1 /*
2 * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_share_linux.hpp"
40 #include "prims/jniFastGetField.hpp"
41 #include "prims/jvm.h"
42 #include "prims/jvm_misc.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/extendedPC.hpp"
45 #include "runtime/globals.hpp"
46 #include "runtime/interfaceSupport.hpp"
47 #include "runtime/init.hpp"
48 #include "runtime/java.hpp"
49 #include "runtime/javaCalls.hpp"
50 #include "runtime/mutexLocker.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/osThread.hpp"
53 #include "runtime/perfMemory.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/statSampler.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "runtime/thread.inline.hpp"
58 #include "runtime/threadCritical.hpp"
59 #include "runtime/timer.hpp"
60 #include "services/attachListener.hpp"
61 #include "services/memTracker.hpp"
62 #include "services/runtimeService.hpp"
63 #include "utilities/decoder.hpp"
64 #include "utilities/defaultStream.hpp"
65 #include "utilities/events.hpp"
66 #include "utilities/elfFile.hpp"
67 #include "utilities/growableArray.hpp"
68 #include "utilities/vmError.hpp"
69
70 // put OS-includes here
71 # include <sys/types.h>
72 # include <sys/mman.h>
73 # include <sys/stat.h>
74 # include <sys/select.h>
75 # include <pthread.h>
76 # include <signal.h>
77 # include <errno.h>
78 # include <dlfcn.h>
79 # include <stdio.h>
80 # include <unistd.h>
81 # include <sys/resource.h>
82 # include <pthread.h>
83 # include <sys/stat.h>
84 # include <sys/time.h>
85 # include <sys/times.h>
86 # include <sys/utsname.h>
87 # include <sys/socket.h>
88 # include <sys/wait.h>
89 # include <pwd.h>
90 # include <poll.h>
91 # include <semaphore.h>
92 # include <fcntl.h>
93 # include <string.h>
94 # include <syscall.h>
95 # include <sys/sysinfo.h>
96 # include <gnu/libc-version.h>
97 # include <sys/ipc.h>
98 # include <sys/shm.h>
99 # include <link.h>
100 # include <stdint.h>
101 # include <inttypes.h>
102 # include <sys/ioctl.h>
103
104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
105 // getrusage() is prepared to handle the associated failure.
106 #ifndef RUSAGE_THREAD
107 #define RUSAGE_THREAD (1) /* only the calling thread */
108 #endif
109
110 #define MAX_PATH (2 * K)
111
112 // for timer info max values which include all bits
113 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
114
115 #define LARGEPAGES_BIT (1 << 6)
116 ////////////////////////////////////////////////////////////////////////////////
117 // global variables
118 julong os::Linux::_physical_memory = 0;
119
120 address os::Linux::_initial_thread_stack_bottom = NULL;
121 uintptr_t os::Linux::_initial_thread_stack_size = 0;
122
123 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
124 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
125 Mutex* os::Linux::_createThread_lock = NULL;
126 pthread_t os::Linux::_main_thread;
127 int os::Linux::_page_size = -1;
128 const int os::Linux::_vm_default_page_size = (8 * K);
129 bool os::Linux::_is_floating_stack = false;
130 bool os::Linux::_is_NPTL = false;
131 bool os::Linux::_supports_fast_thread_cpu_time = false;
132 const char * os::Linux::_glibc_version = NULL;
133 const char * os::Linux::_libpthread_version = NULL;
134 pthread_condattr_t os::Linux::_condattr[1];
135
136 static jlong initial_time_count=0;
137
138 static int clock_tics_per_sec = 100;
139
140 // For diagnostics to print a message once. see run_periodic_checks
141 static sigset_t check_signal_done;
142 static bool check_signals = true;;
143
144 static pid_t _initial_pid = 0;
145
146 /* Signal number used to suspend/resume a thread */
147
148 /* do not use any signal number less than SIGSEGV, see 4355769 */
149 static int SR_signum = SIGUSR2;
150 sigset_t SR_sigset;
151
152 /* Used to protect dlsym() calls */
153 static pthread_mutex_t dl_mutex;
154
155 // Declarations
156 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
157
158 #ifdef JAVASE_EMBEDDED
159 class MemNotifyThread: public Thread {
160 friend class VMStructs;
161 public:
162 virtual void run();
163
164 private:
165 static MemNotifyThread* _memnotify_thread;
166 int _fd;
167
168 public:
169
170 // Constructor
171 MemNotifyThread(int fd);
172
173 // Tester
174 bool is_memnotify_thread() const { return true; }
175
176 // Printing
177 char* name() const { return (char*)"Linux MemNotify Thread"; }
178
179 // Returns the single instance of the MemNotifyThread
180 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
181
182 // Create and start the single instance of MemNotifyThread
183 static void start();
184 };
185 #endif // JAVASE_EMBEDDED
186
187 // utility functions
188
189 static int SR_initialize();
190
191 julong os::available_memory() {
192 return Linux::available_memory();
193 }
194
195 julong os::Linux::available_memory() {
196 // values in struct sysinfo are "unsigned long"
197 struct sysinfo si;
198 sysinfo(&si);
199
200 return (julong)si.freeram * si.mem_unit;
201 }
202
203 julong os::physical_memory() {
204 return Linux::physical_memory();
205 }
206
207 ////////////////////////////////////////////////////////////////////////////////
208 // environment support
209
210 bool os::getenv(const char* name, char* buf, int len) {
211 const char* val = ::getenv(name);
212 if (val != NULL && strlen(val) < (size_t)len) {
213 strcpy(buf, val);
214 return true;
215 }
216 if (len > 0) buf[0] = 0; // return a null string
217 return false;
218 }
219
220
221 // Return true if user is running as root.
222
223 bool os::have_special_privileges() {
224 static bool init = false;
225 static bool privileges = false;
226 if (!init) {
227 privileges = (getuid() != geteuid()) || (getgid() != getegid());
228 init = true;
229 }
230 return privileges;
231 }
232
233
234 #ifndef SYS_gettid
235 // i386: 224, ia64: 1105, amd64: 186, sparc 143
236 #ifdef __ia64__
237 #define SYS_gettid 1105
238 #elif __i386__
239 #define SYS_gettid 224
240 #elif __amd64__
241 #define SYS_gettid 186
242 #elif __sparc__
243 #define SYS_gettid 143
244 #else
245 #error define gettid for the arch
246 #endif
247 #endif
248
249 // Cpu architecture string
250 #if defined(ZERO)
251 static char cpu_arch[] = ZERO_LIBARCH;
252 #elif defined(IA64)
253 static char cpu_arch[] = "ia64";
254 #elif defined(IA32)
255 static char cpu_arch[] = "i386";
256 #elif defined(AMD64)
257 static char cpu_arch[] = "amd64";
258 #elif defined(ARM)
259 static char cpu_arch[] = "arm";
260 #elif defined(PPC)
261 static char cpu_arch[] = "ppc";
262 #elif defined(SPARC)
263 # ifdef _LP64
264 static char cpu_arch[] = "sparcv9";
265 # else
266 static char cpu_arch[] = "sparc";
267 # endif
268 #else
269 #error Add appropriate cpu_arch setting
270 #endif
271
272
273 // pid_t gettid()
274 //
275 // Returns the kernel thread id of the currently running thread. Kernel
276 // thread id is used to access /proc.
277 //
278 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
279 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
280 //
281 pid_t os::Linux::gettid() {
282 int rslt = syscall(SYS_gettid);
283 if (rslt == -1) {
284 // old kernel, no NPTL support
285 return getpid();
286 } else {
287 return (pid_t)rslt;
288 }
289 }
290
291 // Most versions of linux have a bug where the number of processors are
292 // determined by looking at the /proc file system. In a chroot environment,
293 // the system call returns 1. This causes the VM to act as if it is
294 // a single processor and elide locking (see is_MP() call).
295 static bool unsafe_chroot_detected = false;
296 static const char *unstable_chroot_error = "/proc file system not found.\n"
297 "Java may be unstable running multithreaded in a chroot "
298 "environment on Linux when /proc filesystem is not mounted.";
299
300 void os::Linux::initialize_system_info() {
301 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
302 if (processor_count() == 1) {
303 pid_t pid = os::Linux::gettid();
304 char fname[32];
305 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
306 FILE *fp = fopen(fname, "r");
307 if (fp == NULL) {
308 unsafe_chroot_detected = true;
309 } else {
310 fclose(fp);
311 }
312 }
313 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
314 assert(processor_count() > 0, "linux error");
315 }
316
317 void os::init_system_properties_values() {
318 // char arch[12];
319 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
320
321 // The next steps are taken in the product version:
322 //
323 // Obtain the JAVA_HOME value from the location of libjvm.so.
324 // This library should be located at:
325 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
326 //
327 // If "/jre/lib/" appears at the right place in the path, then we
328 // assume libjvm.so is installed in a JDK and we use this path.
329 //
330 // Otherwise exit with message: "Could not create the Java virtual machine."
331 //
332 // The following extra steps are taken in the debugging version:
333 //
334 // If "/jre/lib/" does NOT appear at the right place in the path
335 // instead of exit check for $JAVA_HOME environment variable.
336 //
337 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
338 // then we append a fake suffix "hotspot/libjvm.so" to this path so
339 // it looks like libjvm.so is installed there
340 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
341 //
342 // Otherwise exit.
343 //
344 // Important note: if the location of libjvm.so changes this
345 // code needs to be changed accordingly.
346
347 // The next few definitions allow the code to be verbatim:
348 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
349 #define getenv(n) ::getenv(n)
350
351 /*
352 * See ld(1):
353 * The linker uses the following search paths to locate required
354 * shared libraries:
355 * 1: ...
356 * ...
357 * 7: The default directories, normally /lib and /usr/lib.
358 */
359 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
360 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
361 #else
362 #define DEFAULT_LIBPATH "/lib:/usr/lib"
363 #endif
364
365 #define EXTENSIONS_DIR "/lib/ext"
366 #define ENDORSED_DIR "/lib/endorsed"
367 #define REG_DIR "/usr/java/packages"
368
369 {
370 /* sysclasspath, java_home, dll_dir */
371 {
372 char *home_path;
373 char *dll_path;
374 char *pslash;
375 char buf[MAXPATHLEN];
376 os::jvm_path(buf, sizeof(buf));
377
378 // Found the full path to libjvm.so.
379 // Now cut the path to <java_home>/jre if we can.
380 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
381 pslash = strrchr(buf, '/');
382 if (pslash != NULL)
383 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
384 dll_path = malloc(strlen(buf) + 1);
385 if (dll_path == NULL)
386 return;
387 strcpy(dll_path, buf);
388 Arguments::set_dll_dir(dll_path);
389
390 if (pslash != NULL) {
391 pslash = strrchr(buf, '/');
392 if (pslash != NULL) {
393 *pslash = '\0'; /* get rid of /<arch> */
394 pslash = strrchr(buf, '/');
395 if (pslash != NULL)
396 *pslash = '\0'; /* get rid of /lib */
397 }
398 }
399
400 home_path = malloc(strlen(buf) + 1);
401 if (home_path == NULL)
402 return;
403 strcpy(home_path, buf);
404 Arguments::set_java_home(home_path);
405
406 if (!set_boot_path('/', ':'))
407 return;
408 }
409
410 /*
411 * Where to look for native libraries
412 *
413 * Note: Due to a legacy implementation, most of the library path
414 * is set in the launcher. This was to accomodate linking restrictions
415 * on legacy Linux implementations (which are no longer supported).
416 * Eventually, all the library path setting will be done here.
417 *
418 * However, to prevent the proliferation of improperly built native
419 * libraries, the new path component /usr/java/packages is added here.
420 * Eventually, all the library path setting will be done here.
421 */
422 {
423 char *ld_library_path;
424
425 /*
426 * Construct the invariant part of ld_library_path. Note that the
427 * space for the colon and the trailing null are provided by the
428 * nulls included by the sizeof operator (so actually we allocate
429 * a byte more than necessary).
430 */
431 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
432 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
433 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
434
435 /*
436 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
437 * should always exist (until the legacy problem cited above is
438 * addressed).
439 */
440 char *v = getenv("LD_LIBRARY_PATH");
441 if (v != NULL) {
442 char *t = ld_library_path;
443 /* That's +1 for the colon and +1 for the trailing '\0' */
444 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
445 sprintf(ld_library_path, "%s:%s", v, t);
446 }
447 Arguments::set_library_path(ld_library_path);
448 }
449
450 /*
451 * Extensions directories.
452 *
453 * Note that the space for the colon and the trailing null are provided
454 * by the nulls included by the sizeof operator (so actually one byte more
455 * than necessary is allocated).
456 */
457 {
458 char *buf = malloc(strlen(Arguments::get_java_home()) +
459 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
460 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
461 Arguments::get_java_home());
462 Arguments::set_ext_dirs(buf);
463 }
464
465 /* Endorsed standards default directory. */
466 {
467 char * buf;
468 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
469 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
470 Arguments::set_endorsed_dirs(buf);
471 }
472 }
473
474 #undef malloc
475 #undef getenv
476 #undef EXTENSIONS_DIR
477 #undef ENDORSED_DIR
478
479 // Done
480 return;
481 }
482
483 ////////////////////////////////////////////////////////////////////////////////
484 // breakpoint support
485
486 void os::breakpoint() {
487 BREAKPOINT;
488 }
489
490 extern "C" void breakpoint() {
491 // use debugger to set breakpoint here
492 }
493
494 ////////////////////////////////////////////////////////////////////////////////
495 // signal support
496
497 debug_only(static bool signal_sets_initialized = false);
498 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
499
500 bool os::Linux::is_sig_ignored(int sig) {
501 struct sigaction oact;
502 sigaction(sig, (struct sigaction*)NULL, &oact);
503 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
504 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
505 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
506 return true;
507 else
508 return false;
509 }
510
511 void os::Linux::signal_sets_init() {
512 // Should also have an assertion stating we are still single-threaded.
513 assert(!signal_sets_initialized, "Already initialized");
514 // Fill in signals that are necessarily unblocked for all threads in
515 // the VM. Currently, we unblock the following signals:
516 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
517 // by -Xrs (=ReduceSignalUsage));
518 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
519 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
520 // the dispositions or masks wrt these signals.
521 // Programs embedding the VM that want to use the above signals for their
522 // own purposes must, at this time, use the "-Xrs" option to prevent
523 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
524 // (See bug 4345157, and other related bugs).
525 // In reality, though, unblocking these signals is really a nop, since
526 // these signals are not blocked by default.
527 sigemptyset(&unblocked_sigs);
528 sigemptyset(&allowdebug_blocked_sigs);
529 sigaddset(&unblocked_sigs, SIGILL);
530 sigaddset(&unblocked_sigs, SIGSEGV);
531 sigaddset(&unblocked_sigs, SIGBUS);
532 sigaddset(&unblocked_sigs, SIGFPE);
533 sigaddset(&unblocked_sigs, SR_signum);
534
535 if (!ReduceSignalUsage) {
536 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
537 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
538 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
539 }
540 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
541 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
542 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
543 }
544 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
545 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
546 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
547 }
548 }
549 // Fill in signals that are blocked by all but the VM thread.
550 sigemptyset(&vm_sigs);
551 if (!ReduceSignalUsage)
552 sigaddset(&vm_sigs, BREAK_SIGNAL);
553 debug_only(signal_sets_initialized = true);
554
555 }
556
557 // These are signals that are unblocked while a thread is running Java.
558 // (For some reason, they get blocked by default.)
559 sigset_t* os::Linux::unblocked_signals() {
560 assert(signal_sets_initialized, "Not initialized");
561 return &unblocked_sigs;
562 }
563
564 // These are the signals that are blocked while a (non-VM) thread is
565 // running Java. Only the VM thread handles these signals.
566 sigset_t* os::Linux::vm_signals() {
567 assert(signal_sets_initialized, "Not initialized");
568 return &vm_sigs;
569 }
570
571 // These are signals that are blocked during cond_wait to allow debugger in
572 sigset_t* os::Linux::allowdebug_blocked_signals() {
573 assert(signal_sets_initialized, "Not initialized");
574 return &allowdebug_blocked_sigs;
575 }
576
577 void os::Linux::hotspot_sigmask(Thread* thread) {
578
579 //Save caller's signal mask before setting VM signal mask
580 sigset_t caller_sigmask;
581 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
582
583 OSThread* osthread = thread->osthread();
584 osthread->set_caller_sigmask(caller_sigmask);
585
586 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
587
588 if (!ReduceSignalUsage) {
589 if (thread->is_VM_thread()) {
590 // Only the VM thread handles BREAK_SIGNAL ...
591 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
592 } else {
593 // ... all other threads block BREAK_SIGNAL
594 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
595 }
596 }
597 }
598
599 //////////////////////////////////////////////////////////////////////////////
600 // detecting pthread library
601
602 void os::Linux::libpthread_init() {
603 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
604 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
605 // generic name for earlier versions.
606 // Define macros here so we can build HotSpot on old systems.
607 # ifndef _CS_GNU_LIBC_VERSION
608 # define _CS_GNU_LIBC_VERSION 2
609 # endif
610 # ifndef _CS_GNU_LIBPTHREAD_VERSION
611 # define _CS_GNU_LIBPTHREAD_VERSION 3
612 # endif
613
614 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
615 if (n > 0) {
616 char *str = (char *)malloc(n, mtInternal);
617 confstr(_CS_GNU_LIBC_VERSION, str, n);
618 os::Linux::set_glibc_version(str);
619 } else {
620 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
621 static char _gnu_libc_version[32];
622 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
623 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
624 os::Linux::set_glibc_version(_gnu_libc_version);
625 }
626
627 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
628 if (n > 0) {
629 char *str = (char *)malloc(n, mtInternal);
630 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
631 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
632 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
633 // is the case. LinuxThreads has a hard limit on max number of threads.
634 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
635 // On the other hand, NPTL does not have such a limit, sysconf()
636 // will return -1 and errno is not changed. Check if it is really NPTL.
637 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
638 strstr(str, "NPTL") &&
639 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
640 free(str);
641 os::Linux::set_libpthread_version("linuxthreads");
642 } else {
643 os::Linux::set_libpthread_version(str);
644 }
645 } else {
646 // glibc before 2.3.2 only has LinuxThreads.
647 os::Linux::set_libpthread_version("linuxthreads");
648 }
649
650 if (strstr(libpthread_version(), "NPTL")) {
651 os::Linux::set_is_NPTL();
652 } else {
653 os::Linux::set_is_LinuxThreads();
654 }
655
656 // LinuxThreads have two flavors: floating-stack mode, which allows variable
657 // stack size; and fixed-stack mode. NPTL is always floating-stack.
658 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
659 os::Linux::set_is_floating_stack();
660 }
661 }
662
663 /////////////////////////////////////////////////////////////////////////////
664 // thread stack
665
666 // Force Linux kernel to expand current thread stack. If "bottom" is close
667 // to the stack guard, caller should block all signals.
668 //
669 // MAP_GROWSDOWN:
670 // A special mmap() flag that is used to implement thread stacks. It tells
671 // kernel that the memory region should extend downwards when needed. This
672 // allows early versions of LinuxThreads to only mmap the first few pages
673 // when creating a new thread. Linux kernel will automatically expand thread
674 // stack as needed (on page faults).
675 //
676 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
677 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
678 // region, it's hard to tell if the fault is due to a legitimate stack
679 // access or because of reading/writing non-exist memory (e.g. buffer
680 // overrun). As a rule, if the fault happens below current stack pointer,
681 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
682 // application (see Linux kernel fault.c).
683 //
684 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
685 // stack overflow detection.
686 //
687 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
688 // not use this flag. However, the stack of initial thread is not created
689 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
690 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
691 // and then attach the thread to JVM.
692 //
693 // To get around the problem and allow stack banging on Linux, we need to
694 // manually expand thread stack after receiving the SIGSEGV.
695 //
696 // There are two ways to expand thread stack to address "bottom", we used
697 // both of them in JVM before 1.5:
698 // 1. adjust stack pointer first so that it is below "bottom", and then
699 // touch "bottom"
700 // 2. mmap() the page in question
701 //
702 // Now alternate signal stack is gone, it's harder to use 2. For instance,
703 // if current sp is already near the lower end of page 101, and we need to
704 // call mmap() to map page 100, it is possible that part of the mmap() frame
705 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
706 // That will destroy the mmap() frame and cause VM to crash.
707 //
708 // The following code works by adjusting sp first, then accessing the "bottom"
709 // page to force a page fault. Linux kernel will then automatically expand the
710 // stack mapping.
711 //
712 // _expand_stack_to() assumes its frame size is less than page size, which
713 // should always be true if the function is not inlined.
714
715 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
716 #define NOINLINE
717 #else
718 #define NOINLINE __attribute__ ((noinline))
719 #endif
720
721 static void _expand_stack_to(address bottom) NOINLINE;
722
723 static void _expand_stack_to(address bottom) {
724 address sp;
725 size_t size;
726 volatile char *p;
727
728 // Adjust bottom to point to the largest address within the same page, it
729 // gives us a one-page buffer if alloca() allocates slightly more memory.
730 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
731 bottom += os::Linux::page_size() - 1;
732
733 // sp might be slightly above current stack pointer; if that's the case, we
734 // will alloca() a little more space than necessary, which is OK. Don't use
735 // os::current_stack_pointer(), as its result can be slightly below current
736 // stack pointer, causing us to not alloca enough to reach "bottom".
737 sp = (address)&sp;
738
739 if (sp > bottom) {
740 size = sp - bottom;
741 p = (volatile char *)alloca(size);
742 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
743 p[0] = '\0';
744 }
745 }
746
747 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
748 assert(t!=NULL, "just checking");
749 assert(t->osthread()->expanding_stack(), "expand should be set");
750 assert(t->stack_base() != NULL, "stack_base was not initialized");
751
752 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
753 sigset_t mask_all, old_sigset;
754 sigfillset(&mask_all);
755 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
756 _expand_stack_to(addr);
757 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
758 return true;
759 }
760 return false;
761 }
762
763 //////////////////////////////////////////////////////////////////////////////
764 // create new thread
765
766 static address highest_vm_reserved_address();
767
768 // check if it's safe to start a new thread
769 static bool _thread_safety_check(Thread* thread) {
770 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
771 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
772 // Heap is mmap'ed at lower end of memory space. Thread stacks are
773 // allocated (MAP_FIXED) from high address space. Every thread stack
774 // occupies a fixed size slot (usually 2Mbytes, but user can change
775 // it to other values if they rebuild LinuxThreads).
776 //
777 // Problem with MAP_FIXED is that mmap() can still succeed even part of
778 // the memory region has already been mmap'ed. That means if we have too
779 // many threads and/or very large heap, eventually thread stack will
780 // collide with heap.
781 //
782 // Here we try to prevent heap/stack collision by comparing current
783 // stack bottom with the highest address that has been mmap'ed by JVM
784 // plus a safety margin for memory maps created by native code.
785 //
786 // This feature can be disabled by setting ThreadSafetyMargin to 0
787 //
788 if (ThreadSafetyMargin > 0) {
789 address stack_bottom = os::current_stack_base() - os::current_stack_size();
790
791 // not safe if our stack extends below the safety margin
792 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
793 } else {
794 return true;
795 }
796 } else {
797 // Floating stack LinuxThreads or NPTL:
798 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
799 // there's not enough space left, pthread_create() will fail. If we come
800 // here, that means enough space has been reserved for stack.
801 return true;
802 }
803 }
804
805 // Thread start routine for all newly created threads
806 static void *java_start(Thread *thread) {
807 // Try to randomize the cache line index of hot stack frames.
808 // This helps when threads of the same stack traces evict each other's
809 // cache lines. The threads can be either from the same JVM instance, or
810 // from different JVM instances. The benefit is especially true for
811 // processors with hyperthreading technology.
