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 "interpreter/interpreter.hpp"
33 #include "jvm_linux.h"
34 #include "memory/allocation.inline.hpp"
35 #include "memory/filemap.hpp"
36 #include "mutex_linux.inline.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "os_share_linux.hpp"
39 #include "prims/jniFastGetField.hpp"
40 #include "prims/jvm.h"
41 #include "prims/jvm_misc.hpp"
42 #include "runtime/arguments.hpp"
43 #include "runtime/extendedPC.hpp"
44 #include "runtime/globals.hpp"
45 #include "runtime/interfaceSupport.hpp"
46 #include "runtime/init.hpp"
47 #include "runtime/java.hpp"
48 #include "runtime/javaCalls.hpp"
49 #include "runtime/mutexLocker.hpp"
50 #include "runtime/objectMonitor.hpp"
51 #include "runtime/osThread.hpp"
52 #include "runtime/perfMemory.hpp"
53 #include "runtime/sharedRuntime.hpp"
54 #include "runtime/statSampler.hpp"
55 #include "runtime/stubRoutines.hpp"
56 #include "runtime/threadCritical.hpp"
57 #include "runtime/timer.hpp"
58 #include "services/attachListener.hpp"
59 #include "services/memTracker.hpp"
60 #include "services/runtimeService.hpp"
61 #include "thread_linux.inline.hpp"
62 #include "utilities/decoder.hpp"
63 #include "utilities/defaultStream.hpp"
64 #include "utilities/events.hpp"
65 #include "utilities/elfFile.hpp"
66 #include "utilities/growableArray.hpp"
67 #include "utilities/vmError.hpp"
68 #ifdef TARGET_ARCH_x86
69 # include "assembler_x86.inline.hpp"
70 # include "nativeInst_x86.hpp"
71 #endif
72 #ifdef TARGET_ARCH_sparc
73 # include "assembler_sparc.inline.hpp"
74 # include "nativeInst_sparc.hpp"
75 #endif
76 #ifdef TARGET_ARCH_zero
77 # include "assembler_zero.inline.hpp"
78 # include "nativeInst_zero.hpp"
79 #endif
80 #ifdef TARGET_ARCH_arm
81 # include "assembler_arm.inline.hpp"
82 # include "nativeInst_arm.hpp"
83 #endif
84 #ifdef TARGET_ARCH_ppc
85 # include "assembler_ppc.inline.hpp"
86 # include "nativeInst_ppc.hpp"
87 #endif
88
89 // put OS-includes here
90 # include <sys/types.h>
91 # include <sys/mman.h>
92 # include <sys/stat.h>
93 # include <sys/select.h>
94 # include <pthread.h>
95 # include <signal.h>
96 # include <errno.h>
97 # include <dlfcn.h>
98 # include <stdio.h>
99 # include <unistd.h>
100 # include <sys/resource.h>
101 # include <pthread.h>
102 # include <sys/stat.h>
103 # include <sys/time.h>
104 # include <sys/times.h>
105 # include <sys/utsname.h>
106 # include <sys/socket.h>
107 # include <sys/wait.h>
108 # include <pwd.h>
109 # include <poll.h>
110 # include <semaphore.h>
111 # include <fcntl.h>
112 # include <string.h>
113 # include <syscall.h>
114 # include <sys/sysinfo.h>
115 # include <gnu/libc-version.h>
116 # include <sys/ipc.h>
117 # include <sys/shm.h>
118 # include <link.h>
119 # include <stdint.h>
120 # include <inttypes.h>
121 # include <sys/ioctl.h>
122
123 #define MAX_PATH (2 * K)
124
125 // for timer info max values which include all bits
126 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
127
128 #define LARGEPAGES_BIT (1 << 6)
129 ////////////////////////////////////////////////////////////////////////////////
130 // global variables
131 julong os::Linux::_physical_memory = 0;
132
133 address os::Linux::_initial_thread_stack_bottom = NULL;
134 uintptr_t os::Linux::_initial_thread_stack_size = 0;
135
136 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
137 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
138 Mutex* os::Linux::_createThread_lock = NULL;
139 pthread_t os::Linux::_main_thread;
140 int os::Linux::_page_size = -1;
141 const int os::Linux::_vm_default_page_size = (8 * K);
142 bool os::Linux::_is_floating_stack = false;
143 bool os::Linux::_is_NPTL = false;
144 bool os::Linux::_supports_fast_thread_cpu_time = false;
145 const char * os::Linux::_glibc_version = NULL;
146 const char * os::Linux::_libpthread_version = NULL;
147
148 static jlong initial_time_count=0;
149
150 static int clock_tics_per_sec = 100;
151
152 // For diagnostics to print a message once. see run_periodic_checks
153 static sigset_t check_signal_done;
154 static bool check_signals = true;;
155
156 static pid_t _initial_pid = 0;
157
158 /* Signal number used to suspend/resume a thread */
159
160 /* do not use any signal number less than SIGSEGV, see 4355769 */
161 static int SR_signum = SIGUSR2;
162 sigset_t SR_sigset;
163
164 /* Used to protect dlsym() calls */
165 static pthread_mutex_t dl_mutex;
166
167 // Declarations
168 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
169
170 #ifdef JAVASE_EMBEDDED
171 class MemNotifyThread: public Thread {
172 friend class VMStructs;
173 public:
174 virtual void run();
175
176 private:
177 static MemNotifyThread* _memnotify_thread;
178 int _fd;
179
180 public:
181
182 // Constructor
183 MemNotifyThread(int fd);
184
185 // Tester
186 bool is_memnotify_thread() const { return true; }
187
188 // Printing
189 char* name() const { return (char*)"Linux MemNotify Thread"; }
190
191 // Returns the single instance of the MemNotifyThread
192 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
193
194 // Create and start the single instance of MemNotifyThread
195 static void start();
196 };
197 #endif // JAVASE_EMBEDDED
198
199 // utility functions
200
201 static int SR_initialize();
202 static int SR_finalize();
203
204 julong os::available_memory() {
205 return Linux::available_memory();
206 }
207
208 julong os::Linux::available_memory() {
209 // values in struct sysinfo are "unsigned long"
210 struct sysinfo si;
211 sysinfo(&si);
212
213 return (julong)si.freeram * si.mem_unit;
214 }
215
216 julong os::physical_memory() {
217 return Linux::physical_memory();
218 }
219
220 julong os::allocatable_physical_memory(julong size) {
221 #ifdef _LP64
222 return size;
223 #else
224 julong result = MIN2(size, (julong)3800*M);
225 if (!is_allocatable(result)) {
226 // See comments under solaris for alignment considerations
227 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
228 result = MIN2(size, reasonable_size);
229 }
230 return result;
231 #endif // _LP64
232 }
233
234 ////////////////////////////////////////////////////////////////////////////////
235 // environment support
236
237 bool os::getenv(const char* name, char* buf, int len) {
238 const char* val = ::getenv(name);
239 if (val != NULL && strlen(val) < (size_t)len) {
240 strcpy(buf, val);
241 return true;
242 }
243 if (len > 0) buf[0] = 0; // return a null string
244 return false;
245 }
246
247
248 // Return true if user is running as root.
249
250 bool os::have_special_privileges() {
251 static bool init = false;
252 static bool privileges = false;
253 if (!init) {
254 privileges = (getuid() != geteuid()) || (getgid() != getegid());
255 init = true;
256 }
257 return privileges;
258 }
259
260
261 #ifndef SYS_gettid
262 // i386: 224, ia64: 1105, amd64: 186, sparc 143
263 #ifdef __ia64__
264 #define SYS_gettid 1105
265 #elif __i386__
266 #define SYS_gettid 224
267 #elif __amd64__
268 #define SYS_gettid 186
269 #elif __sparc__
270 #define SYS_gettid 143
271 #else
272 #error define gettid for the arch
273 #endif
274 #endif
275
276 // Cpu architecture string
277 #if defined(ZERO)
278 static char cpu_arch[] = ZERO_LIBARCH;
279 #elif defined(IA64)
280 static char cpu_arch[] = "ia64";
281 #elif defined(IA32)
282 static char cpu_arch[] = "i386";
283 #elif defined(AMD64)
284 static char cpu_arch[] = "amd64";
285 #elif defined(ARM)
286 static char cpu_arch[] = "arm";
287 #elif defined(PPC)
288 static char cpu_arch[] = "ppc";
289 #elif defined(SPARC)
290 # ifdef _LP64
291 static char cpu_arch[] = "sparcv9";
292 # else
293 static char cpu_arch[] = "sparc";
294 # endif
295 #else
296 #error Add appropriate cpu_arch setting
297 #endif
298
299
300 // pid_t gettid()
301 //
302 // Returns the kernel thread id of the currently running thread. Kernel
303 // thread id is used to access /proc.
304 //
305 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
306 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
307 //
308 pid_t os::Linux::gettid() {
309 int rslt = syscall(SYS_gettid);
310 if (rslt == -1) {
311 // old kernel, no NPTL support
312 return getpid();
313 } else {
314 return (pid_t)rslt;
315 }
316 }
317
318 // Most versions of linux have a bug where the number of processors are
319 // determined by looking at the /proc file system. In a chroot environment,
320 // the system call returns 1. This causes the VM to act as if it is
321 // a single processor and elide locking (see is_MP() call).
322 static bool unsafe_chroot_detected = false;
323 static const char *unstable_chroot_error = "/proc file system not found.\n"
324 "Java may be unstable running multithreaded in a chroot "
325 "environment on Linux when /proc filesystem is not mounted.";
326
327 void os::Linux::initialize_system_info() {
328 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
329 if (processor_count() == 1) {
330 pid_t pid = os::Linux::gettid();
331 char fname[32];
332 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
333 FILE *fp = fopen(fname, "r");
334 if (fp == NULL) {
335 unsafe_chroot_detected = true;
336 } else {
337 fclose(fp);
338 }
339 }
340 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
341 assert(processor_count() > 0, "linux error");
342 }
343
344 void os::init_system_properties_values() {
345 // char arch[12];
346 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
347
348 // The next steps are taken in the product version:
349 //
350 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
351 // This library should be located at:
352 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
353 //
354 // If "/jre/lib/" appears at the right place in the path, then we
355 // assume libjvm[_g].so is installed in a JDK and we use this path.
