193 if (time == 0) {
5194 status = pthread_cond_wait (_cond, _mutex) ;
5195 } else {
5196 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5197 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5198 pthread_cond_destroy (_cond) ;
5199 pthread_cond_init (_cond, NULL);
5200 }
5201 }
5202 assert_status(status == 0 || status == EINTR ||
5203 status == ETIME || status == ETIMEDOUT,
5204 status, "cond_timedwait");
5205
5206 #ifdef ASSERT
5207 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5208 #endif
5209
5210 _counter = 0 ;
5211 status = pthread_mutex_unlock(_mutex) ;
5212 assert_status(status == 0, status, "invariant") ;
5213 // If externally suspended while waiting, re-suspend
5214 if (jt->handle_special_suspend_equivalent_condition()) {
5215 jt->java_suspend_self();
5216 }
5217
5218 OrderAccess::fence();
5219 }
5220
5221 void Parker::unpark() {
5222 int s, status ;
5223 status = pthread_mutex_lock(_mutex);
5224 assert (status == 0, "invariant") ;
5225 s = _counter;
5226 _counter = 1;
5227 if (s < 1) {
5228 if (WorkAroundNPTLTimedWaitHang) {
5229 status = pthread_cond_signal (_cond) ;
5230 assert (status == 0, "invariant") ;
5231 status = pthread_mutex_unlock(_mutex);
5232 assert (status == 0, "invariant") ;
5233 } else {
5234 status = pthread_mutex_unlock(_mutex);
5235 assert (status == 0, "invariant") ;
5236 status = pthread_cond_signal (_cond) ;
5237 assert (status == 0, "invariant") ;
5238 }
5239 } else {
5240 pthread_mutex_unlock(_mutex);
5241 assert (status == 0, "invariant") ;
5242 }
5243 }
5244
5245
5246 extern char** environ;
5247
5248 #ifndef __NR_fork
5249 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5250 #endif
5251
5252 #ifndef __NR_execve
5253 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5254 #endif
5255
5256 // Run the specified command in a separate process. Return its exit value,
5257 // or -1 on failure (e.g. can't fork a new process).
5258 // Unlike system(), this function can be called from signal handler. It
5259 // doesn't block SIGINT et al.
5260 int os::fork_and_exec(char* cmd) {
5261 const char * argv[4] = {"sh", "-c", cmd, NULL};
5262
5263 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5264 // pthread_atfork handlers and reset pthread library. All we need is a
5265 // separate process to execve. Make a direct syscall to fork process.
5266 // On IA64 there's no fork syscall, we have to use fork() and hope for
5267 // the best...
5268 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5269 IA64_ONLY(fork();)
5270
5271 if (pid < 0) {
5272 // fork failed
5273 return -1;
5274
5275 } else if (pid == 0) {
5276 // child process
5277
5278 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5279 // first to kill every thread on the thread list. Because this list is
5280 // not reset by fork() (see notes above), execve() will instead kill
5281 // every thread in the parent process. We know this is the only thread
5282 // in the new process, so make a system call directly.
5283 // IA64 should use normal execve() from glibc to match the glibc fork()
5284 // above.
5285 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5286 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5287
5288 // execve failed
5289 _exit(-1);
5290
5291 } else {
5292 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5293 // care about the actual exit code, for now.
5294
5295 int status;
5296
5297 // Wait for the child process to exit. This returns immediately if
5298 // the child has already exited. */
5299 while (waitpid(pid, &status, 0) < 0) {
5300 switch (errno) {
5301 case ECHILD: return 0;
5302 case EINTR: break;
5303 default: return -1;
5304 }
5305 }
5306
5307 if (WIFEXITED(status)) {
5308 // The child exited normally; get its exit code.
5309 return WEXITSTATUS(status);
5310 } else if (WIFSIGNALED(status)) {
5311 // The child exited because of a signal
5312 // The best value to return is 0x80 + signal number,
5313 // because that is what all Unix shells do, and because
5314 // it allows callers to distinguish between process exit and
5315 // process death by signal.
5316 return 0x80 + WTERMSIG(status);
5317 } else {
5318 // Unknown exit code; pass it through
5319 return status;
5320 }
5321 }
5322 }
5323
5324 // is_headless_jre()
5325 //
5326 // Test for the existence of libmawt in motif21 or xawt directories
5327 // in order to report if we are running in a headless jre
5328 //
5329 bool os::is_headless_jre() {
5330 struct stat statbuf;
5331 char buf[MAXPATHLEN];
5332 char libmawtpath[MAXPATHLEN];
5333 const char *xawtstr = "/xawt/libmawt.so";
5334 const char *motifstr = "/motif21/libmawt.so";
5335 char *p;
5336
5337 // Get path to libjvm.so
5338 os::jvm_path(buf, sizeof(buf));
5339
5340 // Get rid of libjvm.so
5341 p = strrchr(buf, '/');
5342 if (p == NULL) return false;
5343 else *p = '\0';
5344
5345 // Get rid of client or server
5346 p = strrchr(buf, '/');
5347 if (p == NULL) return false;
5348 else *p = '\0';
5349
5350 // check xawt/libmawt.so
5351 strcpy(libmawtpath, buf);
5352 strcat(libmawtpath, xawtstr);
5353 if (::stat(libmawtpath, &statbuf) == 0) return false;
5354
5355 // check motif21/libmawt.so
5356 strcpy(libmawtpath, buf);
5357 strcat(libmawtpath, motifstr);
5358 if (::stat(libmawtpath, &statbuf) == 0) return false;
5359
5360 return true;
5361 }
5362