812 static int counter = 0;
813 int pid = os::current_process_id();
814 alloca(((pid ^ counter++) & 7) * 128);
815
816 ThreadLocalStorage::set_thread(thread);
817
818 OSThread* osthread = thread->osthread();
819 Monitor* sync = osthread->startThread_lock();
820
821 // non floating stack LinuxThreads needs extra check, see above
822 if (!_thread_safety_check(thread)) {
823 // notify parent thread
824 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
825 osthread->set_state(ZOMBIE);
826 sync->notify_all();
827 return NULL;
828 }
829
830 // thread_id is kernel thread id (similar to Solaris LWP id)
831 osthread->set_thread_id(os::Linux::gettid());
832
833 if (UseNUMA) {
834 int lgrp_id = os::numa_get_group_id();
835 if (lgrp_id != -1) {
836 thread->set_lgrp_id(lgrp_id);
837 }
838 }
839 // initialize signal mask for this thread
840 os::Linux::hotspot_sigmask(thread);
841
842 // initialize floating point control register
843 os::Linux::init_thread_fpu_state();
844
845 // handshaking with parent thread
846 {
847 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
848
849 // notify parent thread
850 osthread->set_state(INITIALIZED);
851 sync->notify_all();
852
853 // wait until os::start_thread()
854 while (osthread->get_state() == INITIALIZED) {
855 sync->wait(Mutex::_no_safepoint_check_flag);
856 }
857 }
858
859 // call one more level start routine
860 thread->run();
861
862 return 0;
863 }
864
865 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
866 assert(thread->osthread() == NULL, "caller responsible");
867
868 // Allocate the OSThread object
869 OSThread* osthread = new OSThread(NULL, NULL);
870 if (osthread == NULL) {
871 return false;
872 }
873
874 // set the correct thread state
875 osthread->set_thread_type(thr_type);
876
877 // Initial state is ALLOCATED but not INITIALIZED
878 osthread->set_state(ALLOCATED);
879
880 thread->set_osthread(osthread);
881
882 // init thread attributes
883 pthread_attr_t attr;
884 pthread_attr_init(&attr);
885 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
886
887 // stack size
888 if (os::Linux::supports_variable_stack_size()) {
889 // calculate stack size if it's not specified by caller
890 if (stack_size == 0) {
891 stack_size = os::Linux::default_stack_size(thr_type);
892
893 switch (thr_type) {
894 case os::java_thread:
895 // Java threads use ThreadStackSize which default value can be
896 // changed with the flag -Xss
897 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
898 stack_size = JavaThread::stack_size_at_create();
899 break;
900 case os::compiler_thread:
901 if (CompilerThreadStackSize > 0) {
902 stack_size = (size_t)(CompilerThreadStackSize * K);
903 break;
904 } // else fall through:
905 // use VMThreadStackSize if CompilerThreadStackSize is not defined
906 case os::vm_thread:
907 case os::pgc_thread:
908 case os::cgc_thread:
909 case os::watcher_thread:
910 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
911 break;
912 }
913 }
914
915 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
916 pthread_attr_setstacksize(&attr, stack_size);
917 } else {
918 // let pthread_create() pick the default value.
919 }
920
921 // glibc guard page
922 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
923
924 ThreadState state;
925
926 {
927 // Serialize thread creation if we are running with fixed stack LinuxThreads
928 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
929 if (lock) {
930 os::Linux::createThread_lock()->lock_without_safepoint_check();
931 }
932
933 pthread_t tid;
934 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
935
936 pthread_attr_destroy(&attr);
937
938 if (ret != 0) {
939 if (PrintMiscellaneous && (Verbose || WizardMode)) {
940 perror("pthread_create()");
941 }
942 // Need to clean up stuff we've allocated so far
943 thread->set_osthread(NULL);
944 delete osthread;
945 if (lock) os::Linux::createThread_lock()->unlock();
946 return false;
947 }
948
949 // Store pthread info into the OSThread
950 osthread->set_pthread_id(tid);
951
952 // Wait until child thread is either initialized or aborted
953 {
954 Monitor* sync_with_child = osthread->startThread_lock();
955 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
956 while ((state = osthread->get_state()) == ALLOCATED) {
957 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
958 }
959 }
960
961 if (lock) {
962 os::Linux::createThread_lock()->unlock();
963 }
964 }
965
966 // Aborted due to thread limit being reached
967 if (state == ZOMBIE) {
968 thread->set_osthread(NULL);
969 delete osthread;
970 return false;
971 }
972
973 // The thread is returned suspended (in state INITIALIZED),
974 // and is started higher up in the call chain
975 assert(state == INITIALIZED, "race condition");
976 return true;
977 }
978
979 /////////////////////////////////////////////////////////////////////////////
980 // attach existing thread
981
982 // bootstrap the main thread
983 bool os::create_main_thread(JavaThread* thread) {
984 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
985 return create_attached_thread(thread);
986 }
987
988 bool os::create_attached_thread(JavaThread* thread) {
989 #ifdef ASSERT
990 thread->verify_not_published();
991 #endif
992
993 // Allocate the OSThread object
994 OSThread* osthread = new OSThread(NULL, NULL);
995
996 if (osthread == NULL) {
997 return false;
998 }
999
1000 // Store pthread info into the OSThread
1001 osthread->set_thread_id(os::Linux::gettid());
1002 osthread->set_pthread_id(::pthread_self());
1003
1004 // initialize floating point control register
1005 os::Linux::init_thread_fpu_state();
1006
1007 // Initial thread state is RUNNABLE
1008 osthread->set_state(RUNNABLE);
1009
1010 thread->set_osthread(osthread);
1011
1012 if (UseNUMA) {
1013 int lgrp_id = os::numa_get_group_id();
1014 if (lgrp_id != -1) {
1015 thread->set_lgrp_id(lgrp_id);
1016 }
1017 }
1018
1019 if (os::Linux::is_initial_thread()) {
1020 // If current thread is initial thread, its stack is mapped on demand,
1021 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1022 // the entire stack region to avoid SEGV in stack banging.
1023 // It is also useful to get around the heap-stack-gap problem on SuSE
1024 // kernel (see 4821821 for details). We first expand stack to the top
1025 // of yellow zone, then enable stack yellow zone (order is significant,
1026 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1027 // is no gap between the last two virtual memory regions.
1028
1029 JavaThread *jt = (JavaThread *)thread;
1030 address addr = jt->stack_yellow_zone_base();
1031 assert(addr != NULL, "initialization problem?");
1032 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1033
1034 osthread->set_expanding_stack();
1035 os::Linux::manually_expand_stack(jt, addr);
1036 osthread->clear_expanding_stack();
1037 }
1038
1039 // initialize signal mask for this thread
1040 // and save the caller's signal mask
1041 os::Linux::hotspot_sigmask(thread);
1042
1043 return true;
1044 }
1045
1046 void os::pd_start_thread(Thread* thread) {
1047 OSThread * osthread = thread->osthread();
1048 assert(osthread->get_state() != INITIALIZED, "just checking");
1049 Monitor* sync_with_child = osthread->startThread_lock();
1050 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1051 sync_with_child->notify();
1052 }
1053
1054 // Free Linux resources related to the OSThread
1055 void os::free_thread(OSThread* osthread) {
1056 assert(osthread != NULL, "osthread not set");
1057
1058 if (Thread::current()->osthread() == osthread) {
1059 // Restore caller's signal mask
1060 sigset_t sigmask = osthread->caller_sigmask();
1061 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1062 }
1063
1064 delete osthread;
1065 }
1066
1067 //////////////////////////////////////////////////////////////////////////////
1068 // thread local storage
1069
1070 int os::allocate_thread_local_storage() {
1071 pthread_key_t key;
1072 int rslt = pthread_key_create(&key, NULL);
1073 assert(rslt == 0, "cannot allocate thread local storage");
1074 return (int)key;
1075 }
1076
1077 // Note: This is currently not used by VM, as we don't destroy TLS key
1078 // on VM exit.
1079 void os::free_thread_local_storage(int index) {
1080 int rslt = pthread_key_delete((pthread_key_t)index);
1081 assert(rslt == 0, "invalid index");
1082 }
1083
1084 void os::thread_local_storage_at_put(int index, void* value) {
1085 int rslt = pthread_setspecific((pthread_key_t)index, value);
1086 assert(rslt == 0, "pthread_setspecific failed");
1087 }
1088
1089 extern "C" Thread* get_thread() {
1090 return ThreadLocalStorage::thread();
1091 }
1092
1093 //////////////////////////////////////////////////////////////////////////////
1094 // initial thread
1095
1096 // Check if current thread is the initial thread, similar to Solaris thr_main.
1097 bool os::Linux::is_initial_thread(void) {
1098 char dummy;
1099 // If called before init complete, thread stack bottom will be null.
1100 // Can be called if fatal error occurs before initialization.
1101 if (initial_thread_stack_bottom() == NULL) return false;
1102 assert(initial_thread_stack_bottom() != NULL &&
1103 initial_thread_stack_size() != 0,
1104 "os::init did not locate initial thread's stack region");
1105 if ((address)&dummy >= initial_thread_stack_bottom() &&
1106 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1107 return true;
1108 else return false;
1109 }
1110
1111 // Find the virtual memory area that contains addr
1112 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1113 FILE *fp = fopen("/proc/self/maps", "r");
1114 if (fp) {
1115 address low, high;
1116 while (!feof(fp)) {
1117 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1118 if (low <= addr && addr < high) {
1119 if (vma_low) *vma_low = low;
1120 if (vma_high) *vma_high = high;
1121 fclose (fp);
1122 return true;
1123 }
1124 }
1125 for (;;) {
1126 int ch = fgetc(fp);
1127 if (ch == EOF || ch == (int)'\n') break;
1128 }
1129 }
1130 fclose(fp);
1131 }
1132 return false;
1133 }
1134
1135 // Locate initial thread stack. This special handling of initial thread stack
1136 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1137 // bogus value for initial thread.
1138 void os::Linux::capture_initial_stack(size_t max_size) {
1139 // stack size is the easy part, get it from RLIMIT_STACK
1140 size_t stack_size;
1141 struct rlimit rlim;
1142 getrlimit(RLIMIT_STACK, &rlim);
1143 stack_size = rlim.rlim_cur;
1144
1145 // 6308388: a bug in ld.so will relocate its own .data section to the
1146 // lower end of primordial stack; reduce ulimit -s value a little bit
1147 // so we won't install guard page on ld.so's data section.
1148 stack_size -= 2 * page_size();
1149
1150 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1151 // 7.1, in both cases we will get 2G in return value.
1152 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1153 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1154 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1155 // in case other parts in glibc still assumes 2M max stack size.
1156 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1157 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1158 if (stack_size > 2 * K * K IA64_ONLY(*2))
1159 stack_size = 2 * K * K IA64_ONLY(*2);
1160 // Try to figure out where the stack base (top) is. This is harder.
1161 //
1162 // When an application is started, glibc saves the initial stack pointer in
1163 // a global variable "__libc_stack_end", which is then used by system
1164 // libraries. __libc_stack_end should be pretty close to stack top. The
1165 // variable is available since the very early days. However, because it is
1166 // a private interface, it could disappear in the future.
1167 //
1168 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1169 // to __libc_stack_end, it is very close to stack top, but isn't the real
1170 // stack top. Note that /proc may not exist if VM is running as a chroot
1171 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1172 // /proc/<pid>/stat could change in the future (though unlikely).
1173 //
1174 // We try __libc_stack_end first. If that doesn't work, look for
1175 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1176 // as a hint, which should work well in most cases.
1177
1178 uintptr_t stack_start;
1179
1180 // try __libc_stack_end first
1181 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1182 if (p && *p) {
1183 stack_start = *p;
1184 } else {
1185 // see if we can get the start_stack field from /proc/self/stat
1186 FILE *fp;
1187 int pid;
1188 char state;
1189 int ppid;
1190 int pgrp;
1191 int session;
1192 int nr;
1193 int tpgrp;
1194 unsigned long flags;
1195 unsigned long minflt;
1196 unsigned long cminflt;
1197 unsigned long majflt;
1198 unsigned long cmajflt;
1199 unsigned long utime;
1200 unsigned long stime;
1201 long cutime;
1202 long cstime;
1203 long prio;
1204 long nice;
1205 long junk;
1206 long it_real;
1207 uintptr_t start;
1208 uintptr_t vsize;
1209 intptr_t rss;
1210 uintptr_t rsslim;
1211 uintptr_t scodes;
1212 uintptr_t ecode;
1213 int i;
1214
1215 // Figure what the primordial thread stack base is. Code is inspired
1216 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1217 // followed by command name surrounded by parentheses, state, etc.
1218 char stat[2048];
1219 int statlen;
1220
1221 fp = fopen("/proc/self/stat", "r");
1222 if (fp) {
1223 statlen = fread(stat, 1, 2047, fp);
1224 stat[statlen] = '\0';
1225 fclose(fp);
1226
1227 // Skip pid and the command string. Note that we could be dealing with
1228 // weird command names, e.g. user could decide to rename java launcher
1229 // to "java 1.4.2 :)", then the stat file would look like
1230 // 1234 (java 1.4.2 :)) R ... ...
1231 // We don't really need to know the command string, just find the last
1232 // occurrence of ")" and then start parsing from there. See bug 4726580.
1233 char * s = strrchr(stat, ')');
1234
1235 i = 0;
1236 if (s) {
1237 // Skip blank chars
1238 do s++; while (isspace(*s));
1239
1240 #define _UFM UINTX_FORMAT
1241 #define _DFM INTX_FORMAT
1242
1243 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1244 /* 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 */
1245 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1246 &state, /* 3 %c */
1247 &ppid, /* 4 %d */
1248 &pgrp, /* 5 %d */
1249 &session, /* 6 %d */
1250 &nr, /* 7 %d */
1251 &tpgrp, /* 8 %d */
1252 &flags, /* 9 %lu */
1253 &minflt, /* 10 %lu */
1254 &cminflt, /* 11 %lu */
1255 &majflt, /* 12 %lu */
1256 &cmajflt, /* 13 %lu */
1257 &utime, /* 14 %lu */
1258 &stime, /* 15 %lu */
1259 &cutime, /* 16 %ld */
1260 &cstime, /* 17 %ld */
1261 &prio, /* 18 %ld */
1262 &nice, /* 19 %ld */
1263 &junk, /* 20 %ld */
1264 &it_real, /* 21 %ld */
1265 &start, /* 22 UINTX_FORMAT */
1266 &vsize, /* 23 UINTX_FORMAT */
1267 &rss, /* 24 INTX_FORMAT */
1268 &rsslim, /* 25 UINTX_FORMAT */
1269 &scodes, /* 26 UINTX_FORMAT */
1270 &ecode, /* 27 UINTX_FORMAT */
1271 &stack_start); /* 28 UINTX_FORMAT */
1272 }
1273
1274 #undef _UFM
1275 #undef _DFM
1276
1277 if (i != 28 - 2) {
1278 assert(false, "Bad conversion from /proc/self/stat");
1279 // product mode - assume we are the initial thread, good luck in the
1280 // embedded case.
1281 warning("Can't detect initial thread stack location - bad conversion");
1282 stack_start = (uintptr_t) &rlim;
1283 }
1284 } else {
1285 // For some reason we can't open /proc/self/stat (for example, running on
1286 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1287 // most cases, so don't abort:
1288 warning("Can't detect initial thread stack location - no /proc/self/stat");
1289 stack_start = (uintptr_t) &rlim;
1290 }
1291 }
1292
1293 // Now we have a pointer (stack_start) very close to the stack top, the
1294 // next thing to do is to figure out the exact location of stack top. We
1295 // can find out the virtual memory area that contains stack_start by
1296 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1297 // and its upper limit is the real stack top. (again, this would fail if
1298 // running inside chroot, because /proc may not exist.)
1299
1300 uintptr_t stack_top;
1301 address low, high;
1302 if (find_vma((address)stack_start, &low, &high)) {
1303 // success, "high" is the true stack top. (ignore "low", because initial
1304 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1305 stack_top = (uintptr_t)high;
1306 } else {
1307 // failed, likely because /proc/self/maps does not exist
1308 warning("Can't detect initial thread stack location - find_vma failed");
1309 // best effort: stack_start is normally within a few pages below the real
1310 // stack top, use it as stack top, and reduce stack size so we won't put
1311 // guard page outside stack.
1312 stack_top = stack_start;
1313 stack_size -= 16 * page_size();
1314 }
1315
1316 // stack_top could be partially down the page so align it
1317 stack_top = align_size_up(stack_top, page_size());
1318
1319 if (max_size && stack_size > max_size) {
1320 _initial_thread_stack_size = max_size;
1321 } else {
1322 _initial_thread_stack_size = stack_size;
1323 }
1324
1325 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1326 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1327 }
1328
1329 ////////////////////////////////////////////////////////////////////////////////
1330 // time support
1331
1332 // Time since start-up in seconds to a fine granularity.
1333 // Used by VMSelfDestructTimer and the MemProfiler.
1334 double os::elapsedTime() {
1335
1336 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1337 }
1338
1339 jlong os::elapsed_counter() {
1340 return javaTimeNanos() - initial_time_count;
1341 }
1342
1343 jlong os::elapsed_frequency() {
1344 return NANOSECS_PER_SEC; // nanosecond resolution
1345 }
1346
1347 bool os::supports_vtime() { return true; }
1348 bool os::enable_vtime() { return false; }
1349 bool os::vtime_enabled() { return false; }
1350
1351 double os::elapsedVTime() {
1352 struct rusage usage;
1353 int retval = getrusage(RUSAGE_THREAD, &usage);
1354 if (retval == 0) {
1355 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1356 } else {
1357 // better than nothing, but not much
1358 return elapsedTime();
1359 }
1360 }
1361
1362 jlong os::javaTimeMillis() {
1363 timeval time;
1364 int status = gettimeofday(&time, NULL);
1365 assert(status != -1, "linux error");
1366 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1367 }
1368
1369 #ifndef CLOCK_MONOTONIC
1370 #define CLOCK_MONOTONIC (1)
1371 #endif
1372
1373 void os::Linux::clock_init() {
1374 // we do dlopen's in this particular order due to bug in linux
1375 // dynamical loader (see 6348968) leading to crash on exit
1376 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1377 if (handle == NULL) {
1378 handle = dlopen("librt.so", RTLD_LAZY);
1379 }
1380
1381 if (handle) {
1382 int (*clock_getres_func)(clockid_t, struct timespec*) =
1383 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1384 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1385 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1386 if (clock_getres_func && clock_gettime_func) {
1387 // See if monotonic clock is supported by the kernel. Note that some
1388 // early implementations simply return kernel jiffies (updated every
1389 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1390 // for nano time (though the monotonic property is still nice to have).
1391 // It's fixed in newer kernels, however clock_getres() still returns
1392 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1393 // resolution for now. Hopefully as people move to new kernels, this
1394 // won't be a problem.
1395 struct timespec res;
1396 struct timespec tp;
1397 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1398 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1399 // yes, monotonic clock is supported
1400 _clock_gettime = clock_gettime_func;
1401 return;
1402 } else {
1403 // close librt if there is no monotonic clock
1404 dlclose(handle);
1405 }
1406 }
1407 }
1408 warning("No monotonic clock was available - timed services may " \
1409 "be adversely affected if the time-of-day clock changes");
1410 }
1411
1412 #ifndef SYS_clock_getres
1413
1414 #if defined(IA32) || defined(AMD64)
1415 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1416 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1417 #else
1418 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1419 #define sys_clock_getres(x,y) -1
1420 #endif
1421
1422 #else
1423 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1424 #endif
1425
1426 void os::Linux::fast_thread_clock_init() {
1427 if (!UseLinuxPosixThreadCPUClocks) {
1428 return;
1429 }
1430 clockid_t clockid;
1431 struct timespec tp;
1432 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1433 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1434
1435 // Switch to using fast clocks for thread cpu time if
1436 // the sys_clock_getres() returns 0 error code.
1437 // Note, that some kernels may support the current thread
1438 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1439 // returned by the pthread_getcpuclockid().
1440 // If the fast Posix clocks are supported then the sys_clock_getres()
1441 // must return at least tp.tv_sec == 0 which means a resolution
1442 // better than 1 sec. This is extra check for reliability.
1443
1444 if(pthread_getcpuclockid_func &&
1445 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1446 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1447
1448 _supports_fast_thread_cpu_time = true;
1449 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1450 }
1451 }
1452
1453 jlong os::javaTimeNanos() {
1454 if (Linux::supports_monotonic_clock()) {
1455 struct timespec tp;
1456 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1457 assert(status == 0, "gettime error");
1458 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1459 return result;
1460 } else {
1461 timeval time;
1462 int status = gettimeofday(&time, NULL);
1463 assert(status != -1, "linux error");
1464 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1465 return 1000 * usecs;
1466 }
1467 }
1468
1469 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1470 if (Linux::supports_monotonic_clock()) {
1471 info_ptr->max_value = ALL_64_BITS;
1472
1473 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1474 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1475 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1476 } else {
1477 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1478 info_ptr->max_value = ALL_64_BITS;
1479
1480 // gettimeofday is a real time clock so it skips
1481 info_ptr->may_skip_backward = true;
1482 info_ptr->may_skip_forward = true;
1483 }
1484
1485 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1486 }
1487
1488 // Return the real, user, and system times in seconds from an
1489 // arbitrary fixed point in the past.
1490 bool os::getTimesSecs(double* process_real_time,
1491 double* process_user_time,
1492 double* process_system_time) {
1493 struct tms ticks;
1494 clock_t real_ticks = times(&ticks);
1495
1496 if (real_ticks == (clock_t) (-1)) {
1497 return false;
1498 } else {
1499 double ticks_per_second = (double) clock_tics_per_sec;
1500 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1501 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1502 *process_real_time = ((double) real_ticks) / ticks_per_second;
1503
1504 return true;
1505 }
1506 }
1507
1508
1509 char * os::local_time_string(char *buf, size_t buflen) {
1510 struct tm t;
1511 time_t long_time;
1512 time(&long_time);
1513 localtime_r(&long_time, &t);
1514 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1515 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1516 t.tm_hour, t.tm_min, t.tm_sec);
1517 return buf;
1518 }
1519
1520 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1521 return localtime_r(clock, res);
1522 }
1523
1524 ////////////////////////////////////////////////////////////////////////////////
1525 // runtime exit support
1526
1527 // Note: os::shutdown() might be called very early during initialization, or
1528 // called from signal handler. Before adding something to os::shutdown(), make
1529 // sure it is async-safe and can handle partially initialized VM.
1530 void os::shutdown() {
1531
1532 // allow PerfMemory to attempt cleanup of any persistent resources
1533 perfMemory_exit();
1534
1535 // needs to remove object in file system
1536 AttachListener::abort();
1537
1538 // flush buffered output, finish log files
1539 ostream_abort();
1540
1541 // Check for abort hook
1542 abort_hook_t abort_hook = Arguments::abort_hook();
1543 if (abort_hook != NULL) {
1544 abort_hook();
1545 }
1546
1547 }
1548
1549 // Note: os::abort() might be called very early during initialization, or
1550 // called from signal handler. Before adding something to os::abort(), make
1551 // sure it is async-safe and can handle partially initialized VM.
1552 void os::abort(bool dump_core) {
1553 os::shutdown();
1554 if (dump_core) {
1555 #ifndef PRODUCT
1556 fdStream out(defaultStream::output_fd());
1557 out.print_raw("Current thread is ");
1558 char buf[16];
1559 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1560 out.print_raw_cr(buf);
1561 out.print_raw_cr("Dumping core ...");
1562 #endif
1563 ::abort(); // dump core
1564 }
1565
1566 ::exit(1);
1567 }
1568
1569 // Die immediately, no exit hook, no abort hook, no cleanup.