356 //
357 // Otherwise exit with message: "Could not create the Java virtual machine."
358 //
359 // The following extra steps are taken in the debugging version:
360 //
361 // If "/jre/lib/" does NOT appear at the right place in the path
362 // instead of exit check for $JAVA_HOME environment variable.
363 //
364 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
365 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
366 // it looks like libjvm[_g].so is installed there
367 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
368 //
369 // Otherwise exit.
370 //
371 // Important note: if the location of libjvm.so changes this
372 // code needs to be changed accordingly.
373
374 // The next few definitions allow the code to be verbatim:
375 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
376 #define getenv(n) ::getenv(n)
377
378 /*
379 * See ld(1):
380 * The linker uses the following search paths to locate required
381 * shared libraries:
382 * 1: ...
383 * ...
384 * 7: The default directories, normally /lib and /usr/lib.
385 */
386 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
387 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
388 #else
389 #define DEFAULT_LIBPATH "/lib:/usr/lib"
390 #endif
391
392 #define EXTENSIONS_DIR "/lib/ext"
393 #define ENDORSED_DIR "/lib/endorsed"
394 #define REG_DIR "/usr/java/packages"
395
396 {
397 /* sysclasspath, java_home, dll_dir */
398 {
399 char *home_path;
400 char *dll_path;
401 char *pslash;
402 char buf[MAXPATHLEN];
403 os::jvm_path(buf, sizeof(buf));
404
405 // Found the full path to libjvm.so.
406 // Now cut the path to <java_home>/jre if we can.
407 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
408 pslash = strrchr(buf, '/');
409 if (pslash != NULL)
410 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
411 dll_path = malloc(strlen(buf) + 1);
412 if (dll_path == NULL)
413 return;
414 strcpy(dll_path, buf);
415 Arguments::set_dll_dir(dll_path);
416
417 if (pslash != NULL) {
418 pslash = strrchr(buf, '/');
419 if (pslash != NULL) {
420 *pslash = '\0'; /* get rid of /<arch> */
421 pslash = strrchr(buf, '/');
422 if (pslash != NULL)
423 *pslash = '\0'; /* get rid of /lib */
424 }
425 }
426
427 home_path = malloc(strlen(buf) + 1);
428 if (home_path == NULL)
429 return;
430 strcpy(home_path, buf);
431 Arguments::set_java_home(home_path);
432
433 if (!set_boot_path('/', ':'))
434 return;
435 }
436
437 /*
438 * Where to look for native libraries
439 *
440 * Note: Due to a legacy implementation, most of the library path
441 * is set in the launcher. This was to accomodate linking restrictions
442 * on legacy Linux implementations (which are no longer supported).
443 * Eventually, all the library path setting will be done here.
444 *
445 * However, to prevent the proliferation of improperly built native
446 * libraries, the new path component /usr/java/packages is added here.
447 * Eventually, all the library path setting will be done here.
448 */
449 {
450 char *ld_library_path;
451
452 /*
453 * Construct the invariant part of ld_library_path. Note that the
454 * space for the colon and the trailing null are provided by the
455 * nulls included by the sizeof operator (so actually we allocate
456 * a byte more than necessary).
457 */
458 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
459 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
460 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
461
462 /*
463 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
464 * should always exist (until the legacy problem cited above is
465 * addressed).
466 */
467 char *v = getenv("LD_LIBRARY_PATH");
468 if (v != NULL) {
469 char *t = ld_library_path;
470 /* That's +1 for the colon and +1 for the trailing '\0' */
471 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
472 sprintf(ld_library_path, "%s:%s", v, t);
473 }
474 Arguments::set_library_path(ld_library_path);
475 }
476
477 /*
478 * Extensions directories.
479 *
480 * Note that the space for the colon and the trailing null are provided
481 * by the nulls included by the sizeof operator (so actually one byte more
482 * than necessary is allocated).
483 */
484 {
485 char *buf = malloc(strlen(Arguments::get_java_home()) +
486 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
487 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
488 Arguments::get_java_home());
489 Arguments::set_ext_dirs(buf);
490 }
491
492 /* Endorsed standards default directory. */
493 {
494 char * buf;
495 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
496 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
497 Arguments::set_endorsed_dirs(buf);
498 }
499 }
500
501 #undef malloc
502 #undef getenv
503 #undef EXTENSIONS_DIR
504 #undef ENDORSED_DIR
505
506 // Done
507 return;
508 }
509
510 ////////////////////////////////////////////////////////////////////////////////
511 // breakpoint support
512
513 void os::breakpoint() {
514 BREAKPOINT;
515 }
516
517 extern "C" void breakpoint() {
518 // use debugger to set breakpoint here
519 }
520
521 ////////////////////////////////////////////////////////////////////////////////
522 // signal support
523
524 debug_only(static bool signal_sets_initialized = false);
525 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
526
527 bool os::Linux::is_sig_ignored(int sig) {
528 struct sigaction oact;
529 sigaction(sig, (struct sigaction*)NULL, &oact);
530 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
531 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
532 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
533 return true;
534 else
535 return false;
536 }
537
538 void os::Linux::signal_sets_init() {
539 // Should also have an assertion stating we are still single-threaded.
540 assert(!signal_sets_initialized, "Already initialized");
541 // Fill in signals that are necessarily unblocked for all threads in
542 // the VM. Currently, we unblock the following signals:
543 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
544 // by -Xrs (=ReduceSignalUsage));
545 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
546 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
547 // the dispositions or masks wrt these signals.
548 // Programs embedding the VM that want to use the above signals for their
549 // own purposes must, at this time, use the "-Xrs" option to prevent
550 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
551 // (See bug 4345157, and other related bugs).
552 // In reality, though, unblocking these signals is really a nop, since
553 // these signals are not blocked by default.
554 sigemptyset(&unblocked_sigs);
555 sigemptyset(&allowdebug_blocked_sigs);
556 sigaddset(&unblocked_sigs, SIGILL);
557 sigaddset(&unblocked_sigs, SIGSEGV);
558 sigaddset(&unblocked_sigs, SIGBUS);
559 sigaddset(&unblocked_sigs, SIGFPE);
560 sigaddset(&unblocked_sigs, SR_signum);
561
562 if (!ReduceSignalUsage) {
563 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
564 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
565 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
566 }
567 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
568 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
569 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
570 }
571 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
572 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
573 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
574 }
575 }
576 // Fill in signals that are blocked by all but the VM thread.
577 sigemptyset(&vm_sigs);
578 if (!ReduceSignalUsage)
579 sigaddset(&vm_sigs, BREAK_SIGNAL);
580 debug_only(signal_sets_initialized = true);
581
582 }
583
584 // These are signals that are unblocked while a thread is running Java.
585 // (For some reason, they get blocked by default.)
586 sigset_t* os::Linux::unblocked_signals() {
587 assert(signal_sets_initialized, "Not initialized");
588 return &unblocked_sigs;
589 }
590
591 // These are the signals that are blocked while a (non-VM) thread is
592 // running Java. Only the VM thread handles these signals.
593 sigset_t* os::Linux::vm_signals() {
594 assert(signal_sets_initialized, "Not initialized");
595 return &vm_sigs;
596 }
597
598 // These are signals that are blocked during cond_wait to allow debugger in
599 sigset_t* os::Linux::allowdebug_blocked_signals() {
600 assert(signal_sets_initialized, "Not initialized");
601 return &allowdebug_blocked_sigs;
602 }
603
604 void os::Linux::hotspot_sigmask(Thread* thread) {
605
606 //Save caller's signal mask before setting VM signal mask
607 sigset_t caller_sigmask;
608 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
609
610 OSThread* osthread = thread->osthread();
611 osthread->set_caller_sigmask(caller_sigmask);
612
613 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
614
615 if (!ReduceSignalUsage) {
616 if (thread->is_VM_thread()) {
617 // Only the VM thread handles BREAK_SIGNAL ...
618 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
619 } else {
620 // ... all other threads block BREAK_SIGNAL
621 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
622 }
623 }
624 }
625
626 //////////////////////////////////////////////////////////////////////////////
627 // detecting pthread library
628
629 void os::Linux::libpthread_init() {
630 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
631 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
632 // generic name for earlier versions.
633 // Define macros here so we can build HotSpot on old systems.
634 # ifndef _CS_GNU_LIBC_VERSION
635 # define _CS_GNU_LIBC_VERSION 2
636 # endif
637 # ifndef _CS_GNU_LIBPTHREAD_VERSION
638 # define _CS_GNU_LIBPTHREAD_VERSION 3
639 # endif
640
641 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
642 if (n > 0) {
643 char *str = (char *)malloc(n, mtInternal);
644 confstr(_CS_GNU_LIBC_VERSION, str, n);
645 os::Linux::set_glibc_version(str);
646 } else {
647 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
648 static char _gnu_libc_version[32];
649 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
650 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
651 os::Linux::set_glibc_version(_gnu_libc_version);
652 }
653
654 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
655 if (n > 0) {
656 char *str = (char *)malloc(n, mtInternal);
657 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
658 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
659 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
660 // is the case. LinuxThreads has a hard limit on max number of threads.
661 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
662 // On the other hand, NPTL does not have such a limit, sysconf()
663 // will return -1 and errno is not changed. Check if it is really NPTL.
664 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
665 strstr(str, "NPTL") &&
666 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
667 free(str);
668 os::Linux::set_libpthread_version("linuxthreads");
669 } else {
670 os::Linux::set_libpthread_version(str);
671 }
672 } else {
673 // glibc before 2.3.2 only has LinuxThreads.
674 os::Linux::set_libpthread_version("linuxthreads");
675 }
676
677 if (strstr(libpthread_version(), "NPTL")) {
678 os::Linux::set_is_NPTL();
679 } else {
680 os::Linux::set_is_LinuxThreads();
681 }
682
683 // LinuxThreads have two flavors: floating-stack mode, which allows variable
684 // stack size; and fixed-stack mode. NPTL is always floating-stack.
685 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
686 os::Linux::set_is_floating_stack();
687 }
688 }
689
690 /////////////////////////////////////////////////////////////////////////////
691 // thread stack
692
693 // Force Linux kernel to expand current thread stack. If "bottom" is close
694 // to the stack guard, caller should block all signals.
695 //
696 // MAP_GROWSDOWN:
697 // A special mmap() flag that is used to implement thread stacks. It tells
698 // kernel that the memory region should extend downwards when needed. This
699 // allows early versions of LinuxThreads to only mmap the first few pages
700 // when creating a new thread. Linux kernel will automatically expand thread
701 // stack as needed (on page faults).
702 //
703 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
704 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
705 // region, it's hard to tell if the fault is due to a legitimate stack
706 // access or because of reading/writing non-exist memory (e.g. buffer
707 // overrun). As a rule, if the fault happens below current stack pointer,
708 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
709 // application (see Linux kernel fault.c).
710 //
711 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
712 // stack overflow detection.
713 //
714 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
715 // not use this flag. However, the stack of initial thread is not created
716 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
717 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
718 // and then attach the thread to JVM.
719 //
720 // To get around the problem and allow stack banging on Linux, we need to
721 // manually expand thread stack after receiving the SIGSEGV.
722 //
723 // There are two ways to expand thread stack to address "bottom", we used
724 // both of them in JVM before 1.5:
725 // 1. adjust stack pointer first so that it is below "bottom", and then
726 // touch "bottom"
727 // 2. mmap() the page in question
728 //
729 // Now alternate signal stack is gone, it's harder to use 2. For instance,
730 // if current sp is already near the lower end of page 101, and we need to
731 // call mmap() to map page 100, it is possible that part of the mmap() frame
732 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
733 // That will destroy the mmap() frame and cause VM to crash.
734 //
735 // The following code works by adjusting sp first, then accessing the "bottom"
736 // page to force a page fault. Linux kernel will then automatically expand the
737 // stack mapping.
738 //
739 // _expand_stack_to() assumes its frame size is less than page size, which
740 // should always be true if the function is not inlined.
741
742 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
743 #define NOINLINE
744 #else
745 #define NOINLINE __attribute__ ((noinline))
746 #endif
747
748 static void _expand_stack_to(address bottom) NOINLINE;
749
750 static void _expand_stack_to(address bottom) {
751 address sp;
752 size_t size;
753 volatile char *p;
754
755 // Adjust bottom to point to the largest address within the same page, it
756 // gives us a one-page buffer if alloca() allocates slightly more memory.
757 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
758 bottom += os::Linux::page_size() - 1;
759
760 // sp might be slightly above current stack pointer; if that's the case, we
761 // will alloca() a little more space than necessary, which is OK. Don't use
762 // os::current_stack_pointer(), as its result can be slightly below current
763 // stack pointer, causing us to not alloca enough to reach "bottom".
764 sp = (address)&sp;
765
766 if (sp > bottom) {
767 size = sp - bottom;
768 p = (volatile char *)alloca(size);
769 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
770 p[0] = '\0';
771 }
772 }
773
774 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
775 assert(t!=NULL, "just checking");
776 assert(t->osthread()->expanding_stack(), "expand should be set");
777 assert(t->stack_base() != NULL, "stack_base was not initialized");
778
779 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
780 sigset_t mask_all, old_sigset;
781 sigfillset(&mask_all);
782 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
783 _expand_stack_to(addr);
784 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
785 return true;
786 }
787 return false;
788 }
789
790 //////////////////////////////////////////////////////////////////////////////
791 // create new thread
792
793 static address highest_vm_reserved_address();
794
795 // check if it's safe to start a new thread
796 static bool _thread_safety_check(Thread* thread) {
797 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
798 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
799 // Heap is mmap'ed at lower end of memory space. Thread stacks are
800 // allocated (MAP_FIXED) from high address space. Every thread stack
801 // occupies a fixed size slot (usually 2Mbytes, but user can change
802 // it to other values if they rebuild LinuxThreads).
803 //
804 // Problem with MAP_FIXED is that mmap() can still succeed even part of
805 // the memory region has already been mmap'ed. That means if we have too
806 // many threads and/or very large heap, eventually thread stack will
807 // collide with heap.
808 //
809 // Here we try to prevent heap/stack collision by comparing current
810 // stack bottom with the highest address that has been mmap'ed by JVM
811 // plus a safety margin for memory maps created by native code.
812 //
813 // This feature can be disabled by setting ThreadSafetyMargin to 0
814 //
815 if (ThreadSafetyMargin > 0) {
816 address stack_bottom = os::current_stack_base() - os::current_stack_size();
817
818 // not safe if our stack extends below the safety margin
819 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
820 } else {
821 return true;
822 }
823 } else {
824 // Floating stack LinuxThreads or NPTL:
825 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
826 // there's not enough space left, pthread_create() will fail. If we come
827 // here, that means enough space has been reserved for stack.
828 return true;
829 }
830 }
831
832 // Thread start routine for all newly created threads
833 static void *java_start(Thread *thread) {
834 // Try to randomize the cache line index of hot stack frames.
835 // This helps when threads of the same stack traces evict each other's
836 // cache lines. The threads can be either from the same JVM instance, or
837 // from different JVM instances. The benefit is especially true for
838 // processors with hyperthreading technology.
839 static int counter = 0;
840 int pid = os::current_process_id();
841 alloca(((pid ^ counter++) & 7) * 128);
842
843 ThreadLocalStorage::set_thread(thread);
844
845 OSThread* osthread = thread->osthread();
846 Monitor* sync = osthread->startThread_lock();
847
848 // non floating stack LinuxThreads needs extra check, see above
849 if (!_thread_safety_check(thread)) {
850 // notify parent thread
851 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
852 osthread->set_state(ZOMBIE);
853 sync->notify_all();
854 return NULL;
855 }
856
857 // thread_id is kernel thread id (similar to Solaris LWP id)
858 osthread->set_thread_id(os::Linux::gettid());
859
860 if (UseNUMA) {
861 int lgrp_id = os::numa_get_group_id();
862 if (lgrp_id != -1) {
863 thread->set_lgrp_id(lgrp_id);
864 }
865 }
866 // initialize signal mask for this thread
867 os::Linux::hotspot_sigmask(thread);
868
869 // initialize floating point control register
870 os::Linux::init_thread_fpu_state();
871
872 // handshaking with parent thread
873 {
874 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
875
876 // notify parent thread
877 osthread->set_state(INITIALIZED);
878 sync->notify_all();
879
880 // wait until os::start_thread()
881 while (osthread->get_state() == INITIALIZED) {
882 sync->wait(Mutex::_no_safepoint_check_flag);
883 }
884 }
885
886 // call one more level start routine
887 thread->run();
888
889 return 0;
890 }
891
892 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
893 assert(thread->osthread() == NULL, "caller responsible");
894
895 // Allocate the OSThread object
896 OSThread* osthread = new OSThread(NULL, NULL);
897 if (osthread == NULL) {
898 return false;
899 }
900
901 // set the correct thread state
902 osthread->set_thread_type(thr_type);
903
904 // Initial state is ALLOCATED but not INITIALIZED
905 osthread->set_state(ALLOCATED);
906
907 thread->set_osthread(osthread);
908
909 // init thread attributes
910 pthread_attr_t attr;
911 pthread_attr_init(&attr);
912 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
913
914 // stack size
915 if (os::Linux::supports_variable_stack_size()) {
916 // calculate stack size if it's not specified by caller
917 if (stack_size == 0) {
918 stack_size = os::Linux::default_stack_size(thr_type);
919
920 switch (thr_type) {
921 case os::java_thread:
922 // Java threads use ThreadStackSize which default value can be
923 // changed with the flag -Xss
924 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
925 stack_size = JavaThread::stack_size_at_create();
926 break;
927 case os::compiler_thread:
928 if (CompilerThreadStackSize > 0) {
929 stack_size = (size_t)(CompilerThreadStackSize * K);
930 break;
931 } // else fall through:
932 // use VMThreadStackSize if CompilerThreadStackSize is not defined
933 case os::vm_thread:
934 case os::pgc_thread:
935 case os::cgc_thread:
936 case os::watcher_thread:
937 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
938 break;
939 }
940 }
941
942 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
943 pthread_attr_setstacksize(&attr, stack_size);
944 } else {
945 // let pthread_create() pick the default value.
946 }
947
948 // glibc guard page
949 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
950
951 ThreadState state;
952
953 {
954 // Serialize thread creation if we are running with fixed stack LinuxThreads
955 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
956 if (lock) {
957 os::Linux::createThread_lock()->lock_without_safepoint_check();
958 }
959
960 pthread_t tid;
961 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
962
963 pthread_attr_destroy(&attr);
964
965 if (ret != 0) {
966 if (PrintMiscellaneous && (Verbose || WizardMode)) {
967 perror("pthread_create()");
968 }
969 // Need to clean up stuff we've allocated so far
970 thread->set_osthread(NULL);
971 delete osthread;
972 if (lock) os::Linux::createThread_lock()->unlock();
973 return false;
974 }
975
976 // Store pthread info into the OSThread
977 osthread->set_pthread_id(tid);
978
979 // Wait until child thread is either initialized or aborted
980 {
981 Monitor* sync_with_child = osthread->startThread_lock();
982 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
983 while ((state = osthread->get_state()) == ALLOCATED) {
984 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
985 }
986 }
987
988 if (lock) {
989 os::Linux::createThread_lock()->unlock();
990 }
991 }
992
993 // Aborted due to thread limit being reached
994 if (state == ZOMBIE) {
995 thread->set_osthread(NULL);
996 delete osthread;
997 return false;
998 }
999
1000 // The thread is returned suspended (in state INITIALIZED),
1001 // and is started higher up in the call chain
1002 assert(state == INITIALIZED, "race condition");
1003 return true;
1004 }
1005
1006 /////////////////////////////////////////////////////////////////////////////
1007 // attach existing thread
1008
1009 // bootstrap the main thread
1010 bool os::create_main_thread(JavaThread* thread) {
1011 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1012 return create_attached_thread(thread);
1013 }
1014
1015 bool os::create_attached_thread(JavaThread* thread) {
1016 #ifdef ASSERT
1017 thread->verify_not_published();
1018 #endif
1019
1020 // Allocate the OSThread object
1021 OSThread* osthread = new OSThread(NULL, NULL);
1022
1023 if (osthread == NULL) {
1024 return false;
1025 }
1026
1027 // Store pthread info into the OSThread
1028 osthread->set_thread_id(os::Linux::gettid());
1029 osthread->set_pthread_id(::pthread_self());
1030
1031 // initialize floating point control register
1032 os::Linux::init_thread_fpu_state();
1033
1034 // Initial thread state is RUNNABLE
1035 osthread->set_state(RUNNABLE);
1036
1037 thread->set_osthread(osthread);
1038
1039 if (UseNUMA) {
1040 int lgrp_id = os::numa_get_group_id();
1041 if (lgrp_id != -1) {
1042 thread->set_lgrp_id(lgrp_id);
1043 }
1044 }
1045
1046 if (os::Linux::is_initial_thread()) {
1047 // If current thread is initial thread, its stack is mapped on demand,
1048 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1049 // the entire stack region to avoid SEGV in stack banging.