1570 void os::die() {
1571 // _exit() on LinuxThreads only kills current thread
1572 ::abort();
1573 }
1574
1575 // unused on linux for now.
1576 void os::set_error_file(const char *logfile) {}
1577
1578
1579 // This method is a copy of JDK's sysGetLastErrorString
1580 // from src/solaris/hpi/src/system_md.c
1581
1582 size_t os::lasterror(char *buf, size_t len) {
1583
1584 if (errno == 0) return 0;
1585
1586 const char *s = ::strerror(errno);
1587 size_t n = ::strlen(s);
1588 if (n >= len) {
1589 n = len - 1;
1590 }
1591 ::strncpy(buf, s, n);
1592 buf[n] = '\0';
1593 return n;
1594 }
1595
1596 intx os::current_thread_id() { return (intx)pthread_self(); }
1597 int os::current_process_id() {
1598
1599 // Under the old linux thread library, linux gives each thread
1600 // its own process id. Because of this each thread will return
1601 // a different pid if this method were to return the result
1602 // of getpid(2). Linux provides no api that returns the pid
1603 // of the launcher thread for the vm. This implementation
1604 // returns a unique pid, the pid of the launcher thread
1605 // that starts the vm 'process'.
1606
1607 // Under the NPTL, getpid() returns the same pid as the
1608 // launcher thread rather than a unique pid per thread.
1609 // Use gettid() if you want the old pre NPTL behaviour.
1610
1611 // if you are looking for the result of a call to getpid() that
1612 // returns a unique pid for the calling thread, then look at the
1613 // OSThread::thread_id() method in osThread_linux.hpp file
1614
1615 return (int)(_initial_pid ? _initial_pid : getpid());
1616 }
1617
1618 // DLL functions
1619
1620 const char* os::dll_file_extension() { return ".so"; }
1621
1622 // This must be hard coded because it's the system's temporary
1623 // directory not the java application's temp directory, ala java.io.tmpdir.
1624 const char* os::get_temp_directory() { return "/tmp"; }
1625
1626 static bool file_exists(const char* filename) {
1627 struct stat statbuf;
1628 if (filename == NULL || strlen(filename) == 0) {
1629 return false;
1630 }
1631 return os::stat(filename, &statbuf) == 0;
1632 }
1633
1634 bool os::dll_build_name(char* buffer, size_t buflen,
1635 const char* pname, const char* fname) {
1636 bool retval = false;
1637 // Copied from libhpi
1638 const size_t pnamelen = pname ? strlen(pname) : 0;
1639
1640 // Return error on buffer overflow.
1641 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1642 return retval;
1643 }
1644
1645 if (pnamelen == 0) {
1646 snprintf(buffer, buflen, "lib%s.so", fname);
1647 retval = true;
1648 } else if (strchr(pname, *os::path_separator()) != NULL) {
1649 int n;
1650 char** pelements = split_path(pname, &n);
1651 if (pelements == NULL) {
1652 return false;
1653 }
1654 for (int i = 0 ; i < n ; i++) {
1655 // Really shouldn't be NULL, but check can't hurt
1656 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1657 continue; // skip the empty path values
1658 }
1659 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1660 if (file_exists(buffer)) {
1661 retval = true;
1662 break;
1663 }
1664 }
1665 // release the storage
1666 for (int i = 0 ; i < n ; i++) {
1667 if (pelements[i] != NULL) {
1668 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1669 }
1670 }
1671 if (pelements != NULL) {
1672 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1673 }
1674 } else {
1675 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1676 retval = true;
1677 }
1678 return retval;
1679 }
1680
1681 // check if addr is inside libjvm.so
1682 bool os::address_is_in_vm(address addr) {
1683 static address libjvm_base_addr;
1684 Dl_info dlinfo;
1685
1686 if (libjvm_base_addr == NULL) {
1687 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1688 libjvm_base_addr = (address)dlinfo.dli_fbase;
1689 }
1690 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1691 }
1692
1693 if (dladdr((void *)addr, &dlinfo) != 0) {
1694 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1695 }
1696
1697 return false;
1698 }
1699
1700 bool os::dll_address_to_function_name(address addr, char *buf,
1701 int buflen, int *offset) {
1702 // buf is not optional, but offset is optional
1703 assert(buf != NULL, "sanity check");
1704
1705 Dl_info dlinfo;
1706
1707 if (dladdr((void*)addr, &dlinfo) != 0) {
1708 // see if we have a matching symbol
1709 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1710 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1711 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1712 }
1713 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1714 return true;
1715 }
1716 // no matching symbol so try for just file info
1717 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1718 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1719 buf, buflen, offset, dlinfo.dli_fname)) {
1720 return true;
1721 }
1722 }
1723 }
1724
1725 buf[0] = '\0';
1726 if (offset != NULL) *offset = -1;
1727 return false;
1728 }
1729
1730 struct _address_to_library_name {
1731 address addr; // input : memory address
1732 size_t buflen; // size of fname
1733 char* fname; // output: library name
1734 address base; // library base addr
1735 };
1736
1737 static int address_to_library_name_callback(struct dl_phdr_info *info,
1738 size_t size, void *data) {
1739 int i;
1740 bool found = false;
1741 address libbase = NULL;
1742 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1743
1744 // iterate through all loadable segments
1745 for (i = 0; i < info->dlpi_phnum; i++) {
1746 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1747 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1748 // base address of a library is the lowest address of its loaded
1749 // segments.
1750 if (libbase == NULL || libbase > segbase) {
1751 libbase = segbase;
1752 }
1753 // see if 'addr' is within current segment
1754 if (segbase <= d->addr &&
1755 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1756 found = true;
1757 }
1758 }
1759 }
1760
1761 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1762 // so dll_address_to_library_name() can fall through to use dladdr() which
1763 // can figure out executable name from argv[0].
1764 if (found && info->dlpi_name && info->dlpi_name[0]) {
1765 d->base = libbase;
1766 if (d->fname) {
1767 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1768 }
1769 return 1;
1770 }
1771 return 0;
1772 }
1773
1774 bool os::dll_address_to_library_name(address addr, char* buf,
1775 int buflen, int* offset) {
1776 // buf is not optional, but offset is optional
1777 assert(buf != NULL, "sanity check");
1778
1779 Dl_info dlinfo;
1780 struct _address_to_library_name data;
1781
1782 // There is a bug in old glibc dladdr() implementation that it could resolve
1783 // to wrong library name if the .so file has a base address != NULL. Here
1784 // we iterate through the program headers of all loaded libraries to find
1785 // out which library 'addr' really belongs to. This workaround can be
1786 // removed once the minimum requirement for glibc is moved to 2.3.x.
1787 data.addr = addr;
1788 data.fname = buf;
1789 data.buflen = buflen;
1790 data.base = NULL;
1791 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1792
1793 if (rslt) {
1794 // buf already contains library name
1795 if (offset) *offset = addr - data.base;
1796 return true;
1797 }
1798 if (dladdr((void*)addr, &dlinfo) != 0) {
1799 if (dlinfo.dli_fname != NULL) {
1800 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1801 }
1802 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1803 *offset = addr - (address)dlinfo.dli_fbase;
1804 }
1805 return true;
1806 }
1807
1808 buf[0] = '\0';
1809 if (offset) *offset = -1;
1810 return false;
1811 }
1812
1813 // Loads .dll/.so and
1814 // in case of error it checks if .dll/.so was built for the
1815 // same architecture as Hotspot is running on
1816
1817
1818 // Remember the stack's state. The Linux dynamic linker will change
1819 // the stack to 'executable' at most once, so we must safepoint only once.
1820 bool os::Linux::_stack_is_executable = false;
1821
1822 // VM operation that loads a library. This is necessary if stack protection
1823 // of the Java stacks can be lost during loading the library. If we
1824 // do not stop the Java threads, they can stack overflow before the stacks
1825 // are protected again.
1826 class VM_LinuxDllLoad: public VM_Operation {
1827 private:
1828 const char *_filename;
1829 char *_ebuf;
1830 int _ebuflen;
1831 void *_lib;
1832 public:
1833 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1834 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1835 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1836 void doit() {
1837 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1838 os::Linux::_stack_is_executable = true;
1839 }
1840 void* loaded_library() { return _lib; }
1841 };
1842
1843 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1844 {
1845 void * result = NULL;
1846 bool load_attempted = false;
1847
1848 // Check whether the library to load might change execution rights
1849 // of the stack. If they are changed, the protection of the stack
1850 // guard pages will be lost. We need a safepoint to fix this.
1851 //
1852 // See Linux man page execstack(8) for more info.
1853 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1854 ElfFile ef(filename);
1855 if (!ef.specifies_noexecstack()) {
1856 if (!is_init_completed()) {
1857 os::Linux::_stack_is_executable = true;
1858 // This is OK - No Java threads have been created yet, and hence no
1859 // stack guard pages to fix.
1860 //
1861 // This should happen only when you are building JDK7 using a very
1862 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1863 //
1864 // Dynamic loader will make all stacks executable after
1865 // this function returns, and will not do that again.
1866 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1867 } else {
1868 warning("You have loaded library %s which might have disabled stack guard. "
1869 "The VM will try to fix the stack guard now.\n"
1870 "It's highly recommended that you fix the library with "
1871 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1872 filename);
1873
1874 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1875 JavaThread *jt = JavaThread::current();
1876 if (jt->thread_state() != _thread_in_native) {
1877 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1878 // that requires ExecStack. Cannot enter safe point. Let's give up.
1879 warning("Unable to fix stack guard. Giving up.");
1880 } else {
1881 if (!LoadExecStackDllInVMThread) {
1882 // This is for the case where the DLL has an static
1883 // constructor function that executes JNI code. We cannot
1884 // load such DLLs in the VMThread.
1885 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1886 }
1887
1888 ThreadInVMfromNative tiv(jt);
1889 debug_only(VMNativeEntryWrapper vew;)
1890
1891 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1892 VMThread::execute(&op);
1893 if (LoadExecStackDllInVMThread) {
1894 result = op.loaded_library();
1895 }
1896 load_attempted = true;
1897 }
1898 }
1899 }
1900 }
1901
1902 if (!load_attempted) {
1903 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1904 }
1905
1906 if (result != NULL) {
1907 // Successful loading
1908 return result;
1909 }
1910
1911 Elf32_Ehdr elf_head;
1912 int diag_msg_max_length=ebuflen-strlen(ebuf);
1913 char* diag_msg_buf=ebuf+strlen(ebuf);
1914
1915 if (diag_msg_max_length==0) {
1916 // No more space in ebuf for additional diagnostics message
1917 return NULL;
1918 }
1919
1920
1921 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1922
1923 if (file_descriptor < 0) {
1924 // Can't open library, report dlerror() message
1925 return NULL;
1926 }
1927
1928 bool failed_to_read_elf_head=
1929 (sizeof(elf_head)!=
1930 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1931
1932 ::close(file_descriptor);
1933 if (failed_to_read_elf_head) {
1934 // file i/o error - report dlerror() msg
1935 return NULL;
1936 }
1937
1938 typedef struct {
1939 Elf32_Half code; // Actual value as defined in elf.h
1940 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1941 char elf_class; // 32 or 64 bit
1942 char endianess; // MSB or LSB
1943 char* name; // String representation
1944 } arch_t;
1945
1946 #ifndef EM_486
1947 #define EM_486 6 /* Intel 80486 */
1948 #endif
1949
1950 static const arch_t arch_array[]={
1951 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1952 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1953 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1954 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1955 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1956 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1957 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1958 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1959 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1960 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1961 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1962 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1963 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1964 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1965 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1966 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1967 };
1968
1969 #if (defined IA32)
1970 static Elf32_Half running_arch_code=EM_386;
1971 #elif (defined AMD64)
1972 static Elf32_Half running_arch_code=EM_X86_64;
1973 #elif (defined IA64)
1974 static Elf32_Half running_arch_code=EM_IA_64;
1975 #elif (defined __sparc) && (defined _LP64)
1976 static Elf32_Half running_arch_code=EM_SPARCV9;
1977 #elif (defined __sparc) && (!defined _LP64)
1978 static Elf32_Half running_arch_code=EM_SPARC;
1979 #elif (defined __powerpc64__)
1980 static Elf32_Half running_arch_code=EM_PPC64;
1981 #elif (defined __powerpc__)
1982 static Elf32_Half running_arch_code=EM_PPC;
1983 #elif (defined ARM)
1984 static Elf32_Half running_arch_code=EM_ARM;
1985 #elif (defined S390)
1986 static Elf32_Half running_arch_code=EM_S390;
1987 #elif (defined ALPHA)
1988 static Elf32_Half running_arch_code=EM_ALPHA;
1989 #elif (defined MIPSEL)
1990 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1991 #elif (defined PARISC)
1992 static Elf32_Half running_arch_code=EM_PARISC;
1993 #elif (defined MIPS)
1994 static Elf32_Half running_arch_code=EM_MIPS;
1995 #elif (defined M68K)
1996 static Elf32_Half running_arch_code=EM_68K;
1997 #else
1998 #error Method os::dll_load requires that one of following is defined:\
1999 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2000 #endif
2001
2002 // Identify compatability class for VM's architecture and library's architecture
2003 // Obtain string descriptions for architectures
2004
2005 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2006 int running_arch_index=-1;
2007
2008 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2009 if (running_arch_code == arch_array[i].code) {
2010 running_arch_index = i;
2011 }
2012 if (lib_arch.code == arch_array[i].code) {
2013 lib_arch.compat_class = arch_array[i].compat_class;
2014 lib_arch.name = arch_array[i].name;
2015 }
2016 }
2017
2018 assert(running_arch_index != -1,
2019 "Didn't find running architecture code (running_arch_code) in arch_array");
2020 if (running_arch_index == -1) {
2021 // Even though running architecture detection failed
2022 // we may still continue with reporting dlerror() message
2023 return NULL;
2024 }
2025
2026 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2027 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2028 return NULL;
2029 }
2030
2031 #ifndef S390
2032 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2033 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2034 return NULL;
2035 }
2036 #endif // !S390
2037
2038 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2039 if ( lib_arch.name!=NULL ) {
2040 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2041 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2042 lib_arch.name, arch_array[running_arch_index].name);
2043 } else {
2044 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2045 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2046 lib_arch.code,
2047 arch_array[running_arch_index].name);
2048 }
2049 }
2050
2051 return NULL;
2052 }
2053
2054 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2055 void * result = ::dlopen(filename, RTLD_LAZY);
2056 if (result == NULL) {
2057 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2058 ebuf[ebuflen-1] = '\0';
2059 }
2060 return result;
2061 }
2062
2063 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2064 void * result = NULL;
2065 if (LoadExecStackDllInVMThread) {
2066 result = dlopen_helper(filename, ebuf, ebuflen);
2067 }
2068
2069 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2070 // library that requires an executable stack, or which does not have this
2071 // stack attribute set, dlopen changes the stack attribute to executable. The
2072 // read protection of the guard pages gets lost.
2073 //
2074 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2075 // may have been queued at the same time.
2076
2077 if (!_stack_is_executable) {
2078 JavaThread *jt = Threads::first();
2079
2080 while (jt) {
2081 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2082 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2083 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2084 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2085 warning("Attempt to reguard stack yellow zone failed.");
2086 }
2087 }
2088 jt = jt->next();
2089 }
2090 }
2091
2092 return result;
2093 }
2094
2095 /*
2096 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2097 * chances are you might want to run the generated bits against glibc-2.0
2098 * libdl.so, so always use locking for any version of glibc.
2099 */
2100 void* os::dll_lookup(void* handle, const char* name) {
2101 pthread_mutex_lock(&dl_mutex);
2102 void* res = dlsym(handle, name);
2103 pthread_mutex_unlock(&dl_mutex);
2104 return res;
2105 }
2106
2107 void* os::get_default_process_handle() {
2108 return (void*)::dlopen(NULL, RTLD_LAZY);
2109 }
2110
2111 static bool _print_ascii_file(const char* filename, outputStream* st) {
2112 int fd = ::open(filename, O_RDONLY);
2113 if (fd == -1) {
2114 return false;
2115 }
2116
2117 char buf[32];
2118 int bytes;
2119 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2120 st->print_raw(buf, bytes);
2121 }
2122
2123 ::close(fd);
2124
2125 return true;
2126 }
2127
2128 void os::print_dll_info(outputStream *st) {
2129 st->print_cr("Dynamic libraries:");
2130
2131 char fname[32];
2132 pid_t pid = os::Linux::gettid();
2133
2134 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2135
2136 if (!_print_ascii_file(fname, st)) {
2137 st->print("Can not get library information for pid = %d\n", pid);
2138 }
2139 }
2140
2141 void os::print_os_info_brief(outputStream* st) {
2142 os::Linux::print_distro_info(st);
2143
2144 os::Posix::print_uname_info(st);
2145
2146 os::Linux::print_libversion_info(st);
2147
2148 }
2149
2150 void os::print_os_info(outputStream* st) {
2151 st->print("OS:");
2152
2153 os::Linux::print_distro_info(st);
2154
2155 os::Posix::print_uname_info(st);
2156
2157 // Print warning if unsafe chroot environment detected
2158 if (unsafe_chroot_detected) {
2159 st->print("WARNING!! ");
2160 st->print_cr(unstable_chroot_error);
2161 }
2162
2163 os::Linux::print_libversion_info(st);
2164
2165 os::Posix::print_rlimit_info(st);
2166
2167 os::Posix::print_load_average(st);
2168
2169 os::Linux::print_full_memory_info(st);
2170 }
2171
2172 // Try to identify popular distros.
2173 // Most Linux distributions have a /etc/XXX-release file, which contains
2174 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2175 // file that also contains the OS version string. Some have more than one
2176 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2177 // /etc/redhat-release.), so the order is important.
2178 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2179 // their own specific XXX-release file as well as a redhat-release file.
2180 // Because of this the XXX-release file needs to be searched for before the
2181 // redhat-release file.
2182 // Since Red Hat has a lsb-release file that is not very descriptive the
2183 // search for redhat-release needs to be before lsb-release.
2184 // Since the lsb-release file is the new standard it needs to be searched
2185 // before the older style release files.
2186 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2187 // next to last resort. The os-release file is a new standard that contains
2188 // distribution information and the system-release file seems to be an old
2189 // standard that has been replaced by the lsb-release and os-release files.
2190 // Searching for the debian_version file is the last resort. It contains
2191 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2192 // "Debian " is printed before the contents of the debian_version file.
2193 void os::Linux::print_distro_info(outputStream* st) {
2194 if (!_print_ascii_file("/etc/oracle-release", st) &&
2195 !_print_ascii_file("/etc/mandriva-release", st) &&
2196 !_print_ascii_file("/etc/mandrake-release", st) &&
2197 !_print_ascii_file("/etc/sun-release", st) &&
2198 !_print_ascii_file("/etc/redhat-release", st) &&
2199 !_print_ascii_file("/etc/lsb-release", st) &&
2200 !_print_ascii_file("/etc/SuSE-release", st) &&
2201 !_print_ascii_file("/etc/turbolinux-release", st) &&
2202 !_print_ascii_file("/etc/gentoo-release", st) &&
2203 !_print_ascii_file("/etc/ltib-release", st) &&
2204 !_print_ascii_file("/etc/angstrom-version", st) &&
2205 !_print_ascii_file("/etc/system-release", st) &&
2206 !_print_ascii_file("/etc/os-release", st)) {
2207
2208 if (file_exists("/etc/debian_version")) {
2209 st->print("Debian ");
2210 _print_ascii_file("/etc/debian_version", st);
2211 } else {
2212 st->print("Linux");
2213 }
2214 }
2215 st->cr();
2216 }
2217
2218 void os::Linux::print_libversion_info(outputStream* st) {
2219 // libc, pthread
2220 st->print("libc:");
2221 st->print(os::Linux::glibc_version()); st->print(" ");
2222 st->print(os::Linux::libpthread_version()); st->print(" ");
2223 if (os::Linux::is_LinuxThreads()) {
2224 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2225 }
2226 st->cr();
2227 }
2228
2229 void os::Linux::print_full_memory_info(outputStream* st) {
2230 st->print("\n/proc/meminfo:\n");
2231 _print_ascii_file("/proc/meminfo", st);
2232 st->cr();
2233 }
2234
2235 void os::print_memory_info(outputStream* st) {
2236
2237 st->print("Memory:");
2238 st->print(" %dk page", os::vm_page_size()>>10);
2239
2240 // values in struct sysinfo are "unsigned long"
2241 struct sysinfo si;
2242 sysinfo(&si);
2243
2244 st->print(", physical " UINT64_FORMAT "k",
2245 os::physical_memory() >> 10);
2246 st->print("(" UINT64_FORMAT "k free)",
2247 os::available_memory() >> 10);
2248 st->print(", swap " UINT64_FORMAT "k",
2249 ((jlong)si.totalswap * si.mem_unit) >> 10);
2250 st->print("(" UINT64_FORMAT "k free)",
2251 ((jlong)si.freeswap * si.mem_unit) >> 10);
2252 st->cr();
2253 }
2254
2255 void os::pd_print_cpu_info(outputStream* st) {
2256 st->print("\n/proc/cpuinfo:\n");
2257 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2258 st->print(" <Not Available>");
2259 }
2260 st->cr();
2261 }
2262
2263 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2264 // but they're the same for all the linux arch that we support
2265 // and they're the same for solaris but there's no common place to put this.
2266 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2267 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2268 "ILL_COPROC", "ILL_BADSTK" };
2269
2270 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2271 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2272 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2273
2274 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2275
2276 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2277
2278 void os::print_siginfo(outputStream* st, void* siginfo) {
2279 st->print("siginfo:");
2280
2281 const int buflen = 100;
2282 char buf[buflen];
2283 siginfo_t *si = (siginfo_t*)siginfo;
2284 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2285 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2286 st->print("si_errno=%s", buf);
2287 } else {
2288 st->print("si_errno=%d", si->si_errno);
2289 }
2290 const int c = si->si_code;
2291 assert(c > 0, "unexpected si_code");
2292 switch (si->si_signo) {
2293 case SIGILL:
2294 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2295 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2296 break;
2297 case SIGFPE:
2298 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2299 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2300 break;
2301 case SIGSEGV:
2302 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2303 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2304 break;
2305 case SIGBUS:
2306 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2307 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2308 break;
2309 default:
2310 st->print(", si_code=%d", si->si_code);
2311 // no si_addr
2312 }
2313
2314 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2315 UseSharedSpaces) {
2316 FileMapInfo* mapinfo = FileMapInfo::current_info();
2317 if (mapinfo->is_in_shared_space(si->si_addr)) {
2318 st->print("\n\nError accessing class data sharing archive." \
2319 " Mapped file inaccessible during execution, " \
2320 " possible disk/network problem.");
2321 }
2322 }
2323 st->cr();
2324 }
2325
2326
2327 static void print_signal_handler(outputStream* st, int sig,
2328 char* buf, size_t buflen);
2329
2330 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2331 st->print_cr("Signal Handlers:");
2332 print_signal_handler(st, SIGSEGV, buf, buflen);
2333 print_signal_handler(st, SIGBUS , buf, buflen);
2334 print_signal_handler(st, SIGFPE , buf, buflen);
2335 print_signal_handler(st, SIGPIPE, buf, buflen);
2336 print_signal_handler(st, SIGXFSZ, buf, buflen);
2337 print_signal_handler(st, SIGILL , buf, buflen);
2338 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2339 print_signal_handler(st, SR_signum, buf, buflen);
2340 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2341 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2342 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2343 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2344 }
2345
2346 static char saved_jvm_path[MAXPATHLEN] = {0};
2347
2348 // Find the full path to the current module, libjvm.so
2349 void os::jvm_path(char *buf, jint buflen) {
2350 // Error checking.