1050 // It is also useful to get around the heap-stack-gap problem on SuSE
1051 // kernel (see 4821821 for details). We first expand stack to the top
1052 // of yellow zone, then enable stack yellow zone (order is significant,
1053 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1054 // is no gap between the last two virtual memory regions.
1055
1056 JavaThread *jt = (JavaThread *)thread;
1057 address addr = jt->stack_yellow_zone_base();
1058 assert(addr != NULL, "initialization problem?");
1059 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1060
1061 osthread->set_expanding_stack();
1062 os::Linux::manually_expand_stack(jt, addr);
1063 osthread->clear_expanding_stack();
1064 }
1065
1066 // initialize signal mask for this thread
1067 // and save the caller's signal mask
1068 os::Linux::hotspot_sigmask(thread);
1069
1070 return true;
1071 }
1072
1073 void os::pd_start_thread(Thread* thread) {
1074 OSThread * osthread = thread->osthread();
1075 assert(osthread->get_state() != INITIALIZED, "just checking");
1076 Monitor* sync_with_child = osthread->startThread_lock();
1077 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1078 sync_with_child->notify();
1079 }
1080
1081 // Free Linux resources related to the OSThread
1082 void os::free_thread(OSThread* osthread) {
1083 assert(osthread != NULL, "osthread not set");
1084
1085 if (Thread::current()->osthread() == osthread) {
1086 // Restore caller's signal mask
1087 sigset_t sigmask = osthread->caller_sigmask();
1088 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1089 }
1090
1091 delete osthread;
1092 }
1093
1094 //////////////////////////////////////////////////////////////////////////////
1095 // thread local storage
1096
1097 int os::allocate_thread_local_storage() {
1098 pthread_key_t key;
1099 int rslt = pthread_key_create(&key, NULL);
1100 assert(rslt == 0, "cannot allocate thread local storage");
1101 return (int)key;
1102 }
1103
1104 // Note: This is currently not used by VM, as we don't destroy TLS key
1105 // on VM exit.
1106 void os::free_thread_local_storage(int index) {
1107 int rslt = pthread_key_delete((pthread_key_t)index);
1108 assert(rslt == 0, "invalid index");
1109 }
1110
1111 void os::thread_local_storage_at_put(int index, void* value) {
1112 int rslt = pthread_setspecific((pthread_key_t)index, value);
1113 assert(rslt == 0, "pthread_setspecific failed");
1114 }
1115
1116 extern "C" Thread* get_thread() {
1117 return ThreadLocalStorage::thread();
1118 }
1119
1120 //////////////////////////////////////////////////////////////////////////////
1121 // initial thread
1122
1123 // Check if current thread is the initial thread, similar to Solaris thr_main.
1124 bool os::Linux::is_initial_thread(void) {
1125 char dummy;
1126 // If called before init complete, thread stack bottom will be null.
1127 // Can be called if fatal error occurs before initialization.
1128 if (initial_thread_stack_bottom() == NULL) return false;
1129 assert(initial_thread_stack_bottom() != NULL &&
1130 initial_thread_stack_size() != 0,
1131 "os::init did not locate initial thread's stack region");
1132 if ((address)&dummy >= initial_thread_stack_bottom() &&
1133 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1134 return true;
1135 else return false;
1136 }
1137
1138 // Find the virtual memory area that contains addr
1139 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1140 FILE *fp = fopen("/proc/self/maps", "r");
1141 if (fp) {
1142 address low, high;
1143 while (!feof(fp)) {
1144 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1145 if (low <= addr && addr < high) {
1146 if (vma_low) *vma_low = low;
1147 if (vma_high) *vma_high = high;
1148 fclose (fp);
1149 return true;
1150 }
1151 }
1152 for (;;) {
1153 int ch = fgetc(fp);
1154 if (ch == EOF || ch == (int)'\n') break;
1155 }
1156 }
1157 fclose(fp);
1158 }
1159 return false;
1160 }
1161
1162 // Locate initial thread stack. This special handling of initial thread stack
1163 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1164 // bogus value for initial thread.
1165 void os::Linux::capture_initial_stack(size_t max_size) {
1166 // stack size is the easy part, get it from RLIMIT_STACK
1167 size_t stack_size;
1168 struct rlimit rlim;
1169 getrlimit(RLIMIT_STACK, &rlim);
1170 stack_size = rlim.rlim_cur;
1171
1172 // 6308388: a bug in ld.so will relocate its own .data section to the
1173 // lower end of primordial stack; reduce ulimit -s value a little bit
1174 // so we won't install guard page on ld.so's data section.
1175 stack_size -= 2 * page_size();
1176
1177 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1178 // 7.1, in both cases we will get 2G in return value.
1179 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1180 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1181 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1182 // in case other parts in glibc still assumes 2M max stack size.
1183 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1184 #ifndef IA64
1185 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1186 #else
1187 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1188 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1189 #endif
1190
1191 // Try to figure out where the stack base (top) is. This is harder.
1192 //
1193 // When an application is started, glibc saves the initial stack pointer in
1194 // a global variable "__libc_stack_end", which is then used by system
1195 // libraries. __libc_stack_end should be pretty close to stack top. The
1196 // variable is available since the very early days. However, because it is
1197 // a private interface, it could disappear in the future.
1198 //
1199 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1200 // to __libc_stack_end, it is very close to stack top, but isn't the real
1201 // stack top. Note that /proc may not exist if VM is running as a chroot
1202 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1203 // /proc/<pid>/stat could change in the future (though unlikely).
1204 //
1205 // We try __libc_stack_end first. If that doesn't work, look for
1206 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1207 // as a hint, which should work well in most cases.
1208
1209 uintptr_t stack_start;
1210
1211 // try __libc_stack_end first
1212 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1213 if (p && *p) {
1214 stack_start = *p;
1215 } else {
1216 // see if we can get the start_stack field from /proc/self/stat
1217 FILE *fp;
1218 int pid;
1219 char state;
1220 int ppid;
1221 int pgrp;
1222 int session;
1223 int nr;
1224 int tpgrp;
1225 unsigned long flags;
1226 unsigned long minflt;
1227 unsigned long cminflt;
1228 unsigned long majflt;
1229 unsigned long cmajflt;
1230 unsigned long utime;
1231 unsigned long stime;
1232 long cutime;
1233 long cstime;
1234 long prio;
1235 long nice;
1236 long junk;
1237 long it_real;
1238 uintptr_t start;
1239 uintptr_t vsize;
1240 intptr_t rss;
1241 uintptr_t rsslim;
1242 uintptr_t scodes;
1243 uintptr_t ecode;
1244 int i;
1245
1246 // Figure what the primordial thread stack base is. Code is inspired
1247 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1248 // followed by command name surrounded by parentheses, state, etc.
1249 char stat[2048];
1250 int statlen;
1251
1252 fp = fopen("/proc/self/stat", "r");
1253 if (fp) {
1254 statlen = fread(stat, 1, 2047, fp);
1255 stat[statlen] = '\0';
1256 fclose(fp);
1257
1258 // Skip pid and the command string. Note that we could be dealing with
1259 // weird command names, e.g. user could decide to rename java launcher
1260 // to "java 1.4.2 :)", then the stat file would look like
1261 // 1234 (java 1.4.2 :)) R ... ...
1262 // We don't really need to know the command string, just find the last
1263 // occurrence of ")" and then start parsing from there. See bug 4726580.
1264 char * s = strrchr(stat, ')');
1265
1266 i = 0;
1267 if (s) {
1268 // Skip blank chars
1269 do s++; while (isspace(*s));
1270
1271 #define _UFM UINTX_FORMAT
1272 #define _DFM INTX_FORMAT
1273
1274 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1275 /* 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 */
1276 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,
1277 &state, /* 3 %c */
1278 &ppid, /* 4 %d */
1279 &pgrp, /* 5 %d */
1280 &session, /* 6 %d */
1281 &nr, /* 7 %d */
1282 &tpgrp, /* 8 %d */
1283 &flags, /* 9 %lu */
1284 &minflt, /* 10 %lu */
1285 &cminflt, /* 11 %lu */
1286 &majflt, /* 12 %lu */
1287 &cmajflt, /* 13 %lu */
1288 &utime, /* 14 %lu */
1289 &stime, /* 15 %lu */
1290 &cutime, /* 16 %ld */
1291 &cstime, /* 17 %ld */
1292 &prio, /* 18 %ld */
1293 &nice, /* 19 %ld */
1294 &junk, /* 20 %ld */
1295 &it_real, /* 21 %ld */
1296 &start, /* 22 UINTX_FORMAT */
1297 &vsize, /* 23 UINTX_FORMAT */
1298 &rss, /* 24 INTX_FORMAT */
1299 &rsslim, /* 25 UINTX_FORMAT */
1300 &scodes, /* 26 UINTX_FORMAT */
1301 &ecode, /* 27 UINTX_FORMAT */
1302 &stack_start); /* 28 UINTX_FORMAT */
1303 }
1304
1305 #undef _UFM
1306 #undef _DFM
1307
1308 if (i != 28 - 2) {
1309 assert(false, "Bad conversion from /proc/self/stat");
1310 // product mode - assume we are the initial thread, good luck in the
1311 // embedded case.