2351 if (buflen < MAXPATHLEN) {
2352 assert(false, "must use a large-enough buffer");
2353 buf[0] = '\0';
2354 return;
2355 }
2356 // Lazy resolve the path to current module.
2357 if (saved_jvm_path[0] != 0) {
2358 strcpy(buf, saved_jvm_path);
2359 return;
2360 }
2361
2362 char dli_fname[MAXPATHLEN];
2363 bool ret = dll_address_to_library_name(
2364 CAST_FROM_FN_PTR(address, os::jvm_path),
2365 dli_fname, sizeof(dli_fname), NULL);
2366 assert(ret, "cannot locate libjvm");
2367 char *rp = NULL;
2368 if (ret && dli_fname[0] != '\0') {
2369 rp = realpath(dli_fname, buf);
2370 }
2371 if (rp == NULL)
2372 return;
2373
2374 if (Arguments::created_by_gamma_launcher()) {
2375 // Support for the gamma launcher. Typical value for buf is
2376 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2377 // the right place in the string, then assume we are installed in a JDK and
2378 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2379 // up the path so it looks like libjvm.so is installed there (append a
2380 // fake suffix hotspot/libjvm.so).
2381 const char *p = buf + strlen(buf) - 1;
2382 for (int count = 0; p > buf && count < 5; ++count) {
2383 for (--p; p > buf && *p != '/'; --p)
2384 /* empty */ ;
2385 }
2386
2387 if (strncmp(p, "/jre/lib/", 9) != 0) {
2388 // Look for JAVA_HOME in the environment.
2389 char* java_home_var = ::getenv("JAVA_HOME");
2390 if (java_home_var != NULL && java_home_var[0] != 0) {
2391 char* jrelib_p;
2392 int len;
2393
2394 // Check the current module name "libjvm.so".
2395 p = strrchr(buf, '/');
2396 assert(strstr(p, "/libjvm") == p, "invalid library name");
2397
2398 rp = realpath(java_home_var, buf);
2399 if (rp == NULL)
2400 return;
2401
2402 // determine if this is a legacy image or modules image
2403 // modules image doesn't have "jre" subdirectory
2404 len = strlen(buf);
2405 jrelib_p = buf + len;
2406 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2407 if (0 != access(buf, F_OK)) {
2408 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2409 }
2410
2411 if (0 == access(buf, F_OK)) {
2412 // Use current module name "libjvm.so"
2413 len = strlen(buf);
2414 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2415 } else {
2416 // Go back to path of .so
2417 rp = realpath(dli_fname, buf);
2418 if (rp == NULL)
2419 return;
2420 }
2421 }
2422 }
2423 }
2424
2425 strcpy(saved_jvm_path, buf);
2426 }
2427
2428 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2429 // no prefix required, not even "_"
2430 }
2431
2432 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2433 // no suffix required
2434 }
2435
2436 ////////////////////////////////////////////////////////////////////////////////
2437 // sun.misc.Signal support
2438
2439 static volatile jint sigint_count = 0;
2440
2441 static void
2442 UserHandler(int sig, void *siginfo, void *context) {
2443 // 4511530 - sem_post is serialized and handled by the manager thread. When
2444 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2445 // don't want to flood the manager thread with sem_post requests.
2446 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2447 return;
2448
2449 // Ctrl-C is pressed during error reporting, likely because the error
2450 // handler fails to abort. Let VM die immediately.
2451 if (sig == SIGINT && is_error_reported()) {
2452 os::die();
2453 }
2454
2455 os::signal_notify(sig);
2456 }
2457
2458 void* os::user_handler() {
2459 return CAST_FROM_FN_PTR(void*, UserHandler);
2460 }
2461
2462 class Semaphore : public StackObj {
2463 public:
2464 Semaphore();
2465 ~Semaphore();
2466 void signal();
2467 void wait();
2468 bool trywait();
2469 bool timedwait(unsigned int sec, int nsec);
2470 private:
2471 sem_t _semaphore;
2472 };
2473
2474
2475 Semaphore::Semaphore() {
2476 sem_init(&_semaphore, 0, 0);
2477 }
2478
2479 Semaphore::~Semaphore() {
2480 sem_destroy(&_semaphore);
2481 }
2482
2483 void Semaphore::signal() {
2484 sem_post(&_semaphore);
2485 }
2486
2487 void Semaphore::wait() {
2488 sem_wait(&_semaphore);
2489 }
2490
2491 bool Semaphore::trywait() {
2492 return sem_trywait(&_semaphore) == 0;
2493 }
2494
2495 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2496 struct timespec ts;
2497 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2498
2499 while (1) {
2500 int result = sem_timedwait(&_semaphore, &ts);
2501 if (result == 0) {
2502 return true;
2503 } else if (errno == EINTR) {
2504 continue;
2505 } else if (errno == ETIMEDOUT) {
2506 return false;
2507 } else {
2508 return false;
2509 }
2510 }
2511 }
2512
2513 extern "C" {
2514 typedef void (*sa_handler_t)(int);
2515 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2516 }
2517
2518 void* os::signal(int signal_number, void* handler) {
2519 struct sigaction sigAct, oldSigAct;
2520
2521 sigfillset(&(sigAct.sa_mask));
2522 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2523 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2524
2525 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2526 // -1 means registration failed
2527 return (void *)-1;
2528 }
2529
2530 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2531 }
2532
2533 void os::signal_raise(int signal_number) {
2534 ::raise(signal_number);
2535 }
2536
2537 /*
2538 * The following code is moved from os.cpp for making this
2539 * code platform specific, which it is by its very nature.
2540 */
2541
2542 // Will be modified when max signal is changed to be dynamic
2543 int os::sigexitnum_pd() {
2544 return NSIG;
2545 }
2546
2547 // a counter for each possible signal value
2548 static volatile jint pending_signals[NSIG+1] = { 0 };
2549
2550 // Linux(POSIX) specific hand shaking semaphore.
2551 static sem_t sig_sem;
2552 static Semaphore sr_semaphore;
2553
2554 void os::signal_init_pd() {
2555 // Initialize signal structures
2556 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2557
2558 // Initialize signal semaphore
2559 ::sem_init(&sig_sem, 0, 0);
2560 }
2561
2562 void os::signal_notify(int sig) {
2563 Atomic::inc(&pending_signals[sig]);
2564 ::sem_post(&sig_sem);
2565 }
2566
2567 static int check_pending_signals(bool wait) {
2568 Atomic::store(0, &sigint_count);
2569 for (;;) {
2570 for (int i = 0; i < NSIG + 1; i++) {
2571 jint n = pending_signals[i];
2572 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2573 return i;
2574 }
2575 }
2576 if (!wait) {
2577 return -1;
2578 }
2579 JavaThread *thread = JavaThread::current();
2580 ThreadBlockInVM tbivm(thread);
2581
2582 bool threadIsSuspended;
2583 do {
2584 thread->set_suspend_equivalent();
2585 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2586 ::sem_wait(&sig_sem);
2587
2588 // were we externally suspended while we were waiting?
2589 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2590 if (threadIsSuspended) {
2591 //
2592 // The semaphore has been incremented, but while we were waiting
2593 // another thread suspended us. We don't want to continue running
2594 // while suspended because that would surprise the thread that
2595 // suspended us.
2596 //
2597 ::sem_post(&sig_sem);
2598
2599 thread->java_suspend_self();
2600 }
2601 } while (threadIsSuspended);
2602 }
2603 }
2604
2605 int os::signal_lookup() {
2606 return check_pending_signals(false);
2607 }
2608
2609 int os::signal_wait() {
2610 return check_pending_signals(true);
2611 }
2612
2613 ////////////////////////////////////////////////////////////////////////////////
2614 // Virtual Memory
2615
2616 int os::vm_page_size() {
2617 // Seems redundant as all get out
2618 assert(os::Linux::page_size() != -1, "must call os::init");
2619 return os::Linux::page_size();
2620 }
2621
2622 // Solaris allocates memory by pages.
2623 int os::vm_allocation_granularity() {
2624 assert(os::Linux::page_size() != -1, "must call os::init");
2625 return os::Linux::page_size();
2626 }
2627
2628 // Rationale behind this function:
2629 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2630 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2631 // samples for JITted code. Here we create private executable mapping over the code cache
2632 // and then we can use standard (well, almost, as mapping can change) way to provide
2633 // info for the reporting script by storing timestamp and location of symbol
2634 void linux_wrap_code(char* base, size_t size) {
2635 static volatile jint cnt = 0;
2636
2637 if (!UseOprofile) {
2638 return;
2639 }
2640
2641 char buf[PATH_MAX+1];
2642 int num = Atomic::add(1, &cnt);
2643
2644 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2645 os::get_temp_directory(), os::current_process_id(), num);
2646 unlink(buf);
2647
2648 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2649
2650 if (fd != -1) {
2651 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2652 if (rv != (off_t)-1) {
2653 if (::write(fd, "", 1) == 1) {
2654 mmap(base, size,
2655 PROT_READ|PROT_WRITE|PROT_EXEC,
2656 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2657 }
2658 }
2659 ::close(fd);
2660 unlink(buf);
2661 }
2662 }
2663
2664 static bool recoverable_mmap_error(int err) {
2665 // See if the error is one we can let the caller handle. This
2666 // list of errno values comes from JBS-6843484. I can't find a
2667 // Linux man page that documents this specific set of errno
2668 // values so while this list currently matches Solaris, it may
2669 // change as we gain experience with this failure mode.
2670 switch (err) {
2671 case EBADF:
2672 case EINVAL:
2673 case ENOTSUP:
2674 // let the caller deal with these errors
2675 return true;
2676
2677 default:
2678 // Any remaining errors on this OS can cause our reserved mapping
2679 // to be lost. That can cause confusion where different data
2680 // structures think they have the same memory mapped. The worst
2681 // scenario is if both the VM and a library think they have the
2682 // same memory mapped.
2683 return false;
2684 }
2685 }
2686
2687 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2688 int err) {
2689 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2690 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2691 strerror(err), err);
2692 }
2693
2694 static void warn_fail_commit_memory(char* addr, size_t size,
2695 size_t alignment_hint, bool exec,
2696 int err) {
2697 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2698 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2699 alignment_hint, exec, strerror(err), err);
2700 }
2701
2702 // NOTE: Linux kernel does not really reserve the pages for us.
2703 // All it does is to check if there are enough free pages
2704 // left at the time of mmap(). This could be a potential
2705 // problem.
2706 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2707 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2708 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2709 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2710 if (res != (uintptr_t) MAP_FAILED) {
2711 if (UseNUMAInterleaving) {
2712 numa_make_global(addr, size);
2713 }
2714 return 0;
2715 }
2716
2717 int err = errno; // save errno from mmap() call above
2718
2719 if (!recoverable_mmap_error(err)) {
2720 warn_fail_commit_memory(addr, size, exec, err);
2721 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2722 }
2723
2724 return err;
2725 }
2726
2727 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2728 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2729 }
2730
2731 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2732 const char* mesg) {
2733 assert(mesg != NULL, "mesg must be specified");
2734 int err = os::Linux::commit_memory_impl(addr, size, exec);
2735 if (err != 0) {
2736 // the caller wants all commit errors to exit with the specified mesg:
2737 warn_fail_commit_memory(addr, size, exec, err);
2738 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2739 }
2740 }
2741
2742 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2743 #ifndef MAP_HUGETLB
2744 #define MAP_HUGETLB 0x40000
2745 #endif
2746
2747 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2748 #ifndef MADV_HUGEPAGE
2749 #define MADV_HUGEPAGE 14
2750 #endif
2751
2752 int os::Linux::commit_memory_impl(char* addr, size_t size,
2753 size_t alignment_hint, bool exec) {
2754 int err = os::Linux::commit_memory_impl(addr, size, exec);
2755 if (err == 0) {
2756 realign_memory(addr, size, alignment_hint);
2757 }
2758 return err;
2759 }
2760
2761 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2762 bool exec) {
2763 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2764 }
2765
2766 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2767 size_t alignment_hint, bool exec,
2768 const char* mesg) {
2769 assert(mesg != NULL, "mesg must be specified");
2770 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2771 if (err != 0) {
2772 // the caller wants all commit errors to exit with the specified mesg:
2773 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2774 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2775 }
2776 }
2777
2778 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2779 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2780 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2781 // be supported or the memory may already be backed by huge pages.
2782 ::madvise(addr, bytes, MADV_HUGEPAGE);
2783 }
2784 }
2785
2786 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2787 // This method works by doing an mmap over an existing mmaping and effectively discarding
2788 // the existing pages. However it won't work for SHM-based large pages that cannot be
2789 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2790 // small pages on top of the SHM segment. This method always works for small pages, so we
2791 // allow that in any case.
2792 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2793 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2794 }
2795 }
2796
2797 void os::numa_make_global(char *addr, size_t bytes) {
2798 Linux::numa_interleave_memory(addr, bytes);
2799 }
2800
2801 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2802 // bind policy to MPOL_PREFERRED for the current thread.
2803 #define USE_MPOL_PREFERRED 0
2804
2805 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2806 // To make NUMA and large pages more robust when both enabled, we need to ease
2807 // the requirements on where the memory should be allocated. MPOL_BIND is the
2808 // default policy and it will force memory to be allocated on the specified
2809 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2810 // the specified node, but will not force it. Using this policy will prevent
2811 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2812 // free large pages.
2813 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2814 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2815 }
2816
2817 bool os::numa_topology_changed() { return false; }
2818
2819 size_t os::numa_get_groups_num() {
2820 int max_node = Linux::numa_max_node();
2821 return max_node > 0 ? max_node + 1 : 1;
2822 }
2823
2824 int os::numa_get_group_id() {
2825 int cpu_id = Linux::sched_getcpu();
2826 if (cpu_id != -1) {
2827 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2828 if (lgrp_id != -1) {
2829 return lgrp_id;
2830 }
2831 }
2832 return 0;
2833 }
2834
2835 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2836 for (size_t i = 0; i < size; i++) {
2837 ids[i] = i;
2838 }
2839 return size;
2840 }
2841
2842 bool os::get_page_info(char *start, page_info* info) {
2843 return false;
2844 }
2845
2846 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2847 return end;
2848 }
2849
2850
2851 int os::Linux::sched_getcpu_syscall(void) {
2852 unsigned int cpu;
2853 int retval = -1;
2854
2855 #if defined(IA32)
2856 # ifndef SYS_getcpu
2857 # define SYS_getcpu 318
2858 # endif
2859 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2860 #elif defined(AMD64)
2861 // Unfortunately we have to bring all these macros here from vsyscall.h
2862 // to be able to compile on old linuxes.
2863 # define __NR_vgetcpu 2
2864 # define VSYSCALL_START (-10UL << 20)
2865 # define VSYSCALL_SIZE 1024
2866 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2867 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2868 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2869 retval = vgetcpu(&cpu, NULL, NULL);
2870 #endif
2871
2872 return (retval == -1) ? retval : cpu;
2873 }
2874
2875 // Something to do with the numa-aware allocator needs these symbols
2876 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2877 extern "C" JNIEXPORT void numa_error(char *where) { }
2878 extern "C" JNIEXPORT int fork1() { return fork(); }
2879
2880
2881 // If we are running with libnuma version > 2, then we should
2882 // be trying to use symbols with versions 1.1
2883 // If we are running with earlier version, which did not have symbol versions,
2884 // we should use the base version.
2885 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2886 void *f = dlvsym(handle, name, "libnuma_1.1");
2887 if (f == NULL) {
2888 f = dlsym(handle, name);
2889 }
2890 return f;
2891 }
2892
2893 bool os::Linux::libnuma_init() {
2894 // sched_getcpu() should be in libc.
2895 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2896 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2897
2898 // If it's not, try a direct syscall.
2899 if (sched_getcpu() == -1)
2900 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2901
2902 if (sched_getcpu() != -1) { // Does it work?
2903 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2904 if (handle != NULL) {
2905 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2906 libnuma_dlsym(handle, "numa_node_to_cpus")));
2907 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2908 libnuma_dlsym(handle, "numa_max_node")));
2909 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2910 libnuma_dlsym(handle, "numa_available")));
2911 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2912 libnuma_dlsym(handle, "numa_tonode_memory")));
2913 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2914 libnuma_dlsym(handle, "numa_interleave_memory")));
2915 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2916 libnuma_dlsym(handle, "numa_set_bind_policy")));
2917
2918
2919 if (numa_available() != -1) {
2920 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2921 // Create a cpu -> node mapping
2922 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2923 rebuild_cpu_to_node_map();
2924 return true;
2925 }
2926 }
2927 }
2928 return false;
2929 }
2930
2931 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2932 // The table is later used in get_node_by_cpu().
2933 void os::Linux::rebuild_cpu_to_node_map() {
2934 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2935 // in libnuma (possible values are starting from 16,
2936 // and continuing up with every other power of 2, but less
2937 // than the maximum number of CPUs supported by kernel), and
2938 // is a subject to change (in libnuma version 2 the requirements
2939 // are more reasonable) we'll just hardcode the number they use
2940 // in the library.
2941 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2942
2943 size_t cpu_num = os::active_processor_count();
2944 size_t cpu_map_size = NCPUS / BitsPerCLong;
2945 size_t cpu_map_valid_size =
2946 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2947
2948 cpu_to_node()->clear();
2949 cpu_to_node()->at_grow(cpu_num - 1);
2950 size_t node_num = numa_get_groups_num();
2951
2952 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2953 for (size_t i = 0; i < node_num; i++) {
2954 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2955 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2956 if (cpu_map[j] != 0) {
2957 for (size_t k = 0; k < BitsPerCLong; k++) {
2958 if (cpu_map[j] & (1UL << k)) {
2959 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2960 }
2961 }
2962 }
2963 }
2964 }
2965 }
2966 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2967 }
2968
2969 int os::Linux::get_node_by_cpu(int cpu_id) {
2970 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2971 return cpu_to_node()->at(cpu_id);
2972 }
2973 return -1;
2974 }
2975
2976 GrowableArray<int>* os::Linux::_cpu_to_node;
2977 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2978 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2979 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2980 os::Linux::numa_available_func_t os::Linux::_numa_available;
2981 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2982 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2983 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2984 unsigned long* os::Linux::_numa_all_nodes;
2985
2986 bool os::pd_uncommit_memory(char* addr, size_t size) {
2987 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2988 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2989 return res != (uintptr_t) MAP_FAILED;
2990 }
2991
2992 static
2993 address get_stack_commited_bottom(address bottom, size_t size) {
2994 address nbot = bottom;
2995 address ntop = bottom + size;
2996
2997 size_t page_sz = os::vm_page_size();
2998 unsigned pages = size / page_sz;
2999
3000 unsigned char vec[1];
3001 unsigned imin = 1, imax = pages + 1, imid;
3002 int mincore_return_value;
3003
3004 while (imin < imax) {
3005 imid = (imax + imin) / 2;
3006 nbot = ntop - (imid * page_sz);
3007
3008 // Use a trick with mincore to check whether the page is mapped or not.
3009 // mincore sets vec to 1 if page resides in memory and to 0 if page
3010 // is swapped output but if page we are asking for is unmapped
3011 // it returns -1,ENOMEM
3012 mincore_return_value = mincore(nbot, page_sz, vec);
3013
3014 if (mincore_return_value == -1) {
3015 // Page is not mapped go up
3016 // to find first mapped page
3017 if (errno != EAGAIN) {
3018 assert(errno == ENOMEM, "Unexpected mincore errno");
3019 imax = imid;
3020 }
3021 } else {
3022 // Page is mapped go down
3023 // to find first not mapped page
3024 imin = imid + 1;
3025 }
3026 }
3027
3028 nbot = nbot + page_sz;
3029
3030 // Adjust stack bottom one page up if last checked page is not mapped
3031 if (mincore_return_value == -1) {
3032 nbot = nbot + page_sz;
3033 }
3034
3035 return nbot;
3036 }
3037
3038
3039 // Linux uses a growable mapping for the stack, and if the mapping for
3040 // the stack guard pages is not removed when we detach a thread the
3041 // stack cannot grow beyond the pages where the stack guard was
3042 // mapped. If at some point later in the process the stack expands to
3043 // that point, the Linux kernel cannot expand the stack any further
3044 // because the guard pages are in the way, and a segfault occurs.
3045 //
3046 // However, it's essential not to split the stack region by unmapping
3047 // a region (leaving a hole) that's already part of the stack mapping,
3048 // so if the stack mapping has already grown beyond the guard pages at
3049 // the time we create them, we have to truncate the stack mapping.
3050 // So, we need to know the extent of the stack mapping when
3051 // create_stack_guard_pages() is called.
3052
3053 // We only need this for stacks that are growable: at the time of
3054 // writing thread stacks don't use growable mappings (i.e. those
3055 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3056 // only applies to the main thread.
3057
3058 // If the (growable) stack mapping already extends beyond the point
3059 // where we're going to put our guard pages, truncate the mapping at
3060 // that point by munmap()ping it. This ensures that when we later
3061 // munmap() the guard pages we don't leave a hole in the stack
3062 // mapping. This only affects the main/initial thread
3063
3064 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3065
3066 if (os::Linux::is_initial_thread()) {
3067 // As we manually grow stack up to bottom inside create_attached_thread(),
3068 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3069 // we don't need to do anything special.
3070 // Check it first, before calling heavy function.
3071 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3072 unsigned char vec[1];
3073
3074 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3075 // Fallback to slow path on all errors, including EAGAIN
3076 stack_extent = (uintptr_t) get_stack_commited_bottom(
3077 os::Linux::initial_thread_stack_bottom(),
3078 (size_t)addr - stack_extent);
3079 }
3080
3081 if (stack_extent < (uintptr_t)addr) {
3082 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3083 }
3084 }
3085
3086 return os::commit_memory(addr, size, !ExecMem);
3087 }
3088
3089 // If this is a growable mapping, remove the guard pages entirely by
3090 // munmap()ping them. If not, just call uncommit_memory(). This only
3091 // affects the main/initial thread, but guard against future OS changes
3092 // It's safe to always unmap guard pages for initial thread because we
3093 // always place it right after end of the mapped region
3094
3095 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3096 uintptr_t stack_extent, stack_base;
3097
3098 if (os::Linux::is_initial_thread()) {
3099 return ::munmap(addr, size) == 0;
3100 }
3101
3102 return os::uncommit_memory(addr, size);
3103 }
3104
3105 static address _highest_vm_reserved_address = NULL;
3106
3107 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3108 // at 'requested_addr'. If there are existing memory mappings at the same
3109 // location, however, they will be overwritten. If 'fixed' is false,
3110 // 'requested_addr' is only treated as a hint, the return value may or
3111 // may not start from the requested address. Unlike Linux mmap(), this
3112 // function returns NULL to indicate failure.
3113 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3114 char * addr;
3115 int flags;
3116
3117 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3118 if (fixed) {
3119 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3120 flags |= MAP_FIXED;
3121 }
3122
3123 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3124 // touch an uncommitted page. Otherwise, the read/write might
3125 // succeed if we have enough swap space to back the physical page.
3126 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3127 flags, -1, 0);
3128
3129 if (addr != MAP_FAILED) {
3130 // anon_mmap() should only get called during VM initialization,
3131 // don't need lock (actually we can skip locking even it can be called
3132 // from multiple threads, because _highest_vm_reserved_address is just a
3133 // hint about the upper limit of non-stack memory regions.)