1312 warning("Can't detect initial thread stack location - bad conversion");
1313 stack_start = (uintptr_t) &rlim;
1314 }
1315 } else {
1316 // For some reason we can't open /proc/self/stat (for example, running on
1317 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1318 // most cases, so don't abort:
1319 warning("Can't detect initial thread stack location - no /proc/self/stat");
1320 stack_start = (uintptr_t) &rlim;
1321 }
1322 }
1323
1324 // Now we have a pointer (stack_start) very close to the stack top, the
1325 // next thing to do is to figure out the exact location of stack top. We
1326 // can find out the virtual memory area that contains stack_start by
1327 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1328 // and its upper limit is the real stack top. (again, this would fail if
1329 // running inside chroot, because /proc may not exist.)
1330
1331 uintptr_t stack_top;
1332 address low, high;
1333 if (find_vma((address)stack_start, &low, &high)) {
1334 // success, "high" is the true stack top. (ignore "low", because initial
1335 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1336 stack_top = (uintptr_t)high;
1337 } else {
1338 // failed, likely because /proc/self/maps does not exist
1339 warning("Can't detect initial thread stack location - find_vma failed");
1340 // best effort: stack_start is normally within a few pages below the real
1341 // stack top, use it as stack top, and reduce stack size so we won't put
1342 // guard page outside stack.
1343 stack_top = stack_start;
1344 stack_size -= 16 * page_size();
1345 }
1346
1347 // stack_top could be partially down the page so align it
1348 stack_top = align_size_up(stack_top, page_size());
1349
1350 if (max_size && stack_size > max_size) {
1351 _initial_thread_stack_size = max_size;
1352 } else {
1353 _initial_thread_stack_size = stack_size;
1354 }
1355
1356 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1357 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1358 }
1359
1360 ////////////////////////////////////////////////////////////////////////////////
1361 // time support
1362
1363 // Time since start-up in seconds to a fine granularity.
1364 // Used by VMSelfDestructTimer and the MemProfiler.
1365 double os::elapsedTime() {
1366
1367 return (double)(os::elapsed_counter()) * 0.000001;
1368 }
1369
1370 jlong os::elapsed_counter() {
1371 timeval time;
1372 int status = gettimeofday(&time, NULL);
1373 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1374 }
1375
1376 jlong os::elapsed_frequency() {
1377 return (1000 * 1000);
1378 }
1379
1380 // For now, we say that linux does not support vtime. I have no idea
1381 // whether it can actually be made to (DLD, 9/13/05).
1382
1383 bool os::supports_vtime() { return false; }
1384 bool os::enable_vtime() { return false; }
1385 bool os::vtime_enabled() { return false; }
1386 double os::elapsedVTime() {
1387 // better than nothing, but not much
1388 return elapsedTime();
1389 }
1390
1391 jlong os::javaTimeMillis() {
1392 timeval time;
1393 int status = gettimeofday(&time, NULL);
1394 assert(status != -1, "linux error");
1395 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1396 }
1397
1398 #ifndef CLOCK_MONOTONIC
1399 #define CLOCK_MONOTONIC (1)
1400 #endif
1401
1402 void os::Linux::clock_init() {
1403 // we do dlopen's in this particular order due to bug in linux
1404 // dynamical loader (see 6348968) leading to crash on exit
1405 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1406 if (handle == NULL) {
1407 handle = dlopen("librt.so", RTLD_LAZY);
1408 }
1409
1410 if (handle) {
1411 int (*clock_getres_func)(clockid_t, struct timespec*) =
1412 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1413 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1414 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1415 if (clock_getres_func && clock_gettime_func) {
1416 // See if monotonic clock is supported by the kernel. Note that some
1417 // early implementations simply return kernel jiffies (updated every
1418 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1419 // for nano time (though the monotonic property is still nice to have).
1420 // It's fixed in newer kernels, however clock_getres() still returns
1421 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1422 // resolution for now. Hopefully as people move to new kernels, this
1423 // won't be a problem.
1424 struct timespec res;
1425 struct timespec tp;
1426 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1427 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1428 // yes, monotonic clock is supported
1429 _clock_gettime = clock_gettime_func;
1430 } else {
1431 // close librt if there is no monotonic clock
1432 dlclose(handle);
1433 }
1434 }
1435 }
1436 }
1437
1438 #ifndef SYS_clock_getres
1439
1440 #if defined(IA32) || defined(AMD64)
1441 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1442 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1443 #else
1444 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1445 #define sys_clock_getres(x,y) -1
1446 #endif
1447
1448 #else
1449 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1450 #endif
1451
1452 void os::Linux::fast_thread_clock_init() {
1453 if (!UseLinuxPosixThreadCPUClocks) {
1454 return;
1455 }
1456 clockid_t clockid;
1457 struct timespec tp;
1458 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1459 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1460
1461 // Switch to using fast clocks for thread cpu time if
1462 // the sys_clock_getres() returns 0 error code.
1463 // Note, that some kernels may support the current thread
1464 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1465 // returned by the pthread_getcpuclockid().
1466 // If the fast Posix clocks are supported then the sys_clock_getres()
1467 // must return at least tp.tv_sec == 0 which means a resolution
1468 // better than 1 sec. This is extra check for reliability.
1469
1470 if(pthread_getcpuclockid_func &&
1471 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1472 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1473
1474 _supports_fast_thread_cpu_time = true;
1475 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1476 }
1477 }
1478
1479 jlong os::javaTimeNanos() {
1480 if (Linux::supports_monotonic_clock()) {
1481 struct timespec tp;
1482 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1483 assert(status == 0, "gettime error");
1484 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1485 return result;
1486 } else {
1487 timeval time;
1488 int status = gettimeofday(&time, NULL);
1489 assert(status != -1, "linux error");
1490 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1491 return 1000 * usecs;
1492 }
1493 }
1494
1495 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1496 if (Linux::supports_monotonic_clock()) {
1497 info_ptr->max_value = ALL_64_BITS;
1498
1499 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1500 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1501 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1502 } else {
1503 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1504 info_ptr->max_value = ALL_64_BITS;
1505
1506 // gettimeofday is a real time clock so it skips
1507 info_ptr->may_skip_backward = true;
1508 info_ptr->may_skip_forward = true;
1509 }
1510
1511 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1512 }
1513
1514 // Return the real, user, and system times in seconds from an
1515 // arbitrary fixed point in the past.
1516 bool os::getTimesSecs(double* process_real_time,
1517 double* process_user_time,
1518 double* process_system_time) {
1519 struct tms ticks;
1520 clock_t real_ticks = times(&ticks);
1521
1522 if (real_ticks == (clock_t) (-1)) {
1523 return false;
1524 } else {
1525 double ticks_per_second = (double) clock_tics_per_sec;
1526 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1527 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1528 *process_real_time = ((double) real_ticks) / ticks_per_second;
1529
1530 return true;
1531 }
1532 }
1533
1534
1535 char * os::local_time_string(char *buf, size_t buflen) {
1536 struct tm t;
1537 time_t long_time;
1538 time(&long_time);
1539 localtime_r(&long_time, &t);
1540 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1541 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1542 t.tm_hour, t.tm_min, t.tm_sec);
1543 return buf;
1544 }
1545
1546 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1547 return localtime_r(clock, res);
1548 }
1549
1550 ////////////////////////////////////////////////////////////////////////////////
1551 // runtime exit support
1552
1553 // Note: os::shutdown() might be called very early during initialization, or
1554 // called from signal handler. Before adding something to os::shutdown(), make
1555 // sure it is async-safe and can handle partially initialized VM.
1556 void os::shutdown() {
1557
1558 // allow PerfMemory to attempt cleanup of any persistent resources
1559 perfMemory_exit();
1560
1561 // needs to remove object in file system
1562 AttachListener::abort();
1563
1564 // flush buffered output, finish log files
1565 ostream_abort();
1566
1567 // Check for abort hook
1568 abort_hook_t abort_hook = Arguments::abort_hook();
1569 if (abort_hook != NULL) {
1570 abort_hook();
1571 }
1572
1573 }
1574
1575 // Note: os::abort() might be called very early during initialization, or
1576 // called from signal handler. Before adding something to os::abort(), make
1577 // sure it is async-safe and can handle partially initialized VM.
1578 void os::abort(bool dump_core) {
1579 os::shutdown();
1580 if (dump_core) {
1581 #ifndef PRODUCT
1582 fdStream out(defaultStream::output_fd());
1583 out.print_raw("Current thread is ");
1584 char buf[16];
1585 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1586 out.print_raw_cr(buf);
1587 out.print_raw_cr("Dumping core ...");
1588 #endif
1589 ::abort(); // dump core
1590 }
1591
1592 ::exit(1);
1593 }
1594
1595 // Die immediately, no exit hook, no abort hook, no cleanup.
1596 void os::die() {
1597 // _exit() on LinuxThreads only kills current thread
1598 ::abort();
1599 }
1600
1601 // unused on linux for now.
1602 void os::set_error_file(const char *logfile) {}
1603
1604
1605 // This method is a copy of JDK's sysGetLastErrorString
1606 // from src/solaris/hpi/src/system_md.c
1607
1608 size_t os::lasterror(char *buf, size_t len) {
1609
1610 if (errno == 0) return 0;
1611
1612 const char *s = ::strerror(errno);
1613 size_t n = ::strlen(s);
1614 if (n >= len) {
1615 n = len - 1;
1616 }
1617 ::strncpy(buf, s, n);
1618 buf[n] = '\0';
1619 return n;
1620 }
1621
1622 intx os::current_thread_id() { return (intx)pthread_self(); }
1623 int os::current_process_id() {
1624
1625 // Under the old linux thread library, linux gives each thread
1626 // its own process id. Because of this each thread will return
1627 // a different pid if this method were to return the result
1628 // of getpid(2). Linux provides no api that returns the pid
1629 // of the launcher thread for the vm. This implementation
1630 // returns a unique pid, the pid of the launcher thread
1631 // that starts the vm 'process'.
1632
1633 // Under the NPTL, getpid() returns the same pid as the
1634 // launcher thread rather than a unique pid per thread.
1635 // Use gettid() if you want the old pre NPTL behaviour.
1636
1637 // if you are looking for the result of a call to getpid() that
1638 // returns a unique pid for the calling thread, then look at the
1639 // OSThread::thread_id() method in osThread_linux.hpp file
1640
1641 return (int)(_initial_pid ? _initial_pid : getpid());
1642 }
1643
1644 // DLL functions
1645
1646 const char* os::dll_file_extension() { return ".so"; }
1647
1648 // This must be hard coded because it's the system's temporary
1649 // directory not the java application's temp directory, ala java.io.tmpdir.
1650 const char* os::get_temp_directory() { return "/tmp"; }
1651
1652 static bool file_exists(const char* filename) {
1653 struct stat statbuf;
1654 if (filename == NULL || strlen(filename) == 0) {
1655 return false;
1656 }
1657 return os::stat(filename, &statbuf) == 0;
1658 }
1659
1660 void os::dll_build_name(char* buffer, size_t buflen,
1661 const char* pname, const char* fname) {
1662 // Copied from libhpi
1663 const size_t pnamelen = pname ? strlen(pname) : 0;
1664
1665 // Quietly truncate on buffer overflow. Should be an error.
1666 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1667 *buffer = '\0';
1668 return;
1669 }
1670
1671 if (pnamelen == 0) {
1672 snprintf(buffer, buflen, "lib%s.so", fname);
1673 } else if (strchr(pname, *os::path_separator()) != NULL) {
1674 int n;
1675 char** pelements = split_path(pname, &n);
1676 for (int i = 0 ; i < n ; i++) {
1677 // Really shouldn't be NULL, but check can't hurt
1678 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1679 continue; // skip the empty path values
1680 }
1681 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1682 if (file_exists(buffer)) {
1683 break;
1684 }
1685 }
1686 // release the storage
1687 for (int i = 0 ; i < n ; i++) {
1688 if (pelements[i] != NULL) {
1689 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1690 }
1691 }
1692 if (pelements != NULL) {
1693 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1694 }
1695 } else {
1696 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1697 }
1698 }
1699
1700 const char* os::get_current_directory(char *buf, int buflen) {
1701 return getcwd(buf, buflen);
1702 }
1703
1704 // check if addr is inside libjvm[_g].so
1705 bool os::address_is_in_vm(address addr) {
1706 static address libjvm_base_addr;
1707 Dl_info dlinfo;
1708
1709 if (libjvm_base_addr == NULL) {
1710 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1711 libjvm_base_addr = (address)dlinfo.dli_fbase;
1712 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1713 }
1714
1715 if (dladdr((void *)addr, &dlinfo)) {
1716 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1717 }
1718
1719 return false;
1720 }
1721
1722 bool os::dll_address_to_function_name(address addr, char *buf,
1723 int buflen, int *offset) {
1724 Dl_info dlinfo;
1725
1726 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1727 if (buf != NULL) {
1728 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1729 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1730 }
1731 }
1732 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1733 return true;
1734 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1735 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1736 buf, buflen, offset, dlinfo.dli_fname)) {
1737 return true;
1738 }
1739 }
1740
1741 if (buf != NULL) buf[0] = '\0';
1742 if (offset != NULL) *offset = -1;
1743 return false;
1744 }
1745
1746 struct _address_to_library_name {
1747 address addr; // input : memory address
1748 size_t buflen; // size of fname
1749 char* fname; // output: library name
1750 address base; // library base addr
1751 };
1752
1753 static int address_to_library_name_callback(struct dl_phdr_info *info,
1754 size_t size, void *data) {
1755 int i;
1756 bool found = false;
1757 address libbase = NULL;
1758 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1759
1760 // iterate through all loadable segments
1761 for (i = 0; i < info->dlpi_phnum; i++) {
1762 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1763 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1764 // base address of a library is the lowest address of its loaded
1765 // segments.
1766 if (libbase == NULL || libbase > segbase) {
1767 libbase = segbase;
1768 }
1769 // see if 'addr' is within current segment
1770 if (segbase <= d->addr &&
1771 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1772 found = true;
1773 }
1774 }
1775 }
1776
1777 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1778 // so dll_address_to_library_name() can fall through to use dladdr() which
1779 // can figure out executable name from argv[0].
1780 if (found && info->dlpi_name && info->dlpi_name[0]) {
1781 d->base = libbase;
1782 if (d->fname) {
1783 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1784 }
1785 return 1;
1786 }
1787 return 0;
1788 }
1789
1790 bool os::dll_address_to_library_name(address addr, char* buf,
1791 int buflen, int* offset) {
1792 Dl_info dlinfo;
1793 struct _address_to_library_name data;
1794
1795 // There is a bug in old glibc dladdr() implementation that it could resolve
1796 // to wrong library name if the .so file has a base address != NULL. Here
1797 // we iterate through the program headers of all loaded libraries to find
1798 // out which library 'addr' really belongs to. This workaround can be
1799 // removed once the minimum requirement for glibc is moved to 2.3.x.
1800 data.addr = addr;
1801 data.fname = buf;
1802 data.buflen = buflen;
1803 data.base = NULL;
1804 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1805
1806 if (rslt) {
1807 // buf already contains library name
1808 if (offset) *offset = addr - data.base;
1809 return true;
1810 } else if (dladdr((void*)addr, &dlinfo)){
1811 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1812 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1813 return true;
1814 } else {
1815 if (buf) buf[0] = '\0';
1816 if (offset) *offset = -1;
1817 return false;
1818 }
1819 }
1820
1821 // Loads .dll/.so and
1822 // in case of error it checks if .dll/.so was built for the
1823 // same architecture as Hotspot is running on
1824
1825
1826 // Remember the stack's state. The Linux dynamic linker will change
1827 // the stack to 'executable' at most once, so we must safepoint only once.
1828 bool os::Linux::_stack_is_executable = false;
1829
1830 // VM operation that loads a library. This is necessary if stack protection
1831 // of the Java stacks can be lost during loading the library. If we
1832 // do not stop the Java threads, they can stack overflow before the stacks
1833 // are protected again.
1834 class VM_LinuxDllLoad: public VM_Operation {
1835 private:
1836 const char *_filename;
1837 char *_ebuf;
1838 int _ebuflen;
1839 void *_lib;
1840 public:
1841 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1842 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1843 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1844 void doit() {
1845 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1846 os::Linux::_stack_is_executable = true;
1847 }
1848 void* loaded_library() { return _lib; }
1849 };
1850
1851 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1852 {
1853 void * result = NULL;
1854 bool load_attempted = false;
1855
1856 // Check whether the library to load might change execution rights
1857 // of the stack. If they are changed, the protection of the stack
1858 // guard pages will be lost. We need a safepoint to fix this.
1859 //
1860 // See Linux man page execstack(8) for more info.
1861 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1862 ElfFile ef(filename);
1863 if (!ef.specifies_noexecstack()) {
1864 if (!is_init_completed()) {
1865 os::Linux::_stack_is_executable = true;
1866 // This is OK - No Java threads have been created yet, and hence no
1867 // stack guard pages to fix.
1868 //
1869 // This should happen only when you are building JDK7 using a very
1870 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1871 //
1872 // Dynamic loader will make all stacks executable after
1873 // this function returns, and will not do that again.
1874 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1875 } else {
1876 warning("You have loaded library %s which might have disabled stack guard. "
1877 "The VM will try to fix the stack guard now.\n"
1878 "It's highly recommended that you fix the library with "
1879 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1880 filename);
1881
1882 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1883 JavaThread *jt = JavaThread::current();
1884 if (jt->thread_state() != _thread_in_native) {
1885 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1886 // that requires ExecStack. Cannot enter safe point. Let's give up.
1887 warning("Unable to fix stack guard. Giving up.");
1888 } else {
1889 if (!LoadExecStackDllInVMThread) {
1890 // This is for the case where the DLL has an static
1891 // constructor function that executes JNI code. We cannot
1892 // load such DLLs in the VMThread.