3134 if ((address)addr + bytes > _highest_vm_reserved_address) {
3135 _highest_vm_reserved_address = (address)addr + bytes;
3136 }
3137 }
3138
3139 return addr == MAP_FAILED ? NULL : addr;
3140 }
3141
3142 // Don't update _highest_vm_reserved_address, because there might be memory
3143 // regions above addr + size. If so, releasing a memory region only creates
3144 // a hole in the address space, it doesn't help prevent heap-stack collision.
3145 //
3146 static int anon_munmap(char * addr, size_t size) {
3147 return ::munmap(addr, size) == 0;
3148 }
3149
3150 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3151 size_t alignment_hint) {
3152 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3153 }
3154
3155 bool os::pd_release_memory(char* addr, size_t size) {
3156 return anon_munmap(addr, size);
3157 }
3158
3159 static address highest_vm_reserved_address() {
3160 return _highest_vm_reserved_address;
3161 }
3162
3163 static bool linux_mprotect(char* addr, size_t size, int prot) {
3164 // Linux wants the mprotect address argument to be page aligned.
3165 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3166
3167 // According to SUSv3, mprotect() should only be used with mappings
3168 // established by mmap(), and mmap() always maps whole pages. Unaligned
3169 // 'addr' likely indicates problem in the VM (e.g. trying to change
3170 // protection of malloc'ed or statically allocated memory). Check the
3171 // caller if you hit this assert.
3172 assert(addr == bottom, "sanity check");
3173
3174 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3175 return ::mprotect(bottom, size, prot) == 0;
3176 }
3177
3178 // Set protections specified
3179 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3180 bool is_committed) {
3181 unsigned int p = 0;
3182 switch (prot) {
3183 case MEM_PROT_NONE: p = PROT_NONE; break;
3184 case MEM_PROT_READ: p = PROT_READ; break;
3185 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3186 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3187 default:
3188 ShouldNotReachHere();
3189 }
3190 // is_committed is unused.
3191 return linux_mprotect(addr, bytes, p);
3192 }
3193
3194 bool os::guard_memory(char* addr, size_t size) {
3195 return linux_mprotect(addr, size, PROT_NONE);
3196 }
3197
3198 bool os::unguard_memory(char* addr, size_t size) {
3199 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3200 }
3201
3202 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3203 bool result = false;
3204 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3205 MAP_ANONYMOUS|MAP_PRIVATE,
3206 -1, 0);
3207 if (p != MAP_FAILED) {
3208 void *aligned_p = align_ptr_up(p, page_size);
3209
3210 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3211
3212 munmap(p, page_size * 2);
3213 }
3214
3215 if (warn && !result) {
3216 warning("TransparentHugePages is not supported by the operating system.");
3217 }
3218
3219 return result;
3220 }
3221
3222 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3223 bool result = false;
3224 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3225 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3226 -1, 0);
3227
3228 if (p != MAP_FAILED) {
3229 // We don't know if this really is a huge page or not.
3230 FILE *fp = fopen("/proc/self/maps", "r");
3231 if (fp) {
3232 while (!feof(fp)) {
3233 char chars[257];
3234 long x = 0;
3235 if (fgets(chars, sizeof(chars), fp)) {
3236 if (sscanf(chars, "%lx-%*x", &x) == 1
3237 && x == (long)p) {
3238 if (strstr (chars, "hugepage")) {
3239 result = true;
3240 break;
3241 }
3242 }
3243 }
3244 }
3245 fclose(fp);
3246 }
3247 munmap(p, page_size);
3248 }
3249
3250 if (warn && !result) {
3251 warning("HugeTLBFS is not supported by the operating system.");
3252 }
3253
3254 return result;
3255 }
3256
3257 /*
3258 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3259 *
3260 * From the coredump_filter documentation:
3261 *
3262 * - (bit 0) anonymous private memory
3263 * - (bit 1) anonymous shared memory
3264 * - (bit 2) file-backed private memory
3265 * - (bit 3) file-backed shared memory
3266 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3267 * effective only if the bit 2 is cleared)
3268 * - (bit 5) hugetlb private memory
3269 * - (bit 6) hugetlb shared memory
3270 */
3271 static void set_coredump_filter(void) {
3272 FILE *f;
3273 long cdm;
3274
3275 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3276 return;
3277 }
3278
3279 if (fscanf(f, "%lx", &cdm) != 1) {
3280 fclose(f);
3281 return;
3282 }
3283
3284 rewind(f);
3285
3286 if ((cdm & LARGEPAGES_BIT) == 0) {
3287 cdm |= LARGEPAGES_BIT;
3288 fprintf(f, "%#lx", cdm);
3289 }
3290
3291 fclose(f);
3292 }
3293
3294 // Large page support
3295
3296 static size_t _large_page_size = 0;
3297
3298 size_t os::Linux::find_large_page_size() {
3299 size_t large_page_size = 0;
3300
3301 // large_page_size on Linux is used to round up heap size. x86 uses either
3302 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3303 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3304 // page as large as 256M.
3305 //
3306 // Here we try to figure out page size by parsing /proc/meminfo and looking
3307 // for a line with the following format:
3308 // Hugepagesize: 2048 kB
3309 //
3310 // If we can't determine the value (e.g. /proc is not mounted, or the text
3311 // format has been changed), we'll use the largest page size supported by
3312 // the processor.
3313
3314 #ifndef ZERO
3315 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3316 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3317 #endif // ZERO
3318
3319 FILE *fp = fopen("/proc/meminfo", "r");
3320 if (fp) {
3321 while (!feof(fp)) {
3322 int x = 0;
3323 char buf[16];
3324 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3325 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3326 large_page_size = x * K;
3327 break;
3328 }
3329 } else {
3330 // skip to next line
3331 for (;;) {
3332 int ch = fgetc(fp);
3333 if (ch == EOF || ch == (int)'\n') break;
3334 }
3335 }
3336 }
3337 fclose(fp);
3338 }
3339
3340 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3341 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3342 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3343 proper_unit_for_byte_size(large_page_size));
3344 }
3345
3346 return large_page_size;
3347 }
3348
3349 size_t os::Linux::setup_large_page_size() {
3350 _large_page_size = Linux::find_large_page_size();
3351 const size_t default_page_size = (size_t)Linux::page_size();
3352 if (_large_page_size > default_page_size) {
3353 _page_sizes[0] = _large_page_size;
3354 _page_sizes[1] = default_page_size;
3355 _page_sizes[2] = 0;
3356 }
3357
3358 return _large_page_size;
3359 }
3360
3361 bool os::Linux::setup_large_page_type(size_t page_size) {
3362 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3363 FLAG_IS_DEFAULT(UseSHM) &&
3364 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3365
3366 // The type of large pages has not been specified by the user.
3367
3368 // Try UseHugeTLBFS and then UseSHM.
3369 UseHugeTLBFS = UseSHM = true;
3370
3371 // Don't try UseTransparentHugePages since there are known
3372 // performance issues with it turned on. This might change in the future.
3373 UseTransparentHugePages = false;
3374 }
3375
3376 if (UseTransparentHugePages) {
3377 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3378 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3379 UseHugeTLBFS = false;
3380 UseSHM = false;
3381 return true;
3382 }
3383 UseTransparentHugePages = false;
3384 }
3385
3386 if (UseHugeTLBFS) {
3387 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3388 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3389 UseSHM = false;
3390 return true;
3391 }
3392 UseHugeTLBFS = false;
3393 }
3394
3395 return UseSHM;
3396 }
3397
3398 void os::large_page_init() {
3399 if (!UseLargePages &&
3400 !UseTransparentHugePages &&
3401 !UseHugeTLBFS &&
3402 !UseSHM) {
3403 // Not using large pages.
3404 return;
3405 }
3406
3407 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3408 // The user explicitly turned off large pages.
3409 // Ignore the rest of the large pages flags.
3410 UseTransparentHugePages = false;
3411 UseHugeTLBFS = false;
3412 UseSHM = false;
3413 return;
3414 }
3415
3416 size_t large_page_size = Linux::setup_large_page_size();
3417 UseLargePages = Linux::setup_large_page_type(large_page_size);
3418
3419 set_coredump_filter();
3420 }
3421
3422 #ifndef SHM_HUGETLB
3423 #define SHM_HUGETLB 04000
3424 #endif
3425
3426 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3427 // "exec" is passed in but not used. Creating the shared image for
3428 // the code cache doesn't have an SHM_X executable permission to check.
3429 assert(UseLargePages && UseSHM, "only for SHM large pages");
3430 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3431
3432 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3433 return NULL; // Fallback to small pages.
3434 }
3435
3436 key_t key = IPC_PRIVATE;
3437 char *addr;
3438
3439 bool warn_on_failure = UseLargePages &&
3440 (!FLAG_IS_DEFAULT(UseLargePages) ||
3441 !FLAG_IS_DEFAULT(UseSHM) ||
3442 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3443 );
3444 char msg[128];
3445
3446 // Create a large shared memory region to attach to based on size.
3447 // Currently, size is the total size of the heap
3448 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3449 if (shmid == -1) {
3450 // Possible reasons for shmget failure:
3451 // 1. shmmax is too small for Java heap.
3452 // > check shmmax value: cat /proc/sys/kernel/shmmax
3453 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3454 // 2. not enough large page memory.
3455 // > check available large pages: cat /proc/meminfo
3456 // > increase amount of large pages:
3457 // echo new_value > /proc/sys/vm/nr_hugepages
3458 // Note 1: different Linux may use different name for this property,
3459 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3460 // Note 2: it's possible there's enough physical memory available but
3461 // they are so fragmented after a long run that they can't
3462 // coalesce into large pages. Try to reserve large pages when
3463 // the system is still "fresh".
3464 if (warn_on_failure) {
3465 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3466 warning(msg);
3467 }
3468 return NULL;
3469 }
3470
3471 // attach to the region
3472 addr = (char*)shmat(shmid, req_addr, 0);
3473 int err = errno;
3474
3475 // Remove shmid. If shmat() is successful, the actual shared memory segment
3476 // will be deleted when it's detached by shmdt() or when the process
3477 // terminates. If shmat() is not successful this will remove the shared
3478 // segment immediately.
3479 shmctl(shmid, IPC_RMID, NULL);
3480
3481 if ((intptr_t)addr == -1) {
3482 if (warn_on_failure) {
3483 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3484 warning(msg);
3485 }
3486 return NULL;
3487 }
3488
3489 return addr;
3490 }
3491
3492 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3493 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3494
3495 bool warn_on_failure = UseLargePages &&
3496 (!FLAG_IS_DEFAULT(UseLargePages) ||
3497 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3498 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3499
3500 if (warn_on_failure) {
3501 char msg[128];
3502 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3503 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3504 warning(msg);
3505 }
3506 }
3507
3508 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3509 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3510 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3511 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3512
3513 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3514 char* addr = (char*)::mmap(req_addr, bytes, prot,
3515 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3516 -1, 0);
3517
3518 if (addr == MAP_FAILED) {
3519 warn_on_large_pages_failure(req_addr, bytes, errno);
3520 return NULL;
3521 }
3522
3523 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3524
3525 return addr;
3526 }
3527
3528 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3529 size_t large_page_size = os::large_page_size();
3530
3531 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3532
3533 // Allocate small pages.
3534
3535 char* start;
3536 if (req_addr != NULL) {
3537 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3538 assert(is_size_aligned(bytes, alignment), "Must be");
3539 start = os::reserve_memory(bytes, req_addr);
3540 assert(start == NULL || start == req_addr, "Must be");
3541 } else {
3542 start = os::reserve_memory_aligned(bytes, alignment);
3543 }
3544
3545 if (start == NULL) {
3546 return NULL;
3547 }
3548
3549 assert(is_ptr_aligned(start, alignment), "Must be");
3550
3551 // os::reserve_memory_special will record this memory area.
3552 // Need to release it here to prevent overlapping reservations.
3553 MemTracker::record_virtual_memory_release((address)start, bytes);
3554
3555 char* end = start + bytes;
3556
3557 // Find the regions of the allocated chunk that can be promoted to large pages.
3558 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3559 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3560
3561 size_t lp_bytes = lp_end - lp_start;
3562
3563 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3564
3565 if (lp_bytes == 0) {
3566 // The mapped region doesn't even span the start and the end of a large page.
3567 // Fall back to allocate a non-special area.
3568 ::munmap(start, end - start);
3569 return NULL;
3570 }
3571
3572 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3573
3574
3575 void* result;
3576
3577 if (start != lp_start) {
3578 result = ::mmap(start, lp_start - start, prot,
3579 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3580 -1, 0);
3581 if (result == MAP_FAILED) {
3582 ::munmap(lp_start, end - lp_start);
3583 return NULL;
3584 }
3585 }
3586
3587 result = ::mmap(lp_start, lp_bytes, prot,
3588 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3589 -1, 0);
3590 if (result == MAP_FAILED) {
3591 warn_on_large_pages_failure(req_addr, bytes, errno);
3592 // If the mmap above fails, the large pages region will be unmapped and we
3593 // have regions before and after with small pages. Release these regions.
3594 //
3595 // | mapped | unmapped | mapped |
3596 // ^ ^ ^ ^
3597 // start lp_start lp_end end
3598 //
3599 ::munmap(start, lp_start - start);
3600 ::munmap(lp_end, end - lp_end);
3601 return NULL;
3602 }
3603
3604 if (lp_end != end) {
3605 result = ::mmap(lp_end, end - lp_end, prot,
3606 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3607 -1, 0);
3608 if (result == MAP_FAILED) {
3609 ::munmap(start, lp_end - start);
3610 return NULL;
3611 }
3612 }
3613
3614 return start;
3615 }
3616
3617 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3618 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3619 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3620 assert(is_power_of_2(alignment), "Must be");
3621 assert(is_power_of_2(os::large_page_size()), "Must be");
3622 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3623
3624 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3625 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3626 } else {
3627 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3628 }
3629 }
3630
3631 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3632 assert(UseLargePages, "only for large pages");
3633
3634 char* addr;
3635 if (UseSHM) {
3636 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3637 } else {
3638 assert(UseHugeTLBFS, "must be");
3639 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3640 }
3641
3642 if (addr != NULL) {
3643 if (UseNUMAInterleaving) {
3644 numa_make_global(addr, bytes);
3645 }
3646
3647 // The memory is committed
3648 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3649 }
3650
3651 return addr;
3652 }
3653
3654 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3655 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3656 return shmdt(base) == 0;
3657 }
3658
3659 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3660 return pd_release_memory(base, bytes);
3661 }
3662
3663 bool os::release_memory_special(char* base, size_t bytes) {
3664 assert(UseLargePages, "only for large pages");
3665
3666 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3667
3668 bool res;
3669 if (UseSHM) {
3670 res = os::Linux::release_memory_special_shm(base, bytes);
3671 } else {
3672 assert(UseHugeTLBFS, "must be");
3673 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3674 }
3675
3676 if (res) {
3677 tkr.record((address)base, bytes);
3678 } else {
3679 tkr.discard();
3680 }
3681
3682 return res;
3683 }
3684
3685 size_t os::large_page_size() {
3686 return _large_page_size;
3687 }
3688
3689 // With SysV SHM the entire memory region must be allocated as shared
3690 // memory.
3691 // HugeTLBFS allows application to commit large page memory on demand.
3692 // However, when committing memory with HugeTLBFS fails, the region
3693 // that was supposed to be committed will lose the old reservation
3694 // and allow other threads to steal that memory region. Because of this
3695 // behavior we can't commit HugeTLBFS memory.
3696 bool os::can_commit_large_page_memory() {
3697 return UseTransparentHugePages;
3698 }
3699
3700 bool os::can_execute_large_page_memory() {
3701 return UseTransparentHugePages || UseHugeTLBFS;
3702 }
3703
3704 // Reserve memory at an arbitrary address, only if that area is
3705 // available (and not reserved for something else).
3706
3707 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3708 const int max_tries = 10;
3709 char* base[max_tries];
3710 size_t size[max_tries];
3711 const size_t gap = 0x000000;
3712
3713 // Assert only that the size is a multiple of the page size, since
3714 // that's all that mmap requires, and since that's all we really know
3715 // about at this low abstraction level. If we need higher alignment,
3716 // we can either pass an alignment to this method or verify alignment
3717 // in one of the methods further up the call chain. See bug 5044738.
3718 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3719
3720 // Repeatedly allocate blocks until the block is allocated at the
3721 // right spot. Give up after max_tries. Note that reserve_memory() will
3722 // automatically update _highest_vm_reserved_address if the call is
3723 // successful. The variable tracks the highest memory address every reserved
3724 // by JVM. It is used to detect heap-stack collision if running with
3725 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3726 // space than needed, it could confuse the collision detecting code. To
3727 // solve the problem, save current _highest_vm_reserved_address and
3728 // calculate the correct value before return.
3729 address old_highest = _highest_vm_reserved_address;
3730
3731 // Linux mmap allows caller to pass an address as hint; give it a try first,
3732 // if kernel honors the hint then we can return immediately.
3733 char * addr = anon_mmap(requested_addr, bytes, false);
3734 if (addr == requested_addr) {
3735 return requested_addr;
3736 }
3737
3738 if (addr != NULL) {
3739 // mmap() is successful but it fails to reserve at the requested address
3740 anon_munmap(addr, bytes);
3741 }
3742
3743 int i;
3744 for (i = 0; i < max_tries; ++i) {
3745 base[i] = reserve_memory(bytes);
3746
3747 if (base[i] != NULL) {
3748 // Is this the block we wanted?
3749 if (base[i] == requested_addr) {
3750 size[i] = bytes;
3751 break;
3752 }
3753
3754 // Does this overlap the block we wanted? Give back the overlapped
3755 // parts and try again.
3756
3757 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3758 if (top_overlap >= 0 && top_overlap < bytes) {
3759 unmap_memory(base[i], top_overlap);
3760 base[i] += top_overlap;
3761 size[i] = bytes - top_overlap;
3762 } else {
3763 size_t bottom_overlap = base[i] + bytes - requested_addr;
3764 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3765 unmap_memory(requested_addr, bottom_overlap);
3766 size[i] = bytes - bottom_overlap;
3767 } else {
3768 size[i] = bytes;
3769 }
3770 }
3771 }
3772 }
3773
3774 // Give back the unused reserved pieces.
3775
3776 for (int j = 0; j < i; ++j) {
3777 if (base[j] != NULL) {
3778 unmap_memory(base[j], size[j]);
3779 }
3780 }
3781
3782 if (i < max_tries) {
3783 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3784 return requested_addr;
3785 } else {
3786 _highest_vm_reserved_address = old_highest;
3787 return NULL;
3788 }
3789 }
3790
3791 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3792 return ::read(fd, buf, nBytes);
3793 }
3794
3795 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3796 // Solaris uses poll(), linux uses park().
3797 // Poll() is likely a better choice, assuming that Thread.interrupt()
3798 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3799 // SIGSEGV, see 4355769.
3800
3801 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3802 assert(thread == Thread::current(), "thread consistency check");
3803
3804 ParkEvent * const slp = thread->_SleepEvent ;
3805 slp->reset() ;
3806 OrderAccess::fence() ;
3807
3808 if (interruptible) {
3809 jlong prevtime = javaTimeNanos();
3810
3811 for (;;) {
3812 if (os::is_interrupted(thread, true)) {
3813 return OS_INTRPT;
3814 }
3815
3816 jlong newtime = javaTimeNanos();
3817
3818 if (newtime - prevtime < 0) {
3819 // time moving backwards, should only happen if no monotonic clock
3820 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3821 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3822 } else {
3823 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3824 }
3825
3826 if(millis <= 0) {
3827 return OS_OK;
3828 }
3829
3830 prevtime = newtime;
3831
3832 {
3833 assert(thread->is_Java_thread(), "sanity check");
3834 JavaThread *jt = (JavaThread *) thread;
3835 ThreadBlockInVM tbivm(jt);
3836 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3837
3838 jt->set_suspend_equivalent();
3839 // cleared by handle_special_suspend_equivalent_condition() or
3840 // java_suspend_self() via check_and_wait_while_suspended()
3841
3842 slp->park(millis);
3843
3844 // were we externally suspended while we were waiting?
3845 jt->check_and_wait_while_suspended();
3846 }
3847 }
3848 } else {
3849 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3850 jlong prevtime = javaTimeNanos();
3851
3852 for (;;) {
3853 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3854 // the 1st iteration ...
3855 jlong newtime = javaTimeNanos();
3856
3857 if (newtime - prevtime < 0) {
3858 // time moving backwards, should only happen if no monotonic clock
3859 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3860 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3861 } else {
3862 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3863 }
3864
3865 if(millis <= 0) break ;
3866
3867 prevtime = newtime;
3868 slp->park(millis);
3869 }
3870 return OS_OK ;
3871 }
3872 }
3873
3874 //
3875 // Short sleep, direct OS call.
3876 //
3877 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3878 // sched_yield(2) will actually give up the CPU:
3879 //
3880 // * Alone on this pariticular CPU, keeps running.
3881 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3882 // (pre 2.6.39).
3883 //
3884 // So calling this with 0 is an alternative.
3885 //
3886 void os::naked_short_sleep(jlong ms) {
3887 struct timespec req;
3888
3889 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3890 req.tv_sec = 0;
3891 if (ms > 0) {
3892 req.tv_nsec = (ms % 1000) * 1000000;
3893 }
3894 else {
3895 req.tv_nsec = 1;
3896 }
3897
3898 nanosleep(&req, NULL);
3899
3900 return;
3901 }
3902
3903 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3904 void os::infinite_sleep() {
3905 while (true) { // sleep forever ...
3906 ::sleep(100); // ... 100 seconds at a time
3907 }
3908 }
3909
3910 // Used to convert frequent JVM_Yield() to nops
3911 bool os::dont_yield() {
3912 return DontYieldALot;
3913 }
3914
3915 void os::yield() {
3916 sched_yield();
3917 }
3918
3919 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3920
3921 void os::yield_all(int attempts) {
3922 // Yields to all threads, including threads with lower priorities
3923 // Threads on Linux are all with same priority. The Solaris style
3924 // os::yield_all() with nanosleep(1ms) is not necessary.
3925 sched_yield();
3926 }
3927
3928 // Called from the tight loops to possibly influence time-sharing heuristics
3929 void os::loop_breaker(int attempts) {
3930 os::yield_all(attempts);
3931 }
3932
3933 ////////////////////////////////////////////////////////////////////////////////
3934 // thread priority support
3935
3936 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3937 // only supports dynamic priority, static priority must be zero. For real-time
3938 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3939 // However, for large multi-threaded applications, SCHED_RR is not only slower
3940 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3941 // of 5 runs - Sep 2005).
3942 //
3943 // The following code actually changes the niceness of kernel-thread/LWP. It
3944 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3945 // not the entire user process, and user level threads are 1:1 mapped to kernel
3946 // threads. It has always been the case, but could change in the future. For
3947 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3948 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3949
3950 int os::java_to_os_priority[CriticalPriority + 1] = {
3951 19, // 0 Entry should never be used
3952
3953 4, // 1 MinPriority
3954 3, // 2
3955 2, // 3
3956
3957 1, // 4
3958 0, // 5 NormPriority
3959 -1, // 6
3960
3961 -2, // 7
3962 -3, // 8
3963 -4, // 9 NearMaxPriority
3964
3965 -5, // 10 MaxPriority
3966
3967 -5 // 11 CriticalPriority
3968 };
3969
3970 static int prio_init() {
3971 if (ThreadPriorityPolicy == 1) {
3972 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3973 // if effective uid is not root. Perhaps, a more elegant way of doing
3974 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3975 if (geteuid() != 0) {
3976 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3977 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3978 }
3979 ThreadPriorityPolicy = 0;
3980 }
3981 }
3982 if (UseCriticalJavaThreadPriority) {
3983 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3984 }
3985 return 0;
3986 }
3987
3988 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3989 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3990
3991 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3992 return (ret == 0) ? OS_OK : OS_ERR;
3993 }
3994
3995 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3996 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3997 *priority_ptr = java_to_os_priority[NormPriority];
3998 return OS_OK;
3999 }
4000
4001 errno = 0;
4002 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4003 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4004 }
4005
4006 // Hint to the underlying OS that a task switch would not be good.
4007 // Void return because it's a hint and can fail.
4008 void os::hint_no_preempt() {}
4009
4010 ////////////////////////////////////////////////////////////////////////////////
4011 // suspend/resume support
4012
4013 // the low-level signal-based suspend/resume support is a remnant from the
4014 // old VM-suspension that used to be for java-suspension, safepoints etc,
4015 // within hotspot. Now there is a single use-case for this:
4016 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4017 // that runs in the watcher thread.
4018 // The remaining code is greatly simplified from the more general suspension
4019 // code that used to be used.
4020 //
4021 // The protocol is quite simple:
4022 // - suspend:
4023 // - sends a signal to the target thread
4024 // - polls the suspend state of the osthread using a yield loop
4025 // - target thread signal handler (SR_handler) sets suspend state
4026 // and blocks in sigsuspend until continued
4027 // - resume:
4028 // - sets target osthread state to continue
4029 // - sends signal to end the sigsuspend loop in the SR_handler
4030 //
4031 // Note that the SR_lock plays no role in this suspend/resume protocol.
4032 //
4033
4034 static void resume_clear_context(OSThread *osthread) {
4035 osthread->set_ucontext(NULL);
4036 osthread->set_siginfo(NULL);
4037 }
4038
4039 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4040 osthread->set_ucontext(context);
4041 osthread->set_siginfo(siginfo);
4042 }
4043
4044 //
4045 // Handler function invoked when a thread's execution is suspended or
4046 // resumed. We have to be careful that only async-safe functions are
4047 // called here (Note: most pthread functions are not async safe and
4048 // should be avoided.)
4049 //
4050 // Note: sigwait() is a more natural fit than sigsuspend() from an
4051 // interface point of view, but sigwait() prevents the signal hander
4052 // from being run. libpthread would get very confused by not having
4053 // its signal handlers run and prevents sigwait()'s use with the
4054 // mutex granting granting signal.
4055 //
4056 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4057 //
4058 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4059 // Save and restore errno to avoid confusing native code with EINTR
4060 // after sigsuspend.
4061 int old_errno = errno;
4062
4063 Thread* thread = Thread::current();
4064 OSThread* osthread = thread->osthread();
4065 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4066
4067 os::SuspendResume::State current = osthread->sr.state();
4068 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4069 suspend_save_context(osthread, siginfo, context);
4070
4071 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4072 os::SuspendResume::State state = osthread->sr.suspended();
4073 if (state == os::SuspendResume::SR_SUSPENDED) {
4074 sigset_t suspend_set; // signals for sigsuspend()
4075
4076 // get current set of blocked signals and unblock resume signal
4077 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4078 sigdelset(&suspend_set, SR_signum);
4079
4080 sr_semaphore.signal();
4081 // wait here until we are resumed
4082 while (1) {
4083 sigsuspend(&suspend_set);
4084
4085 os::SuspendResume::State result = osthread->sr.running();
4086 if (result == os::SuspendResume::SR_RUNNING) {
4087 sr_semaphore.signal();
4088 break;
4089 }
4090 }
4091
4092 } else if (state == os::SuspendResume::SR_RUNNING) {
4093 // request was cancelled, continue
4094 } else {
4095 ShouldNotReachHere();
4096 }
4097
4098 resume_clear_context(osthread);
4099 } else if (current == os::SuspendResume::SR_RUNNING) {
4100 // request was cancelled, continue
4101 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4102 // ignore
4103 } else {
4104 // ignore
4105 }
4106
4107 errno = old_errno;
4108 }
4109
4110
4111 static int SR_initialize() {
4112 struct sigaction act;
4113 char *s;
4114 /* Get signal number to use for suspend/resume */
4115 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4116 int sig = ::strtol(s, 0, 10);
4117 if (sig > 0 || sig < _NSIG) {
4118 SR_signum = sig;
4119 }
4120 }
4121
4122 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4123 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4124
4125 sigemptyset(&SR_sigset);
4126 sigaddset(&SR_sigset, SR_signum);
4127
4128 /* Set up signal handler for suspend/resume */
4129 act.sa_flags = SA_RESTART|SA_SIGINFO;
4130 act.sa_handler = (void (*)(int)) SR_handler;
4131
4132 // SR_signum is blocked by default.
4133 // 4528190 - We also need to block pthread restart signal (32 on all
4134 // supported Linux platforms). Note that LinuxThreads need to block
4135 // this signal for all threads to work properly. So we don't have
4136 // to use hard-coded signal number when setting up the mask.
4137 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4138
4139 if (sigaction(SR_signum, &act, 0) == -1) {
4140 return -1;
4141 }
4142
4143 // Save signal flag
4144 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4145 return 0;
4146 }
4147
4148 static int sr_notify(OSThread* osthread) {
4149 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4150 assert_status(status == 0, status, "pthread_kill");
4151 return status;
4152 }
4153
4154 // "Randomly" selected value for how long we want to spin
4155 // before bailing out on suspending a thread, also how often
4156 // we send a signal to a thread we want to resume
4157 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4158 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4159
4160 // returns true on success and false on error - really an error is fatal
4161 // but this seems the normal response to library errors
4162 static bool do_suspend(OSThread* osthread) {
4163 assert(osthread->sr.is_running(), "thread should be running");
4164 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4165
4166 // mark as suspended and send signal
4167 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4168 // failed to switch, state wasn't running?
4169 ShouldNotReachHere();
4170 return false;
4171 }
4172
4173 if (sr_notify(osthread) != 0) {
4174 ShouldNotReachHere();
4175 }
4176
4177 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4178 while (true) {
4179 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4180 break;
4181 } else {
4182 // timeout
4183 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4184 if (cancelled == os::SuspendResume::SR_RUNNING) {
4185 return false;
4186 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4187 // make sure that we consume the signal on the semaphore as well
4188 sr_semaphore.wait();
4189 break;
4190 } else {
4191 ShouldNotReachHere();
4192 return false;
4193 }
4194 }
4195 }
4196
4197 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4198 return true;
4199 }
4200
4201 static void do_resume(OSThread* osthread) {
4202 assert(osthread->sr.is_suspended(), "thread should be suspended");
4203 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4204
4205 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4206 // failed to switch to WAKEUP_REQUEST
4207 ShouldNotReachHere();
4208 return;
4209 }
4210
4211 while (true) {
4212 if (sr_notify(osthread) == 0) {
4213 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4214 if (osthread->sr.is_running()) {
4215 return;
4216 }
4217 }
4218 } else {
4219 ShouldNotReachHere();
4220 }
4221 }
4222
4223 guarantee(osthread->sr.is_running(), "Must be running!");
4224 }
4225
4226 ////////////////////////////////////////////////////////////////////////////////
4227 // interrupt support
4228
4229 void os::interrupt(Thread* thread) {
4230 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4231 "possibility of dangling Thread pointer");
4232
4233 OSThread* osthread = thread->osthread();
4234
4235 if (!osthread->interrupted()) {
4236 osthread->set_interrupted(true);
4237 // More than one thread can get here with the same value of osthread,
4238 // resulting in multiple notifications. We do, however, want the store
4239 // to interrupted() to be visible to other threads before we execute unpark().
4240 OrderAccess::fence();
4241 ParkEvent * const slp = thread->_SleepEvent ;
4242 if (slp != NULL) slp->unpark() ;
4243 }
4244
4245 // For JSR166. Unpark even if interrupt status already was set
4246 if (thread->is_Java_thread())
4247 ((JavaThread*)thread)->parker()->unpark();
4248
4249 ParkEvent * ev = thread->_ParkEvent ;
4250 if (ev != NULL) ev->unpark() ;
4251
4252 }
4253
4254 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4255 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4256 "possibility of dangling Thread pointer");
4257
4258 OSThread* osthread = thread->osthread();
4259
4260 bool interrupted = osthread->interrupted();
4261
4262 if (interrupted && clear_interrupted) {
4263 osthread->set_interrupted(false);
4264 // consider thread->_SleepEvent->reset() ... optional optimization
4265 }
4266
4267 return interrupted;
4268 }
4269
4270 ///////////////////////////////////////////////////////////////////////////////////
4271 // signal handling (except suspend/resume)
4272
4273 // This routine may be used by user applications as a "hook" to catch signals.
4274 // The user-defined signal handler must pass unrecognized signals to this
4275 // routine, and if it returns true (non-zero), then the signal handler must
4276 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4277 // routine will never retun false (zero), but instead will execute a VM panic
4278 // routine kill the process.
4279 //
4280 // If this routine returns false, it is OK to call it again. This allows
4281 // the user-defined signal handler to perform checks either before or after
4282 // the VM performs its own checks. Naturally, the user code would be making
4283 // a serious error if it tried to handle an exception (such as a null check
4284 // or breakpoint) that the VM was generating for its own correct operation.
4285 //
4286 // This routine may recognize any of the following kinds of signals:
4287 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4288 // It should be consulted by handlers for any of those signals.
4289 //
4290 // The caller of this routine must pass in the three arguments supplied
4291 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4292 // field of the structure passed to sigaction(). This routine assumes that
4293 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4294 //
4295 // Note that the VM will print warnings if it detects conflicting signal
4296 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4297 //
4298 extern "C" JNIEXPORT int
4299 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4300 void* ucontext, int abort_if_unrecognized);
4301
4302 void signalHandler(int sig, siginfo_t* info, void* uc) {
4303 assert(info != NULL && uc != NULL, "it must be old kernel");
4304 int orig_errno = errno; // Preserve errno value over signal handler.
4305 JVM_handle_linux_signal(sig, info, uc, true);
4306 errno = orig_errno;
4307 }
4308
4309
4310 // This boolean allows users to forward their own non-matching signals
4311 // to JVM_handle_linux_signal, harmlessly.
4312 bool os::Linux::signal_handlers_are_installed = false;
4313
4314 // For signal-chaining
4315 struct sigaction os::Linux::sigact[MAXSIGNUM];
4316 unsigned int os::Linux::sigs = 0;
4317 bool os::Linux::libjsig_is_loaded = false;
4318 typedef struct sigaction *(*get_signal_t)(int);
4319 get_signal_t os::Linux::get_signal_action = NULL;
4320
4321 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4322 struct sigaction *actp = NULL;
4323
4324 if (libjsig_is_loaded) {
4325 // Retrieve the old signal handler from libjsig
4326 actp = (*get_signal_action)(sig);
4327 }
4328 if (actp == NULL) {
4329 // Retrieve the preinstalled signal handler from jvm
4330 actp = get_preinstalled_handler(sig);
4331 }
4332
4333 return actp;
4334 }
4335
4336 static bool call_chained_handler(struct sigaction *actp, int sig,
4337 siginfo_t *siginfo, void *context) {
4338 // Call the old signal handler
4339 if (actp->sa_handler == SIG_DFL) {
4340 // It's more reasonable to let jvm treat it as an unexpected exception
4341 // instead of taking the default action.
4342 return false;
4343 } else if (actp->sa_handler != SIG_IGN) {
4344 if ((actp->sa_flags & SA_NODEFER) == 0) {
4345 // automaticlly block the signal
4346 sigaddset(&(actp->sa_mask), sig);
4347 }
4348
4349 sa_handler_t hand;
4350 sa_sigaction_t sa;
4351 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4352 // retrieve the chained handler
4353 if (siginfo_flag_set) {
4354 sa = actp->sa_sigaction;
4355 } else {
4356 hand = actp->sa_handler;
4357 }
4358
4359 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4360 actp->sa_handler = SIG_DFL;
4361 }
4362
4363 // try to honor the signal mask
4364 sigset_t oset;
4365 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4366
4367 // call into the chained handler
4368 if (siginfo_flag_set) {
4369 (*sa)(sig, siginfo, context);
4370 } else {
4371 (*hand)(sig);
4372 }
4373
4374 // restore the signal mask
4375 pthread_sigmask(SIG_SETMASK, &oset, 0);
4376 }
4377 // Tell jvm's signal handler the signal is taken care of.
4378 return true;
4379 }
4380
4381 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4382 bool chained = false;
4383 // signal-chaining
4384 if (UseSignalChaining) {
4385 struct sigaction *actp = get_chained_signal_action(sig);
4386 if (actp != NULL) {
4387 chained = call_chained_handler(actp, sig, siginfo, context);
4388 }
4389 }
4390 return chained;
4391 }
4392
4393 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4394 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4395 return &sigact[sig];
4396 }
4397 return NULL;
4398 }
4399
4400 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4401 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4402 sigact[sig] = oldAct;
4403 sigs |= (unsigned int)1 << sig;
4404 }
4405
4406 // for diagnostic
4407 int os::Linux::sigflags[MAXSIGNUM];
4408
4409 int os::Linux::get_our_sigflags(int sig) {
4410 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4411 return sigflags[sig];
4412 }
4413
4414 void os::Linux::set_our_sigflags(int sig, int flags) {
4415 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4416 sigflags[sig] = flags;
4417 }
4418
4419 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4420 // Check for overwrite.
4421 struct sigaction oldAct;
4422 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4423
4424 void* oldhand = oldAct.sa_sigaction
4425 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4426 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4427 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4428 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4429 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4430 if (AllowUserSignalHandlers || !set_installed) {
4431 // Do not overwrite; user takes responsibility to forward to us.
4432 return;
4433 } else if (UseSignalChaining) {
4434 // save the old handler in jvm
4435 save_preinstalled_handler(sig, oldAct);
4436 // libjsig also interposes the sigaction() call below and saves the
4437 // old sigaction on it own.
4438 } else {
4439 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4440 "%#lx for signal %d.", (long)oldhand, sig));
4441 }
4442 }
4443
4444 struct sigaction sigAct;
4445 sigfillset(&(sigAct.sa_mask));
4446 sigAct.sa_handler = SIG_DFL;
4447 if (!set_installed) {
4448 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4449 } else {
4450 sigAct.sa_sigaction = signalHandler;
4451 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4452 }
4453 // Save flags, which are set by ours
4454 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4455 sigflags[sig] = sigAct.sa_flags;
4456
4457 int ret = sigaction(sig, &sigAct, &oldAct);
4458 assert(ret == 0, "check");
4459
4460 void* oldhand2 = oldAct.sa_sigaction
4461 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4462 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4463 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4464 }
4465
4466 // install signal handlers for signals that HotSpot needs to
4467 // handle in order to support Java-level exception handling.
4468
4469 void os::Linux::install_signal_handlers() {
4470 if (!signal_handlers_are_installed) {
4471 signal_handlers_are_installed = true;
4472
4473 // signal-chaining
4474 typedef void (*signal_setting_t)();
4475 signal_setting_t begin_signal_setting = NULL;
4476 signal_setting_t end_signal_setting = NULL;
4477 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4478 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4479 if (begin_signal_setting != NULL) {
4480 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4481 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4482 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4483 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4484 libjsig_is_loaded = true;
4485 assert(UseSignalChaining, "should enable signal-chaining");
4486 }
4487 if (libjsig_is_loaded) {
4488 // Tell libjsig jvm is setting signal handlers
4489 (*begin_signal_setting)();
4490 }
4491
4492 set_signal_handler(SIGSEGV, true);
4493 set_signal_handler(SIGPIPE, true);
4494 set_signal_handler(SIGBUS, true);
4495 set_signal_handler(SIGILL, true);
4496 set_signal_handler(SIGFPE, true);
4497 set_signal_handler(SIGXFSZ, true);
4498
4499 if (libjsig_is_loaded) {
4500 // Tell libjsig jvm finishes setting signal handlers
4501 (*end_signal_setting)();
4502 }
4503
4504 // We don't activate signal checker if libjsig is in place, we trust ourselves
4505 // and if UserSignalHandler is installed all bets are off.
4506 // Log that signal checking is off only if -verbose:jni is specified.
4507 if (CheckJNICalls) {
4508 if (libjsig_is_loaded) {
4509 if (PrintJNIResolving) {
4510 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4511 }
4512 check_signals = false;
4513 }
4514 if (AllowUserSignalHandlers) {
4515 if (PrintJNIResolving) {
4516 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4517 }
4518 check_signals = false;
4519 }
4520 }
4521 }
4522 }
4523
4524 // This is the fastest way to get thread cpu time on Linux.
4525 // Returns cpu time (user+sys) for any thread, not only for current.
4526 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4527 // It might work on 2.6.10+ with a special kernel/glibc patch.
4528 // For reference, please, see IEEE Std 1003.1-2004:
4529 // http://www.unix.org/single_unix_specification
4530
4531 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4532 struct timespec tp;
4533 int rc = os::Linux::clock_gettime(clockid, &tp);
4534 assert(rc == 0, "clock_gettime is expected to return 0 code");
4535
4536 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4537 }
4538
4539 /////
4540 // glibc on Linux platform uses non-documented flag
4541 // to indicate, that some special sort of signal
4542 // trampoline is used.
4543 // We will never set this flag, and we should
4544 // ignore this flag in our diagnostic
4545 #ifdef SIGNIFICANT_SIGNAL_MASK
4546 #undef SIGNIFICANT_SIGNAL_MASK
4547 #endif
4548 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4549
4550 static const char* get_signal_handler_name(address handler,
4551 char* buf, int buflen) {
4552 int offset;
4553 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4554 if (found) {
4555 // skip directory names
4556 const char *p1, *p2;
4557 p1 = buf;
4558 size_t len = strlen(os::file_separator());
4559 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4560 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4561 } else {
4562 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4563 }
4564 return buf;
4565 }
4566
4567 static void print_signal_handler(outputStream* st, int sig,
4568 char* buf, size_t buflen) {
4569 struct sigaction sa;
4570
4571 sigaction(sig, NULL, &sa);
4572
4573 // See comment for SIGNIFICANT_SIGNAL_MASK define
4574 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4575
4576 st->print("%s: ", os::exception_name(sig, buf, buflen));
4577
4578 address handler = (sa.sa_flags & SA_SIGINFO)
4579 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4580 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4581
4582 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4583 st->print("SIG_DFL");
4584 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4585 st->print("SIG_IGN");
4586 } else {
4587 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4588 }
4589
4590 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4591
4592 address rh = VMError::get_resetted_sighandler(sig);
4593 // May be, handler was resetted by VMError?
4594 if(rh != NULL) {
4595 handler = rh;
4596 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4597 }
4598
4599 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4600
4601 // Check: is it our handler?
4602 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4603 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4604 // It is our signal handler
4605 // check for flags, reset system-used one!
4606 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4607 st->print(
4608 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4609 os::Linux::get_our_sigflags(sig));
4610 }
4611 }
4612 st->cr();
4613 }
4614
4615
4616 #define DO_SIGNAL_CHECK(sig) \
4617 if (!sigismember(&check_signal_done, sig)) \
4618 os::Linux::check_signal_handler(sig)
4619
4620 // This method is a periodic task to check for misbehaving JNI applications
4621 // under CheckJNI, we can add any periodic checks here
4622
4623 void os::run_periodic_checks() {
4624
4625 if (check_signals == false) return;
4626
4627 // SEGV and BUS if overridden could potentially prevent
4628 // generation of hs*.log in the event of a crash, debugging
4629 // such a case can be very challenging, so we absolutely
4630 // check the following for a good measure:
4631 DO_SIGNAL_CHECK(SIGSEGV);
4632 DO_SIGNAL_CHECK(SIGILL);
4633 DO_SIGNAL_CHECK(SIGFPE);
4634 DO_SIGNAL_CHECK(SIGBUS);
4635 DO_SIGNAL_CHECK(SIGPIPE);
4636 DO_SIGNAL_CHECK(SIGXFSZ);
4637
4638
4639 // ReduceSignalUsage allows the user to override these handlers
4640 // see comments at the very top and jvm_solaris.h
4641 if (!ReduceSignalUsage) {
4642 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4643 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4644 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4645 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4646 }
4647
4648 DO_SIGNAL_CHECK(SR_signum);
4649 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4650 }
4651
4652 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4653
4654 static os_sigaction_t os_sigaction = NULL;
4655
4656 void os::Linux::check_signal_handler(int sig) {
4657 char buf[O_BUFLEN];
4658 address jvmHandler = NULL;
4659
4660
4661 struct sigaction act;
4662 if (os_sigaction == NULL) {
4663 // only trust the default sigaction, in case it has been interposed
4664 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4665 if (os_sigaction == NULL) return;
4666 }
4667
4668 os_sigaction(sig, (struct sigaction*)NULL, &act);
4669
4670
4671 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4672
4673 address thisHandler = (act.sa_flags & SA_SIGINFO)
4674 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4675 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4676
4677
4678 switch(sig) {
4679 case SIGSEGV:
4680 case SIGBUS:
4681 case SIGFPE:
4682 case SIGPIPE:
4683 case SIGILL:
4684 case SIGXFSZ:
4685 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4686 break;
4687
4688 case SHUTDOWN1_SIGNAL:
4689 case SHUTDOWN2_SIGNAL:
4690 case SHUTDOWN3_SIGNAL:
4691 case BREAK_SIGNAL:
4692 jvmHandler = (address)user_handler();
4693 break;
4694
4695 case INTERRUPT_SIGNAL:
4696 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4697 break;
4698
4699 default:
4700 if (sig == SR_signum) {
4701 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4702 } else {
4703 return;
4704 }
4705 break;
4706 }
4707
4708 if (thisHandler != jvmHandler) {
4709 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4710 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4711 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4712 // No need to check this sig any longer
4713 sigaddset(&check_signal_done, sig);
4714 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4715 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4716 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4717 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4718 // No need to check this sig any longer
4719 sigaddset(&check_signal_done, sig);
4720 }
4721
4722 // Dump all the signal
4723 if (sigismember(&check_signal_done, sig)) {
4724 print_signal_handlers(tty, buf, O_BUFLEN);
4725 }
4726 }
4727
4728 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4729
4730 extern bool signal_name(int signo, char* buf, size_t len);
4731
4732 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4733 if (0 < exception_code && exception_code <= SIGRTMAX) {
4734 // signal
4735 if (!signal_name(exception_code, buf, size)) {
4736 jio_snprintf(buf, size, "SIG%d", exception_code);
4737 }
4738 return buf;
4739 } else {
4740 return NULL;
4741 }
4742 }
4743
4744 // this is called _before_ the most of global arguments have been parsed
4745 void os::init(void) {
4746 char dummy; /* used to get a guess on initial stack address */
4747 // first_hrtime = gethrtime();
4748
4749 // With LinuxThreads the JavaMain thread pid (primordial thread)
4750 // is different than the pid of the java launcher thread.
4751 // So, on Linux, the launcher thread pid is passed to the VM
4752 // via the sun.java.launcher.pid property.
4753 // Use this property instead of getpid() if it was correctly passed.
4754 // See bug 6351349.
4755 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4756
4757 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4758
4759 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4760
4761 init_random(1234567);
4762
4763 ThreadCritical::initialize();
4764
4765 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4766 if (Linux::page_size() == -1) {
4767 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4768 strerror(errno)));
4769 }
4770 init_page_sizes((size_t) Linux::page_size());
4771
4772 Linux::initialize_system_info();
4773
4774 // main_thread points to the aboriginal thread
4775 Linux::_main_thread = pthread_self();
4776
4777 Linux::clock_init();
4778 initial_time_count = javaTimeNanos();
4779
4780 // pthread_condattr initialization for monotonic clock
4781 int status;
4782 pthread_condattr_t* _condattr = os::Linux::condAttr();
4783 if ((status = pthread_condattr_init(_condattr)) != 0) {
4784 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4785 }
4786 // Only set the clock if CLOCK_MONOTONIC is available
4787 if (Linux::supports_monotonic_clock()) {
4788 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4789 if (status == EINVAL) {
4790 warning("Unable to use monotonic clock with relative timed-waits" \
4791 " - changes to the time-of-day clock may have adverse affects");
4792 } else {
4793 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4794 }
4795 }
4796 }
4797 // else it defaults to CLOCK_REALTIME
4798
4799 pthread_mutex_init(&dl_mutex, NULL);
4800
4801 // If the pagesize of the VM is greater than 8K determine the appropriate
4802 // number of initial guard pages. The user can change this with the
4803 // command line arguments, if needed.