1893 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1894 }
1895
1896 ThreadInVMfromNative tiv(jt);
1897 debug_only(VMNativeEntryWrapper vew;)
1898
1899 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1900 VMThread::execute(&op);
1901 if (LoadExecStackDllInVMThread) {
1902 result = op.loaded_library();
1903 }
1904 load_attempted = true;
1905 }
1906 }
1907 }
1908 }
1909
1910 if (!load_attempted) {
1911 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1912 }
1913
1914 if (result != NULL) {
1915 // Successful loading
1916 return result;
1917 }
1918
1919 Elf32_Ehdr elf_head;
1920 int diag_msg_max_length=ebuflen-strlen(ebuf);
1921 char* diag_msg_buf=ebuf+strlen(ebuf);
1922
1923 if (diag_msg_max_length==0) {
1924 // No more space in ebuf for additional diagnostics message
1925 return NULL;
1926 }
1927
1928
1929 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1930
1931 if (file_descriptor < 0) {
1932 // Can't open library, report dlerror() message
1933 return NULL;
1934 }
1935
1936 bool failed_to_read_elf_head=
1937 (sizeof(elf_head)!=
1938 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1939
1940 ::close(file_descriptor);
1941 if (failed_to_read_elf_head) {
1942 // file i/o error - report dlerror() msg
1943 return NULL;
1944 }
1945
1946 typedef struct {
1947 Elf32_Half code; // Actual value as defined in elf.h
1948 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1949 char elf_class; // 32 or 64 bit
1950 char endianess; // MSB or LSB
1951 char* name; // String representation
1952 } arch_t;
1953
1954 #ifndef EM_486
1955 #define EM_486 6 /* Intel 80486 */
1956 #endif
1957
1958 static const arch_t arch_array[]={
1959 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1960 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1961 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1962 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1963 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1964 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1965 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1966 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1967 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1968 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1969 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1970 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1971 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1972 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1973 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1974 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1975 };
1976
1977 #if (defined IA32)
1978 static Elf32_Half running_arch_code=EM_386;
1979 #elif (defined AMD64)
1980 static Elf32_Half running_arch_code=EM_X86_64;
1981 #elif (defined IA64)
1982 static Elf32_Half running_arch_code=EM_IA_64;
1983 #elif (defined __sparc) && (defined _LP64)
1984 static Elf32_Half running_arch_code=EM_SPARCV9;
1985 #elif (defined __sparc) && (!defined _LP64)
1986 static Elf32_Half running_arch_code=EM_SPARC;
1987 #elif (defined __powerpc64__)
1988 static Elf32_Half running_arch_code=EM_PPC64;
1989 #elif (defined __powerpc__)
1990 static Elf32_Half running_arch_code=EM_PPC;
1991 #elif (defined ARM)
1992 static Elf32_Half running_arch_code=EM_ARM;
1993 #elif (defined S390)
1994 static Elf32_Half running_arch_code=EM_S390;
1995 #elif (defined ALPHA)
1996 static Elf32_Half running_arch_code=EM_ALPHA;
1997 #elif (defined MIPSEL)
1998 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1999 #elif (defined PARISC)
2000 static Elf32_Half running_arch_code=EM_PARISC;
2001 #elif (defined MIPS)
2002 static Elf32_Half running_arch_code=EM_MIPS;
2003 #elif (defined M68K)
2004 static Elf32_Half running_arch_code=EM_68K;
2005 #else
2006 #error Method os::dll_load requires that one of following is defined:\
2007 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2008 #endif
2009
2010 // Identify compatability class for VM's architecture and library's architecture
2011 // Obtain string descriptions for architectures
2012
2013 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2014 int running_arch_index=-1;
2015
2016 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2017 if (running_arch_code == arch_array[i].code) {
2018 running_arch_index = i;
2019 }
2020 if (lib_arch.code == arch_array[i].code) {
2021 lib_arch.compat_class = arch_array[i].compat_class;
2022 lib_arch.name = arch_array[i].name;
2023 }
2024 }
2025
2026 assert(running_arch_index != -1,
2027 "Didn't find running architecture code (running_arch_code) in arch_array");
2028 if (running_arch_index == -1) {
2029 // Even though running architecture detection failed
2030 // we may still continue with reporting dlerror() message
2031 return NULL;
2032 }
2033
2034 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2035 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2036 return NULL;
2037 }
2038
2039 #ifndef S390
2040 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2041 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2042 return NULL;
2043 }
2044 #endif // !S390
2045
2046 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2047 if ( lib_arch.name!=NULL ) {
2048 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2049 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2050 lib_arch.name, arch_array[running_arch_index].name);
2051 } else {
2052 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2053 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2054 lib_arch.code,
2055 arch_array[running_arch_index].name);
2056 }
2057 }
2058
2059 return NULL;
2060 }
2061
2062 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2063 void * result = ::dlopen(filename, RTLD_LAZY);
2064 if (result == NULL) {
2065 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2066 ebuf[ebuflen-1] = '\0';
2067 }
2068 return result;
2069 }
2070
2071 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2072 void * result = NULL;
2073 if (LoadExecStackDllInVMThread) {
2074 result = dlopen_helper(filename, ebuf, ebuflen);
2075 }
2076
2077 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2078 // library that requires an executable stack, or which does not have this
2079 // stack attribute set, dlopen changes the stack attribute to executable. The
2080 // read protection of the guard pages gets lost.
2081 //
2082 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2083 // may have been queued at the same time.
2084
2085 if (!_stack_is_executable) {
2086 JavaThread *jt = Threads::first();
2087
2088 while (jt) {
2089 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2090 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2091 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2092 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2093 warning("Attempt to reguard stack yellow zone failed.");
2094 }
2095 }
2096 jt = jt->next();
2097 }
2098 }
2099
2100 return result;
2101 }
2102
2103 /*
2104 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2105 * chances are you might want to run the generated bits against glibc-2.0
2106 * libdl.so, so always use locking for any version of glibc.
2107 */
2108 void* os::dll_lookup(void* handle, const char* name) {
2109 pthread_mutex_lock(&dl_mutex);
2110 void* res = dlsym(handle, name);
2111 pthread_mutex_unlock(&dl_mutex);
2112 return res;
2113 }
2114
2115
2116 static bool _print_ascii_file(const char* filename, outputStream* st) {
2117 int fd = ::open(filename, O_RDONLY);
2118 if (fd == -1) {
2119 return false;
2120 }
2121
2122 char buf[32];
2123 int bytes;
2124 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2125 st->print_raw(buf, bytes);
2126 }
2127
2128 ::close(fd);
2129
2130 return true;
2131 }
2132
2133 void os::print_dll_info(outputStream *st) {
2134 st->print_cr("Dynamic libraries:");
2135
2136 char fname[32];
2137 pid_t pid = os::Linux::gettid();
2138
2139 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2140
2141 if (!_print_ascii_file(fname, st)) {
2142 st->print("Can not get library information for pid = %d\n", pid);
2143 }
2144 }
2145
2146 void os::print_os_info_brief(outputStream* st) {
2147 os::Linux::print_distro_info(st);
2148
2149 os::Posix::print_uname_info(st);
2150
2151 os::Linux::print_libversion_info(st);
2152
2153 }
2154
2155 void os::print_os_info(outputStream* st) {
2156 st->print("OS:");
2157
2158 os::Linux::print_distro_info(st);
2159
2160 os::Posix::print_uname_info(st);
2161
2162 // Print warning if unsafe chroot environment detected
2163 if (unsafe_chroot_detected) {
2164 st->print("WARNING!! ");
2165 st->print_cr(unstable_chroot_error);
2166 }
2167
2168 os::Linux::print_libversion_info(st);
2169
2170 os::Posix::print_rlimit_info(st);
2171
2172 os::Posix::print_load_average(st);
2173
2174 os::Linux::print_full_memory_info(st);
2175 }
2176
2177 // Try to identify popular distros.
2178 // Most Linux distributions have /etc/XXX-release file, which contains
2179 // the OS version string. Some have more than one /etc/XXX-release file
2180 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2181 // so the order is important.
2182 void os::Linux::print_distro_info(outputStream* st) {
2183 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2184 !_print_ascii_file("/etc/sun-release", st) &&
2185 !_print_ascii_file("/etc/redhat-release", st) &&
2186 !_print_ascii_file("/etc/SuSE-release", st) &&
2187 !_print_ascii_file("/etc/turbolinux-release", st) &&
2188 !_print_ascii_file("/etc/gentoo-release", st) &&
2189 !_print_ascii_file("/etc/debian_version", st) &&
2190 !_print_ascii_file("/etc/ltib-release", st) &&
2191 !_print_ascii_file("/etc/angstrom-version", st)) {
2192 st->print("Linux");
2193 }
2194 st->cr();
2195 }
2196
2197 void os::Linux::print_libversion_info(outputStream* st) {
2198 // libc, pthread
2199 st->print("libc:");
2200 st->print(os::Linux::glibc_version()); st->print(" ");
2201 st->print(os::Linux::libpthread_version()); st->print(" ");
2202 if (os::Linux::is_LinuxThreads()) {
2203 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2204 }
2205 st->cr();
2206 }
2207
2208 void os::Linux::print_full_memory_info(outputStream* st) {
2209 st->print("\n/proc/meminfo:\n");
2210 _print_ascii_file("/proc/meminfo", st);
2211 st->cr();
2212 }
2213
2214 void os::print_memory_info(outputStream* st) {
2215
2216 st->print("Memory:");
2217 st->print(" %dk page", os::vm_page_size()>>10);
2218
2219 // values in struct sysinfo are "unsigned long"
2220 struct sysinfo si;
2221 sysinfo(&si);
2222
2223 st->print(", physical " UINT64_FORMAT "k",
2224 os::physical_memory() >> 10);
2225 st->print("(" UINT64_FORMAT "k free)",
2226 os::available_memory() >> 10);
2227 st->print(", swap " UINT64_FORMAT "k",
2228 ((jlong)si.totalswap * si.mem_unit) >> 10);
2229 st->print("(" UINT64_FORMAT "k free)",
2230 ((jlong)si.freeswap * si.mem_unit) >> 10);
2231 st->cr();
2232 }
2233
2234 void os::pd_print_cpu_info(outputStream* st) {
2235 st->print("\n/proc/cpuinfo:\n");
2236 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2237 st->print(" <Not Available>");
2238 }
2239 st->cr();
2240 }
2241
2242 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2243 // but they're the same for all the linux arch that we support
2244 // and they're the same for solaris but there's no common place to put this.