4804 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4805 StackYellowPages = 1;
4806 StackRedPages = 1;
4807 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4808 }
4809 }
4810
4811 // To install functions for atexit system call
4812 extern "C" {
4813 static void perfMemory_exit_helper() {
4814 perfMemory_exit();
4815 }
4816 }
4817
4818 // this is called _after_ the global arguments have been parsed
4819 jint os::init_2(void)
4820 {
4821 Linux::fast_thread_clock_init();
4822
4823 // Allocate a single page and mark it as readable for safepoint polling
4824 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4825 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4826
4827 os::set_polling_page( polling_page );
4828
4829 #ifndef PRODUCT
4830 if(Verbose && PrintMiscellaneous)
4831 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4832 #endif
4833
4834 if (!UseMembar) {
4835 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4836 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4837 os::set_memory_serialize_page( mem_serialize_page );
4838
4839 #ifndef PRODUCT
4840 if(Verbose && PrintMiscellaneous)
4841 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4842 #endif
4843 }
4844
4845 // initialize suspend/resume support - must do this before signal_sets_init()
4846 if (SR_initialize() != 0) {
4847 perror("SR_initialize failed");
4848 return JNI_ERR;
4849 }
4850
4851 Linux::signal_sets_init();
4852 Linux::install_signal_handlers();
4853
4854 // Check minimum allowable stack size for thread creation and to initialize
4855 // the java system classes, including StackOverflowError - depends on page
4856 // size. Add a page for compiler2 recursion in main thread.
4857 // Add in 2*BytesPerWord times page size to account for VM stack during
4858 // class initialization depending on 32 or 64 bit VM.
4859 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4860 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4861 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4862
4863 size_t threadStackSizeInBytes = ThreadStackSize * K;
4864 if (threadStackSizeInBytes != 0 &&
4865 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4866 tty->print_cr("\nThe stack size specified is too small, "
4867 "Specify at least %dk",
4868 os::Linux::min_stack_allowed/ K);
4869 return JNI_ERR;
4870 }
4871
4872 // Make the stack size a multiple of the page size so that
4873 // the yellow/red zones can be guarded.
4874 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4875 vm_page_size()));
4876
4877 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4878
4879 #if defined(IA32)
4880 workaround_expand_exec_shield_cs_limit();
4881 #endif
4882
4883 Linux::libpthread_init();
4884 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4885 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4886 Linux::glibc_version(), Linux::libpthread_version(),
4887 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4888 }
4889
4890 if (UseNUMA) {
4891 if (!Linux::libnuma_init()) {
4892 UseNUMA = false;
4893 } else {
4894 if ((Linux::numa_max_node() < 1)) {
4895 // There's only one node(they start from 0), disable NUMA.
4896 UseNUMA = false;
4897 }
4898 }
4899 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4900 // we can make the adaptive lgrp chunk resizing work. If the user specified
4901 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4902 // disable adaptive resizing.
4903 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4904 if (FLAG_IS_DEFAULT(UseNUMA)) {
4905 UseNUMA = false;
4906 } else {
4907 if (FLAG_IS_DEFAULT(UseLargePages) &&
4908 FLAG_IS_DEFAULT(UseSHM) &&
4909 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4910 UseLargePages = false;
4911 } else {
4912 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4913 UseAdaptiveSizePolicy = false;
4914 UseAdaptiveNUMAChunkSizing = false;
4915 }
4916 }
4917 }
4918 if (!UseNUMA && ForceNUMA) {
4919 UseNUMA = true;
4920 }
4921 }
4922
4923 if (MaxFDLimit) {
4924 // set the number of file descriptors to max. print out error
4925 // if getrlimit/setrlimit fails but continue regardless.
4926 struct rlimit nbr_files;
4927 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4928 if (status != 0) {
4929 if (PrintMiscellaneous && (Verbose || WizardMode))
4930 perror("os::init_2 getrlimit failed");
4931 } else {
4932 nbr_files.rlim_cur = nbr_files.rlim_max;
4933 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4934 if (status != 0) {
4935 if (PrintMiscellaneous && (Verbose || WizardMode))
4936 perror("os::init_2 setrlimit failed");
4937 }
4938 }
4939 }
4940
4941 // Initialize lock used to serialize thread creation (see os::create_thread)
4942 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4943
4944 // at-exit methods are called in the reverse order of their registration.
4945 // atexit functions are called on return from main or as a result of a
4946 // call to exit(3C). There can be only 32 of these functions registered
4947 // and atexit() does not set errno.
4948
4949 if (PerfAllowAtExitRegistration) {
4950 // only register atexit functions if PerfAllowAtExitRegistration is set.
4951 // atexit functions can be delayed until process exit time, which
4952 // can be problematic for embedded VM situations. Embedded VMs should
4953 // call DestroyJavaVM() to assure that VM resources are released.
4954
4955 // note: perfMemory_exit_helper atexit function may be removed in
4956 // the future if the appropriate cleanup code can be added to the
4957 // VM_Exit VMOperation's doit method.
4958 if (atexit(perfMemory_exit_helper) != 0) {
4959 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4960 }
4961 }
4962
4963 // initialize thread priority policy
4964 prio_init();
4965
4966 return JNI_OK;
4967 }
4968
4969 // this is called at the end of vm_initialization
4970 void os::init_3(void)
4971 {
4972 #ifdef JAVASE_EMBEDDED
4973 // Start the MemNotifyThread
4974 if (LowMemoryProtection) {
4975 MemNotifyThread::start();
4976 }
4977 return;
4978 #endif
4979 }
4980
4981 // Mark the polling page as unreadable
4982 void os::make_polling_page_unreadable(void) {
4983 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4984 fatal("Could not disable polling page");
4985 };
4986
4987 // Mark the polling page as readable
4988 void os::make_polling_page_readable(void) {
4989 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4990 fatal("Could not enable polling page");
4991 }
4992 };
4993
4994 int os::active_processor_count() {
4995 // Linux doesn't yet have a (official) notion of processor sets,
4996 // so just return the number of online processors.
4997 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4998 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4999 return online_cpus;
5000 }
5001
5002 void os::set_native_thread_name(const char *name) {
5003 // Not yet implemented.
5004 return;
5005 }
5006
5007 bool os::distribute_processes(uint length, uint* distribution) {
5008 // Not yet implemented.
5009 return false;
5010 }
5011
5012 bool os::bind_to_processor(uint processor_id) {
5013 // Not yet implemented.
5014 return false;
5015 }
5016
5017 ///
5018
5019 void os::SuspendedThreadTask::internal_do_task() {
5020 if (do_suspend(_thread->osthread())) {
5021 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5022 do_task(context);
5023 do_resume(_thread->osthread());
5024 }
5025 }
5026
5027 class PcFetcher : public os::SuspendedThreadTask {
5028 public:
5029 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5030 ExtendedPC result();
5031 protected:
5032 void do_task(const os::SuspendedThreadTaskContext& context);
5033 private:
5034 ExtendedPC _epc;
5035 };
5036
5037 ExtendedPC PcFetcher::result() {
5038 guarantee(is_done(), "task is not done yet.");
5039 return _epc;
5040 }
5041
5042 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5043 Thread* thread = context.thread();
5044 OSThread* osthread = thread->osthread();
5045 if (osthread->ucontext() != NULL) {
5046 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5047 } else {
5048 // NULL context is unexpected, double-check this is the VMThread
5049 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5050 }
5051 }
5052
5053 // Suspends the target using the signal mechanism and then grabs the PC before
5054 // resuming the target. Used by the flat-profiler only
5055 ExtendedPC os::get_thread_pc(Thread* thread) {
5056 // Make sure that it is called by the watcher for the VMThread
5057 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5058 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5059
5060 PcFetcher fetcher(thread);
5061 fetcher.run();
5062 return fetcher.result();
5063 }
5064
5065 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5066 {
5067 if (is_NPTL()) {
5068 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5069 } else {
5070 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5071 // word back to default 64bit precision if condvar is signaled. Java
5072 // wants 53bit precision. Save and restore current value.
5073 int fpu = get_fpu_control_word();
5074 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5075 set_fpu_control_word(fpu);
5076 return status;
5077 }
5078 }
5079
5080 ////////////////////////////////////////////////////////////////////////////////
5081 // debug support
5082
5083 bool os::find(address addr, outputStream* st) {
5084 Dl_info dlinfo;
5085 memset(&dlinfo, 0, sizeof(dlinfo));
5086 if (dladdr(addr, &dlinfo) != 0) {
5087 st->print(PTR_FORMAT ": ", addr);
5088 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5089 st->print("%s+%#x", dlinfo.dli_sname,
5090 addr - (intptr_t)dlinfo.dli_saddr);
5091 } else if (dlinfo.dli_fbase != NULL) {
5092 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5093 } else {
5094 st->print("<absolute address>");
5095 }
5096 if (dlinfo.dli_fname != NULL) {
5097 st->print(" in %s", dlinfo.dli_fname);
5098 }
5099 if (dlinfo.dli_fbase != NULL) {
5100 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5101 }
5102 st->cr();
5103
5104 if (Verbose) {
5105 // decode some bytes around the PC
5106 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5107 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5108 address lowest = (address) dlinfo.dli_sname;
5109 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5110 if (begin < lowest) begin = lowest;
5111 Dl_info dlinfo2;
5112 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5113 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5114 end = (address) dlinfo2.dli_saddr;
5115 Disassembler::decode(begin, end, st);
5116 }
5117 return true;
5118 }
5119 return false;
5120 }
5121
5122 ////////////////////////////////////////////////////////////////////////////////
5123 // misc
5124
5125 // This does not do anything on Linux. This is basically a hook for being
5126 // able to use structured exception handling (thread-local exception filters)
5127 // on, e.g., Win32.
5128 void
5129 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5130 JavaCallArguments* args, Thread* thread) {
5131 f(value, method, args, thread);
5132 }
5133
5134 void os::print_statistics() {
5135 }
5136
5137 int os::message_box(const char* title, const char* message) {
5138 int i;
5139 fdStream err(defaultStream::error_fd());
5140 for (i = 0; i < 78; i++) err.print_raw("=");
5141 err.cr();
5142 err.print_raw_cr(title);
5143 for (i = 0; i < 78; i++) err.print_raw("-");
5144 err.cr();
5145 err.print_raw_cr(message);
5146 for (i = 0; i < 78; i++) err.print_raw("=");
5147 err.cr();
5148
5149 char buf[16];
5150 // Prevent process from exiting upon "read error" without consuming all CPU
5151 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5152
5153 return buf[0] == 'y' || buf[0] == 'Y';
5154 }
5155
5156 int os::stat(const char *path, struct stat *sbuf) {
5157 char pathbuf[MAX_PATH];
5158 if (strlen(path) > MAX_PATH - 1) {
5159 errno = ENAMETOOLONG;
5160 return -1;
5161 }
5162 os::native_path(strcpy(pathbuf, path));
5163 return ::stat(pathbuf, sbuf);
5164 }
5165
5166 bool os::check_heap(bool force) {
5167 return true;
5168 }
5169
5170 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5171 return ::vsnprintf(buf, count, format, args);
5172 }
5173
5174 // Is a (classpath) directory empty?
5175 bool os::dir_is_empty(const char* path) {
5176 DIR *dir = NULL;
5177 struct dirent *ptr;
5178
5179 dir = opendir(path);
5180 if (dir == NULL) return true;
5181
5182 /* Scan the directory */
5183 bool result = true;
5184 char buf[sizeof(struct dirent) + MAX_PATH];
5185 while (result && (ptr = ::readdir(dir)) != NULL) {
5186 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5187 result = false;
5188 }
5189 }
5190 closedir(dir);
5191 return result;
5192 }
5193
5194 // This code originates from JDK's sysOpen and open64_w
5195 // from src/solaris/hpi/src/system_md.c
5196
5197 #ifndef O_DELETE
5198 #define O_DELETE 0x10000
5199 #endif
5200
5201 // Open a file. Unlink the file immediately after open returns
5202 // if the specified oflag has the O_DELETE flag set.
5203 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5204
5205 int os::open(const char *path, int oflag, int mode) {
5206
5207 if (strlen(path) > MAX_PATH - 1) {
5208 errno = ENAMETOOLONG;
5209 return -1;
5210 }
5211 int fd;
5212 int o_delete = (oflag & O_DELETE);
5213 oflag = oflag & ~O_DELETE;
5214
5215 fd = ::open64(path, oflag, mode);
5216 if (fd == -1) return -1;
5217
5218 //If the open succeeded, the file might still be a directory
5219 {
5220 struct stat64 buf64;
5221 int ret = ::fstat64(fd, &buf64);
5222 int st_mode = buf64.st_mode;
5223
5224 if (ret != -1) {
5225 if ((st_mode & S_IFMT) == S_IFDIR) {
5226 errno = EISDIR;
5227 ::close(fd);
5228 return -1;
5229 }
5230 } else {
5231 ::close(fd);
5232 return -1;
5233 }
5234 }
5235
5236 /*
5237 * All file descriptors that are opened in the JVM and not
5238 * specifically destined for a subprocess should have the
5239 * close-on-exec flag set. If we don't set it, then careless 3rd
5240 * party native code might fork and exec without closing all
5241 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5242 * UNIXProcess.c), and this in turn might:
5243 *
5244 * - cause end-of-file to fail to be detected on some file
5245 * descriptors, resulting in mysterious hangs, or
5246 *
5247 * - might cause an fopen in the subprocess to fail on a system
5248 * suffering from bug 1085341.
5249 *
5250 * (Yes, the default setting of the close-on-exec flag is a Unix
5251 * design flaw)
5252 *
5253 * See:
5254 * 1085341: 32-bit stdio routines should support file descriptors >255
5255 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5256 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5257 */
5258 #ifdef FD_CLOEXEC
5259 {
5260 int flags = ::fcntl(fd, F_GETFD);
5261 if (flags != -1)
5262 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5263 }
5264 #endif
5265
5266 if (o_delete != 0) {
5267 ::unlink(path);
5268 }
5269 return fd;
5270 }
5271
5272
5273 // create binary file, rewriting existing file if required
5274 int os::create_binary_file(const char* path, bool rewrite_existing) {
5275 int oflags = O_WRONLY | O_CREAT;
5276 if (!rewrite_existing) {
5277 oflags |= O_EXCL;
5278 }
5279 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5280 }
5281
5282 // return current position of file pointer
5283 jlong os::current_file_offset(int fd) {
5284 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5285 }
5286
5287 // move file pointer to the specified offset
5288 jlong os::seek_to_file_offset(int fd, jlong offset) {
5289 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5290 }
5291
5292 // This code originates from JDK's sysAvailable
5293 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5294
5295 int os::available(int fd, jlong *bytes) {
5296 jlong cur, end;
5297 int mode;
5298 struct stat64 buf64;
5299
5300 if (::fstat64(fd, &buf64) >= 0) {
5301 mode = buf64.st_mode;
5302 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5303 /*
5304 * XXX: is the following call interruptible? If so, this might
5305 * need to go through the INTERRUPT_IO() wrapper as for other
5306 * blocking, interruptible calls in this file.
5307 */
5308 int n;
5309 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5310 *bytes = n;
5311 return 1;
5312 }
5313 }
5314 }
5315 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5316 return 0;
5317 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5318 return 0;
5319 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5320 return 0;
5321 }
5322 *bytes = end - cur;
5323 return 1;
5324 }
5325
5326 int os::socket_available(int fd, jint *pbytes) {
5327 // Linux doc says EINTR not returned, unlike Solaris
5328 int ret = ::ioctl(fd, FIONREAD, pbytes);
5329
5330 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5331 // is expected to return 0 on failure and 1 on success to the jdk.
5332 return (ret < 0) ? 0 : 1;
5333 }
5334
5335 // Map a block of memory.
5336 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5337 char *addr, size_t bytes, bool read_only,
5338 bool allow_exec) {
5339 int prot;
5340 int flags = MAP_PRIVATE;
5341
5342 if (read_only) {
5343 prot = PROT_READ;
5344 } else {
5345 prot = PROT_READ | PROT_WRITE;
5346 }
5347
5348 if (allow_exec) {
5349 prot |= PROT_EXEC;
5350 }
5351
5352 if (addr != NULL) {
5353 flags |= MAP_FIXED;
5354 }
5355
5356 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5357 fd, file_offset);
5358 if (mapped_address == MAP_FAILED) {
5359 return NULL;
5360 }
5361 return mapped_address;
5362 }
5363
5364
5365 // Remap a block of memory.
5366 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5367 char *addr, size_t bytes, bool read_only,
5368 bool allow_exec) {
5369 // same as map_memory() on this OS
5370 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5371 allow_exec);
5372 }
5373
5374
5375 // Unmap a block of memory.
5376 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5377 return munmap(addr, bytes) == 0;
5378 }
5379
5380 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5381
5382 static clockid_t thread_cpu_clockid(Thread* thread) {
5383 pthread_t tid = thread->osthread()->pthread_id();
5384 clockid_t clockid;
5385
5386 // Get thread clockid
5387 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5388 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5389 return clockid;
5390 }
5391
5392 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5393 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5394 // of a thread.
5395 //
5396 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5397 // the fast estimate available on the platform.
5398
5399 jlong os::current_thread_cpu_time() {
5400 if (os::Linux::supports_fast_thread_cpu_time()) {
5401 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5402 } else {
5403 // return user + sys since the cost is the same
5404 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5405 }
5406 }
5407
5408 jlong os::thread_cpu_time(Thread* thread) {
5409 // consistent with what current_thread_cpu_time() returns
5410 if (os::Linux::supports_fast_thread_cpu_time()) {
5411 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5412 } else {
5413 return slow_thread_cpu_time(thread, true /* user + sys */);
5414 }
5415 }
5416
5417 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5418 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5419 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5420 } else {
5421 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5422 }
5423 }
5424
5425 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5426 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5427 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5428 } else {
5429 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5430 }
5431 }
5432
5433 //
5434 // -1 on error.
5435 //
5436
5437 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5438 static bool proc_task_unchecked = true;
5439 static const char *proc_stat_path = "/proc/%d/stat";
5440 pid_t tid = thread->osthread()->thread_id();
5441 char *s;
5442 char stat[2048];
5443 int statlen;
5444 char proc_name[64];
5445 int count;
5446 long sys_time, user_time;
5447 char cdummy;
5448 int idummy;
5449 long ldummy;
5450 FILE *fp;
5451
5452 // The /proc/<tid>/stat aggregates per-process usage on
5453 // new Linux kernels 2.6+ where NPTL is supported.
5454 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5455 // See bug 6328462.
5456 // There possibly can be cases where there is no directory
5457 // /proc/self/task, so we check its availability.
5458 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5459 // This is executed only once
5460 proc_task_unchecked = false;
5461 fp = fopen("/proc/self/task", "r");
5462 if (fp != NULL) {
5463 proc_stat_path = "/proc/self/task/%d/stat";
5464 fclose(fp);
5465 }
5466 }
5467
5468 sprintf(proc_name, proc_stat_path, tid);
5469 fp = fopen(proc_name, "r");
5470 if ( fp == NULL ) return -1;
5471 statlen = fread(stat, 1, 2047, fp);
5472 stat[statlen] = '\0';
5473 fclose(fp);
5474
5475 // Skip pid and the command string. Note that we could be dealing with
5476 // weird command names, e.g. user could decide to rename java launcher
5477 // to "java 1.4.2 :)", then the stat file would look like
5478 // 1234 (java 1.4.2 :)) R ... ...
5479 // We don't really need to know the command string, just find the last
5480 // occurrence of ")" and then start parsing from there. See bug 4726580.
5481 s = strrchr(stat, ')');
5482 if (s == NULL ) return -1;
5483
5484 // Skip blank chars
5485 do s++; while (isspace(*s));
5486
5487 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5488 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5489 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5490 &user_time, &sys_time);
5491 if ( count != 13 ) return -1;
5492 if (user_sys_cpu_time) {
5493 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5494 } else {
5495 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5496 }
5497 }
5498
5499 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5500 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5501 info_ptr->may_skip_backward = false; // elapsed time not wall time
5502 info_ptr->may_skip_forward = false; // elapsed time not wall time
5503 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5504 }
5505
5506 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5507 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5508 info_ptr->may_skip_backward = false; // elapsed time not wall time
5509 info_ptr->may_skip_forward = false; // elapsed time not wall time
5510 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5511 }
5512
5513 bool os::is_thread_cpu_time_supported() {
5514 return true;
5515 }
5516
5517 // System loadavg support. Returns -1 if load average cannot be obtained.
5518 // Linux doesn't yet have a (official) notion of processor sets,
5519 // so just return the system wide load average.
5520 int os::loadavg(double loadavg[], int nelem) {
5521 return ::getloadavg(loadavg, nelem);
5522 }
5523
5524 void os::pause() {
5525 char filename[MAX_PATH];
5526 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5527 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5528 } else {
5529 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5530 }
5531
5532 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5533 if (fd != -1) {
5534 struct stat buf;
5535 ::close(fd);
5536 while (::stat(filename, &buf) == 0) {
5537 (void)::poll(NULL, 0, 100);
5538 }
5539 } else {
5540 jio_fprintf(stderr,
5541 "Could not open pause file '%s', continuing immediately.\n", filename);
5542 }
5543 }
5544
5545
5546 // Refer to the comments in os_solaris.cpp park-unpark.
5547 //
5548 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5549 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5550 // For specifics regarding the bug see GLIBC BUGID 261237 :
5551 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5552 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5553 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5554 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5555 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5556 // and monitorenter when we're using 1-0 locking. All those operations may result in
5557 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5558 // of libpthread avoids the problem, but isn't practical.
5559 //
5560 // Possible remedies:
5561 //
5562 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5563 // This is palliative and probabilistic, however. If the thread is preempted
5564 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5565 // than the minimum period may have passed, and the abstime may be stale (in the
5566 // past) resultin in a hang. Using this technique reduces the odds of a hang
5567 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5568 //
5569 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5570 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5571 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5572 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5573 // thread.
5574 //
5575 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5576 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5577 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5578 // This also works well. In fact it avoids kernel-level scalability impediments
5579 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5580 // timers in a graceful fashion.
5581 //
5582 // 4. When the abstime value is in the past it appears that control returns
5583 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5584 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5585 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5586 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5587 // It may be possible to avoid reinitialization by checking the return
5588 // value from pthread_cond_timedwait(). In addition to reinitializing the
5589 // condvar we must establish the invariant that cond_signal() is only called
5590 // within critical sections protected by the adjunct mutex. This prevents
5591 // cond_signal() from "seeing" a condvar that's in the midst of being
5592 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5593 // desirable signal-after-unlock optimization that avoids futile context switching.
5594 //
5595 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5596 // structure when a condvar is used or initialized. cond_destroy() would
5597 // release the helper structure. Our reinitialize-after-timedwait fix
5598 // put excessive stress on malloc/free and locks protecting the c-heap.
5599 //
5600 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5601 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5602 // and only enabling the work-around for vulnerable environments.
5603
5604 // utility to compute the abstime argument to timedwait:
5605 // millis is the relative timeout time
5606 // abstime will be the absolute timeout time
5607 // TODO: replace compute_abstime() with unpackTime()
5608
5609 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5610 if (millis < 0) millis = 0;
5611
5612 jlong seconds = millis / 1000;
5613 millis %= 1000;
5614 if (seconds > 50000000) { // see man cond_timedwait(3T)
5615 seconds = 50000000;
5616 }
5617
5618 if (os::Linux::supports_monotonic_clock()) {
5619 struct timespec now;
5620 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5621 assert_status(status == 0, status, "clock_gettime");
5622 abstime->tv_sec = now.tv_sec + seconds;
5623 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5624 if (nanos >= NANOSECS_PER_SEC) {
5625 abstime->tv_sec += 1;
5626 nanos -= NANOSECS_PER_SEC;
5627 }
5628 abstime->tv_nsec = nanos;
5629 } else {
5630 struct timeval now;
5631 int status = gettimeofday(&now, NULL);
5632 assert(status == 0, "gettimeofday");
5633 abstime->tv_sec = now.tv_sec + seconds;
5634 long usec = now.tv_usec + millis * 1000;
5635 if (usec >= 1000000) {
5636 abstime->tv_sec += 1;
5637 usec -= 1000000;
5638 }
5639 abstime->tv_nsec = usec * 1000;
5640 }
5641 return abstime;
5642 }
5643
5644
5645 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5646 // Conceptually TryPark() should be equivalent to park(0).
5647
5648 int os::PlatformEvent::TryPark() {
5649 for (;;) {
5650 const int v = _Event ;
5651 guarantee ((v == 0) || (v == 1), "invariant") ;
5652 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5653 }
5654 }
5655
5656 void os::PlatformEvent::park() { // AKA "down()"
5657 // Invariant: Only the thread associated with the Event/PlatformEvent
5658 // may call park().
5659 // TODO: assert that _Assoc != NULL or _Assoc == Self
5660 int v ;
5661 for (;;) {
5662 v = _Event ;
5663 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5664 }
5665 guarantee (v >= 0, "invariant") ;
5666 if (v == 0) {
5667 // Do this the hard way by blocking ...
5668 int status = pthread_mutex_lock(_mutex);
5669 assert_status(status == 0, status, "mutex_lock");
5670 guarantee (_nParked == 0, "invariant") ;
5671 ++ _nParked ;
5672 while (_Event < 0) {
5673 status = pthread_cond_wait(_cond, _mutex);
5674 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5675 // Treat this the same as if the wait was interrupted
5676 if (status == ETIME) { status = EINTR; }
5677 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5678 }
5679 -- _nParked ;
5680
5681 _Event = 0 ;
5682 status = pthread_mutex_unlock(_mutex);
5683 assert_status(status == 0, status, "mutex_unlock");
5684 // Paranoia to ensure our locked and lock-free paths interact
5685 // correctly with each other.
5686 OrderAccess::fence();
5687 }
5688 guarantee (_Event >= 0, "invariant") ;
5689 }
5690
5691 int os::PlatformEvent::park(jlong millis) {
5692 guarantee (_nParked == 0, "invariant") ;
5693
5694 int v ;
5695 for (;;) {
5696 v = _Event ;
5697 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5698 }
5699 guarantee (v >= 0, "invariant") ;
5700 if (v != 0) return OS_OK ;
5701
5702 // We do this the hard way, by blocking the thread.
5703 // Consider enforcing a minimum timeout value.
5704 struct timespec abst;
5705 compute_abstime(&abst, millis);
5706
5707 int ret = OS_TIMEOUT;
5708 int status = pthread_mutex_lock(_mutex);
5709 assert_status(status == 0, status, "mutex_lock");
5710 guarantee (_nParked == 0, "invariant") ;
5711 ++_nParked ;
5712
5713 // Object.wait(timo) will return because of
5714 // (a) notification
5715 // (b) timeout
5716 // (c) thread.interrupt
5717 //
5718 // Thread.interrupt and object.notify{All} both call Event::set.
5719 // That is, we treat thread.interrupt as a special case of notification.
5720 // The underlying Solaris implementation, cond_timedwait, admits
5721 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5722 // JVM from making those visible to Java code. As such, we must
5723 // filter out spurious wakeups. We assume all ETIME returns are valid.
5724 //
5725 // TODO: properly differentiate simultaneous notify+interrupt.
5726 // In that case, we should propagate the notify to another waiter.
5727
5728 while (_Event < 0) {
5729 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5730 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5731 pthread_cond_destroy (_cond);
5732 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5733 }
5734 assert_status(status == 0 || status == EINTR ||
5735 status == ETIME || status == ETIMEDOUT,
5736 status, "cond_timedwait");
5737 if (!FilterSpuriousWakeups) break ; // previous semantics
5738 if (status == ETIME || status == ETIMEDOUT) break ;
5739 // We consume and ignore EINTR and spurious wakeups.
5740 }
5741 --_nParked ;
5742 if (_Event >= 0) {
5743 ret = OS_OK;
5744 }
5745 _Event = 0 ;
5746 status = pthread_mutex_unlock(_mutex);
5747 assert_status(status == 0, status, "mutex_unlock");
5748 assert (_nParked == 0, "invariant") ;
5749 // Paranoia to ensure our locked and lock-free paths interact
5750 // correctly with each other.
5751 OrderAccess::fence();
5752 return ret;
5753 }
5754
5755 void os::PlatformEvent::unpark() {
5756 // Transitions for _Event:
5757 // 0 :=> 1
5758 // 1 :=> 1
5759 // -1 :=> either 0 or 1; must signal target thread
5760 // That is, we can safely transition _Event from -1 to either
5761 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5762 // unpark() calls.
5763 // See also: "Semaphores in Plan 9" by Mullender & Cox
5764 //
5765 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5766 // that it will take two back-to-back park() calls for the owning
5767 // thread to block. This has the benefit of forcing a spurious return
5768 // from the first park() call after an unpark() call which will help
5769 // shake out uses of park() and unpark() without condition variables.
5770
5771 if (Atomic::xchg(1, &_Event) >= 0) return;
5772
5773 // Wait for the thread associated with the event to vacate
5774 int status = pthread_mutex_lock(_mutex);
5775 assert_status(status == 0, status, "mutex_lock");
5776 int AnyWaiters = _nParked;
5777 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5778 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5779 AnyWaiters = 0;
5780 pthread_cond_signal(_cond);
5781 }
5782 status = pthread_mutex_unlock(_mutex);
5783 assert_status(status == 0, status, "mutex_unlock");
5784 if (AnyWaiters != 0) {
5785 status = pthread_cond_signal(_cond);
5786 assert_status(status == 0, status, "cond_signal");
5787 }
5788
5789 // Note that we signal() _after dropping the lock for "immortal" Events.
5790 // This is safe and avoids a common class of futile wakeups. In rare
5791 // circumstances this can cause a thread to return prematurely from
5792 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5793 // simply re-test the condition and re-park itself.
5794 }
5795
5796
5797 // JSR166
5798 // -------------------------------------------------------
5799
5800 /*
5801 * The solaris and linux implementations of park/unpark are fairly
5802 * conservative for now, but can be improved. They currently use a
5803 * mutex/condvar pair, plus a a count.
5804 * Park decrements count if > 0, else does a condvar wait. Unpark
5805 * sets count to 1 and signals condvar. Only one thread ever waits
5806 * on the condvar. Contention seen when trying to park implies that someone
5807 * is unparking you, so don't wait. And spurious returns are fine, so there
5808 * is no need to track notifications.
5809 */
5810
5811 #define MAX_SECS 100000000
5812 /*
5813 * This code is common to linux and solaris and will be moved to a
5814 * common place in dolphin.
5815 *
5816 * The passed in time value is either a relative time in nanoseconds
5817 * or an absolute time in milliseconds. Either way it has to be unpacked
5818 * into suitable seconds and nanoseconds components and stored in the
5819 * given timespec structure.
5820 * Given time is a 64-bit value and the time_t used in the timespec is only
5821 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5822 * overflow if times way in the future are given. Further on Solaris versions
5823 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5824 * number of seconds, in abstime, is less than current_time + 100,000,000.
5825 * As it will be 28 years before "now + 100000000" will overflow we can
5826 * ignore overflow and just impose a hard-limit on seconds using the value
5827 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5828 * years from "now".
5829 */
5830
5831 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5832 assert (time > 0, "convertTime");
5833 time_t max_secs = 0;
5834
5835 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5836 struct timeval now;
5837 int status = gettimeofday(&now, NULL);
5838 assert(status == 0, "gettimeofday");
5839
5840 max_secs = now.tv_sec + MAX_SECS;
5841
5842 if (isAbsolute) {
5843 jlong secs = time / 1000;
5844 if (secs > max_secs) {
5845 absTime->tv_sec = max_secs;
5846 } else {
5847 absTime->tv_sec = secs;
5848 }
5849 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5850 } else {
5851 jlong secs = time / NANOSECS_PER_SEC;
5852 if (secs >= MAX_SECS) {
5853 absTime->tv_sec = max_secs;
5854 absTime->tv_nsec = 0;
5855 } else {
5856 absTime->tv_sec = now.tv_sec + secs;
5857 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5858 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5859 absTime->tv_nsec -= NANOSECS_PER_SEC;
5860 ++absTime->tv_sec; // note: this must be <= max_secs
5861 }
5862 }
5863 }
5864 } else {
5865 // must be relative using monotonic clock
5866 struct timespec now;
5867 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5868 assert_status(status == 0, status, "clock_gettime");
5869 max_secs = now.tv_sec + MAX_SECS;
5870 jlong secs = time / NANOSECS_PER_SEC;
5871 if (secs >= MAX_SECS) {
5872 absTime->tv_sec = max_secs;
5873 absTime->tv_nsec = 0;
5874 } else {
5875 absTime->tv_sec = now.tv_sec + secs;
5876 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5877 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5878 absTime->tv_nsec -= NANOSECS_PER_SEC;
5879 ++absTime->tv_sec; // note: this must be <= max_secs
5880 }
5881 }
5882 }
5883 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5884 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5885 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5886 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5887 }
5888
5889 void Parker::park(bool isAbsolute, jlong time) {
5890 // Ideally we'd do something useful while spinning, such
5891 // as calling unpackTime().
5892
5893 // Optional fast-path check:
5894 // Return immediately if a permit is available.
5895 // We depend on Atomic::xchg() having full barrier semantics
5896 // since we are doing a lock-free update to _counter.
5897 if (Atomic::xchg(0, &_counter) > 0) return;
5898
5899 Thread* thread = Thread::current();
5900 assert(thread->is_Java_thread(), "Must be JavaThread");
5901 JavaThread *jt = (JavaThread *)thread;
5902
5903 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5904 // Check interrupt before trying to wait
5905 if (Thread::is_interrupted(thread, false)) {
5906 return;
5907 }
5908
5909 // Next, demultiplex/decode time arguments
5910 timespec absTime;
5911 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5912 return;
5913 }
5914 if (time > 0) {
5915 unpackTime(&absTime, isAbsolute, time);
5916 }
5917
5918
5919 // Enter safepoint region
5920 // Beware of deadlocks such as 6317397.
5921 // The per-thread Parker:: mutex is a classic leaf-lock.
5922 // In particular a thread must never block on the Threads_lock while
5923 // holding the Parker:: mutex. If safepoints are pending both the
5924 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5925 ThreadBlockInVM tbivm(jt);
5926
5927 // Don't wait if cannot get lock since interference arises from
5928 // unblocking. Also. check interrupt before trying wait
5929 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5930 return;
5931 }
5932
5933 int status ;
5934 if (_counter > 0) { // no wait needed
5935 _counter = 0;
5936 status = pthread_mutex_unlock(_mutex);
5937 assert (status == 0, "invariant") ;
5938 // Paranoia to ensure our locked and lock-free paths interact
5939 // correctly with each other and Java-level accesses.
5940 OrderAccess::fence();
5941 return;
5942 }
5943
5944 #ifdef ASSERT
5945 // Don't catch signals while blocked; let the running threads have the signals.
5946 // (This allows a debugger to break into the running thread.)
5947 sigset_t oldsigs;
5948 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5949 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5950 #endif
5951
5952 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5953 jt->set_suspend_equivalent();
5954 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5955
5956 assert(_cur_index == -1, "invariant");
5957 if (time == 0) {
5958 _cur_index = REL_INDEX; // arbitrary choice when not timed
5959 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5960 } else {
5961 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5962 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5963 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5964 pthread_cond_destroy (&_cond[_cur_index]) ;
5965 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5966 }
5967 }
5968 _cur_index = -1;
5969 assert_status(status == 0 || status == EINTR ||
5970 status == ETIME || status == ETIMEDOUT,
5971 status, "cond_timedwait");
5972
5973 #ifdef ASSERT
5974 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5975 #endif
5976
5977 _counter = 0 ;
5978 status = pthread_mutex_unlock(_mutex) ;
5979 assert_status(status == 0, status, "invariant") ;
5980 // Paranoia to ensure our locked and lock-free paths interact
5981 // correctly with each other and Java-level accesses.
5982 OrderAccess::fence();
5983
5984 // If externally suspended while waiting, re-suspend
5985 if (jt->handle_special_suspend_equivalent_condition()) {
5986 jt->java_suspend_self();
5987 }
5988 }
5989
5990 void Parker::unpark() {
5991 int s, status ;
5992 status = pthread_mutex_lock(_mutex);
5993 assert (status == 0, "invariant") ;
5994 s = _counter;
5995 _counter = 1;
5996 if (s < 1) {
5997 // thread might be parked
5998 if (_cur_index != -1) {
5999 // thread is definitely parked
6000 if (WorkAroundNPTLTimedWaitHang) {
6001 status = pthread_cond_signal (&_cond[_cur_index]);
6002 assert (status == 0, "invariant");
6003 status = pthread_mutex_unlock(_mutex);
6004 assert (status == 0, "invariant");
6005 } else {
6006 status = pthread_mutex_unlock(_mutex);
6007 assert (status == 0, "invariant");
6008 status = pthread_cond_signal (&_cond[_cur_index]);
6009 assert (status == 0, "invariant");
6010 }
6011 } else {
6012 pthread_mutex_unlock(_mutex);
6013 assert (status == 0, "invariant") ;
6014 }
6015 } else {
6016 pthread_mutex_unlock(_mutex);
6017 assert (status == 0, "invariant") ;
6018 }
6019 }
6020
6021
6022 extern char** environ;
6023
6024 #ifndef __NR_fork
6025 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
6026 #endif
6027
6028 #ifndef __NR_execve
6029 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
6030 #endif
6031
6032 // Run the specified command in a separate process. Return its exit value,
6033 // or -1 on failure (e.g. can't fork a new process).
6034 // Unlike system(), this function can be called from signal handler. It
6035 // doesn't block SIGINT et al.
6036 int os::fork_and_exec(char* cmd) {
6037 const char * argv[4] = {"sh", "-c", cmd, NULL};
6038
6039 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
6040 // pthread_atfork handlers and reset pthread library. All we need is a
6041 // separate process to execve. Make a direct syscall to fork process.
6042 // On IA64 there's no fork syscall, we have to use fork() and hope for
6043 // the best...
6044 pid_t pid = NOT_IA64(syscall(__NR_fork);)
6045 IA64_ONLY(fork();)
6046
6047 if (pid < 0) {
6048 // fork failed
6049 return -1;
6050
6051 } else if (pid == 0) {
6052 // child process
6053
6054 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6055 // first to kill every thread on the thread list. Because this list is
6056 // not reset by fork() (see notes above), execve() will instead kill
6057 // every thread in the parent process. We know this is the only thread
6058 // in the new process, so make a system call directly.
6059 // IA64 should use normal execve() from glibc to match the glibc fork()
6060 // above.
6061 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6062 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6063
6064 // execve failed
6065 _exit(-1);
6066
6067 } else {
6068 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6069 // care about the actual exit code, for now.
6070
6071 int status;
6072
6073 // Wait for the child process to exit. This returns immediately if
6074 // the child has already exited. */
6075 while (waitpid(pid, &status, 0) < 0) {
6076 switch (errno) {
6077 case ECHILD: return 0;
6078 case EINTR: break;
6079 default: return -1;
6080 }
6081 }
6082
6083 if (WIFEXITED(status)) {
6084 // The child exited normally; get its exit code.
6085 return WEXITSTATUS(status);
6086 } else if (WIFSIGNALED(status)) {
6087 // The child exited because of a signal
6088 // The best value to return is 0x80 + signal number,
6089 // because that is what all Unix shells do, and because
6090 // it allows callers to distinguish between process exit and
6091 // process death by signal.
6092 return 0x80 + WTERMSIG(status);
6093 } else {
6094 // Unknown exit code; pass it through
6095 return status;
6096 }
6097 }
6098 }
6099
6100 // is_headless_jre()
6101 //
6102 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6103 // in order to report if we are running in a headless jre
6104 //
6105 // Since JDK8 xawt/libmawt.so was moved into the same directory
6106 // as libawt.so, and renamed libawt_xawt.so
6107 //
6108 bool os::is_headless_jre() {
6109 struct stat statbuf;
6110 char buf[MAXPATHLEN];
6111 char libmawtpath[MAXPATHLEN];
6112 const char *xawtstr = "/xawt/libmawt.so";
6113 const char *new_xawtstr = "/libawt_xawt.so";
6114 char *p;
6115
6116 // Get path to libjvm.so
6117 os::jvm_path(buf, sizeof(buf));
6118
6119 // Get rid of libjvm.so
6120 p = strrchr(buf, '/');
6121 if (p == NULL) return false;
6122 else *p = '\0';
6123
6124 // Get rid of client or server
6125 p = strrchr(buf, '/');
6126 if (p == NULL) return false;
6127 else *p = '\0';
6128
6129 // check xawt/libmawt.so
6130 strcpy(libmawtpath, buf);
6131 strcat(libmawtpath, xawtstr);
6132 if (::stat(libmawtpath, &statbuf) == 0) return false;
6133
6134 // check libawt_xawt.so
6135 strcpy(libmawtpath, buf);
6136 strcat(libmawtpath, new_xawtstr);
6137 if (::stat(libmawtpath, &statbuf) == 0) return false;
6138
6139 return true;
6140 }
6141
6142 // Get the default path to the core file
6143 // Returns the length of the string
6144 int os::get_core_path(char* buffer, size_t bufferSize) {
6145 const char* p = get_current_directory(buffer, bufferSize);
6146
6147 if (p == NULL) {
6148 assert(p != NULL, "failed to get current directory");
6149 return 0;
6150 }
6151
6152 return strlen(buffer);
6153 }
6154
6155 #ifdef JAVASE_EMBEDDED
6156 //
6157 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6158 //
6159 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6160
6161 // ctor
6162 //
6163 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6164 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6165 _fd = fd;
6166
6167 if (os::create_thread(this, os::os_thread)) {
6168 _memnotify_thread = this;
6169 os::set_priority(this, NearMaxPriority);
6170 os::start_thread(this);
6171 }
6172 }
6173
6174 // Where all the work gets done
6175 //
6176 void MemNotifyThread::run() {
6177 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6178
6179 // Set up the select arguments
6180 fd_set rfds;
6181 if (_fd != -1) {
6182 FD_ZERO(&rfds);
6183 FD_SET(_fd, &rfds);
6184 }
6185
6186 // Now wait for the mem_notify device to wake up
6187 while (1) {
6188 // Wait for the mem_notify device to signal us..
6189 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6190 if (rc == -1) {
6191 perror("select!\n");
6192 break;
6193 } else if (rc) {
6194 //ssize_t free_before = os::available_memory();
6195 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6196
6197 // The kernel is telling us there is not much memory left...
6198 // try to do something about that
6199
6200 // If we are not already in a GC, try one.
6201 if (!Universe::heap()->is_gc_active()) {
6202 Universe::heap()->collect(GCCause::_allocation_failure);
6203
6204 //ssize_t free_after = os::available_memory();
6205 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6206 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6207 }
6208 // We might want to do something like the following if we find the GC's are not helping...
6209 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6210 }
6211 }
6212 }
6213
6214 //
6215 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6216 //
6217 void MemNotifyThread::start() {
6218 int fd;
6219 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6220 if (fd < 0) {
6221 return;
6222 }
6223
6224 if (memnotify_thread() == NULL) {
6225 new MemNotifyThread(fd);
6226 }
6227 }
6228
6229 #endif // JAVASE_EMBEDDED
6230
6231
6232 /////////////// Unit tests ///////////////
6233
6234 #ifndef PRODUCT
6235
6236 #define test_log(...) \
6237 do {\
6238 if (VerboseInternalVMTests) { \
6239 tty->print_cr(__VA_ARGS__); \
6240 tty->flush(); \
6241 }\
6242 } while (false)
6243
6244 class TestReserveMemorySpecial : AllStatic {
6245 public:
6246 static void small_page_write(void* addr, size_t size) {
6247 size_t page_size = os::vm_page_size();
6248
6249 char* end = (char*)addr + size;
6250 for (char* p = (char*)addr; p < end; p += page_size) {
6251 *p = 1;
6252 }
6253 }
6254
6255 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6256 if (!UseHugeTLBFS) {
6257 return;
6258 }
6259
6260 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6261
6262 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6263
6264 if (addr != NULL) {
6265 small_page_write(addr, size);
6266
6267 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6268 }
6269 }
6270
6271 static void test_reserve_memory_special_huge_tlbfs_only() {
6272 if (!UseHugeTLBFS) {
6273 return;
6274 }
6275
6276 size_t lp = os::large_page_size();
6277
6278 for (size_t size = lp; size <= lp * 10; size += lp) {
6279 test_reserve_memory_special_huge_tlbfs_only(size);
6280 }
6281 }
6282
6283 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6284 if (!UseHugeTLBFS) {
6285 return;
6286 }
6287
6288 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6289 size, alignment);
6290
6291 assert(size >= os::large_page_size(), "Incorrect input to test");
6292
6293 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6294
6295 if (addr != NULL) {
6296 small_page_write(addr, size);
6297
6298 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6299 }
6300 }
6301
6302 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6303 size_t lp = os::large_page_size();
6304 size_t ag = os::vm_allocation_granularity();
6305
6306 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6307 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6308 }
6309 }
6310
6311 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6312 size_t lp = os::large_page_size();
6313 size_t ag = os::vm_allocation_granularity();
6314
6315 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6316 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6317 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6318 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6319 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6320 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6321 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6322 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6323 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6324 }
6325
6326 static void test_reserve_memory_special_huge_tlbfs() {
6327 if (!UseHugeTLBFS) {
6328 return;
6329 }
6330
6331 test_reserve_memory_special_huge_tlbfs_only();
6332 test_reserve_memory_special_huge_tlbfs_mixed();
6333 }
6334
6335 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6336 if (!UseSHM) {
6337 return;
6338 }
6339
6340 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6341
6342 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6343
6344 if (addr != NULL) {
6345 assert(is_ptr_aligned(addr, alignment), "Check");
6346 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6347
6348 small_page_write(addr, size);
6349
6350 os::Linux::release_memory_special_shm(addr, size);
6351 }
6352 }
6353
6354 static void test_reserve_memory_special_shm() {
6355 size_t lp = os::large_page_size();
6356 size_t ag = os::vm_allocation_granularity();
6357
6358 for (size_t size = ag; size < lp * 3; size += ag) {
6359 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6360 test_reserve_memory_special_shm(size, alignment);
6361 }
6362 }
6363 }
6364
6365 static void test() {
6366 test_reserve_memory_special_huge_tlbfs();
6367 test_reserve_memory_special_shm();
6368 }
6369 };
6370
6371 void TestReserveMemorySpecial_test() {
6372 TestReserveMemorySpecial::test();
6373 }
6374
6375 #endif