2245 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2246 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2247 "ILL_COPROC", "ILL_BADSTK" };
2248
2249 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2250 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2251 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2252
2253 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2254
2255 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2256
2257 void os::print_siginfo(outputStream* st, void* siginfo) {
2258 st->print("siginfo:");
2259
2260 const int buflen = 100;
2261 char buf[buflen];
2262 siginfo_t *si = (siginfo_t*)siginfo;
2263 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2264 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2265 st->print("si_errno=%s", buf);
2266 } else {
2267 st->print("si_errno=%d", si->si_errno);
2268 }
2269 const int c = si->si_code;
2270 assert(c > 0, "unexpected si_code");
2271 switch (si->si_signo) {
2272 case SIGILL:
2273 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2274 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2275 break;
2276 case SIGFPE:
2277 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2278 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2279 break;
2280 case SIGSEGV:
2281 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2282 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2283 break;
2284 case SIGBUS:
2285 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2286 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2287 break;
2288 default:
2289 st->print(", si_code=%d", si->si_code);
2290 // no si_addr
2291 }
2292
2293 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2294 UseSharedSpaces) {
2295 FileMapInfo* mapinfo = FileMapInfo::current_info();
2296 if (mapinfo->is_in_shared_space(si->si_addr)) {
2297 st->print("\n\nError accessing class data sharing archive." \
2298 " Mapped file inaccessible during execution, " \
2299 " possible disk/network problem.");
2300 }
2301 }
2302 st->cr();
2303 }
2304
2305
2306 static void print_signal_handler(outputStream* st, int sig,
2307 char* buf, size_t buflen);
2308
2309 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2310 st->print_cr("Signal Handlers:");
2311 print_signal_handler(st, SIGSEGV, buf, buflen);
2312 print_signal_handler(st, SIGBUS , buf, buflen);
2313 print_signal_handler(st, SIGFPE , buf, buflen);
2314 print_signal_handler(st, SIGPIPE, buf, buflen);
2315 print_signal_handler(st, SIGXFSZ, buf, buflen);
2316 print_signal_handler(st, SIGILL , buf, buflen);
2317 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2318 print_signal_handler(st, SR_signum, buf, buflen);
2319 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2320 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2321 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2322 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2323 }
2324
2325 static char saved_jvm_path[MAXPATHLEN] = {0};
2326
2327 // Find the full path to the current module, libjvm.so or libjvm_g.so
2328 void os::jvm_path(char *buf, jint buflen) {
2329 // Error checking.
2330 if (buflen < MAXPATHLEN) {
2331 assert(false, "must use a large-enough buffer");
2332 buf[0] = '\0';
2333 return;
2334 }
2335 // Lazy resolve the path to current module.
2336 if (saved_jvm_path[0] != 0) {
2337 strcpy(buf, saved_jvm_path);
2338 return;
2339 }
2340
2341 char dli_fname[MAXPATHLEN];
2342 bool ret = dll_address_to_library_name(
2343 CAST_FROM_FN_PTR(address, os::jvm_path),
2344 dli_fname, sizeof(dli_fname), NULL);
2345 assert(ret != 0, "cannot locate libjvm");
2346 char *rp = realpath(dli_fname, buf);
2347 if (rp == NULL)
2348 return;
2349
2350 if (Arguments::created_by_gamma_launcher()) {
2351 // Support for the gamma launcher. Typical value for buf is
2352 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2353 // the right place in the string, then assume we are installed in a JDK and
2354 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2355 // up the path so it looks like libjvm.so is installed there (append a
2356 // fake suffix hotspot/libjvm.so).
2357 const char *p = buf + strlen(buf) - 1;
2358 for (int count = 0; p > buf && count < 5; ++count) {
2359 for (--p; p > buf && *p != '/'; --p)
2360 /* empty */ ;
2361 }
2362
2363 if (strncmp(p, "/jre/lib/", 9) != 0) {
2364 // Look for JAVA_HOME in the environment.
2365 char* java_home_var = ::getenv("JAVA_HOME");
2366 if (java_home_var != NULL && java_home_var[0] != 0) {
2367 char* jrelib_p;
2368 int len;
2369
2370 // Check the current module name "libjvm.so" or "libjvm_g.so".
2371 p = strrchr(buf, '/');
2372 assert(strstr(p, "/libjvm") == p, "invalid library name");
2373 p = strstr(p, "_g") ? "_g" : "";
2374
2375 rp = realpath(java_home_var, buf);
2376 if (rp == NULL)
2377 return;
2378
2379 // determine if this is a legacy image or modules image
2380 // modules image doesn't have "jre" subdirectory
2381 len = strlen(buf);
2382 jrelib_p = buf + len;
2383 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2384 if (0 != access(buf, F_OK)) {
2385 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2386 }
2387
2388 if (0 == access(buf, F_OK)) {
2389 // Use current module name "libjvm[_g].so" instead of
2390 // "libjvm"debug_only("_g")".so" since for fastdebug version
2391 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2392 len = strlen(buf);
2393 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2394 } else {
2395 // Go back to path of .so
2396 rp = realpath(dli_fname, buf);
2397 if (rp == NULL)
2398 return;
2399 }
2400 }
2401 }
2402 }
2403
2404 strcpy(saved_jvm_path, buf);
2405 }
2406
2407 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2408 // no prefix required, not even "_"
2409 }
2410
2411 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2412 // no suffix required
2413 }
2414
2415 ////////////////////////////////////////////////////////////////////////////////
2416 // sun.misc.Signal support
2417
2418 static volatile jint sigint_count = 0;
2419
2420 static void
2421 UserHandler(int sig, void *siginfo, void *context) {
2422 // 4511530 - sem_post is serialized and handled by the manager thread. When
2423 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2424 // don't want to flood the manager thread with sem_post requests.
2425 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2426 return;
2427
2428 // Ctrl-C is pressed during error reporting, likely because the error
2429 // handler fails to abort. Let VM die immediately.
2430 if (sig == SIGINT && is_error_reported()) {
2431 os::die();
2432 }
2433
2434 os::signal_notify(sig);
2435 }
2436
2437 void* os::user_handler() {
2438 return CAST_FROM_FN_PTR(void*, UserHandler);
2439 }
2440
2441 class Semaphore : public StackObj {
2442 public:
2443 Semaphore();
2444 ~Semaphore();
2445 void signal();
2446 void wait();
2447 bool trywait();
2448 bool timedwait(unsigned int sec, int nsec);
2449 private:
2450 sem_t _semaphore;
2451 };
2452
2453
2454 Semaphore::Semaphore() {
2455 sem_init(&_semaphore, 0, 0);
2456 }
2457
2458 Semaphore::~Semaphore() {
2459 sem_destroy(&_semaphore);
2460 }
2461
2462 void Semaphore::signal() {
2463 sem_post(&_semaphore);
2464 }
2465
2466 void Semaphore::wait() {
2467 sem_wait(&_semaphore);
2468 }
2469
2470 bool Semaphore::trywait() {
2471 return sem_trywait(&_semaphore) == 0;
2472 }
2473
2474 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2475 struct timespec ts;
2476 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2477
2478 while (1) {
2479 int result = sem_timedwait(&_semaphore, &ts);
2480 if (result == 0) {
2481 return true;
2482 } else if (errno == EINTR) {
2483 continue;
2484 } else if (errno == ETIMEDOUT) {
2485 return false;
2486 } else {
2487 return false;
2488 }
2489 }
2490 }
2491
2492 extern "C" {
2493 typedef void (*sa_handler_t)(int);
2494 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2495 }
2496
2497 void* os::signal(int signal_number, void* handler) {
2498 struct sigaction sigAct, oldSigAct;
2499
2500 sigfillset(&(sigAct.sa_mask));
2501 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2502 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2503
2504 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2505 // -1 means registration failed
2506 return (void *)-1;
2507 }
2508
2509 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2510 }
2511
2512 void os::signal_raise(int signal_number) {
2513 ::raise(signal_number);
2514 }
2515
2516 /*
2517 * The following code is moved from os.cpp for making this
2518 * code platform specific, which it is by its very nature.
2519 */
2520
2521 // Will be modified when max signal is changed to be dynamic
2522 int os::sigexitnum_pd() {
2523 return NSIG;
2524 }
2525
2526 // a counter for each possible signal value
2527 static volatile jint pending_signals[NSIG+1] = { 0 };
2528
2529 // Linux(POSIX) specific hand shaking semaphore.
2530 static sem_t sig_sem;
2531 static Semaphore sr_semaphore;
2532
2533 void os::signal_init_pd() {
2534 // Initialize signal structures
2535 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2536
2537 // Initialize signal semaphore
2538 ::sem_init(&sig_sem, 0, 0);
2539 }
2540
2541 void os::signal_notify(int sig) {
2542 Atomic::inc(&pending_signals[sig]);
2543 ::sem_post(&sig_sem);
2544 }
2545
2546 static int check_pending_signals(bool wait) {
2547 Atomic::store(0, &sigint_count);
2548 for (;;) {
2549 for (int i = 0; i < NSIG + 1; i++) {
2550 jint n = pending_signals[i];
2551 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2552 return i;
2553 }
2554 }
2555 if (!wait) {
2556 return -1;
2557 }
2558 JavaThread *thread = JavaThread::current();
2559 ThreadBlockInVM tbivm(thread);
2560
2561 bool threadIsSuspended;
2562 do {
2563 thread->set_suspend_equivalent();
2564 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2565 ::sem_wait(&sig_sem);
2566
2567 // were we externally suspended while we were waiting?
2568 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2569 if (threadIsSuspended) {
2570 //
2571 // The semaphore has been incremented, but while we were waiting
2572 // another thread suspended us. We don't want to continue running
2573 // while suspended because that would surprise the thread that
2574 // suspended us.
2575 //
2576 ::sem_post(&sig_sem);
2577
2578 thread->java_suspend_self();
2579 }
2580 } while (threadIsSuspended);
2581 }
2582 }
2583
2584 int os::signal_lookup() {
2585 return check_pending_signals(false);
2586 }
2587
2588 int os::signal_wait() {
2589 return check_pending_signals(true);
2590 }
2591
2592 ////////////////////////////////////////////////////////////////////////////////
2593 // Virtual Memory
2594
2595 int os::vm_page_size() {
2596 // Seems redundant as all get out
2597 assert(os::Linux::page_size() != -1, "must call os::init");
2598 return os::Linux::page_size();
2599 }
2600
2601 // Solaris allocates memory by pages.
2602 int os::vm_allocation_granularity() {
2603 assert(os::Linux::page_size() != -1, "must call os::init");
2604 return os::Linux::page_size();
2605 }
2606
2607 // Rationale behind this function:
2608 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2609 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2610 // samples for JITted code. Here we create private executable mapping over the code cache
2611 // and then we can use standard (well, almost, as mapping can change) way to provide
2612 // info for the reporting script by storing timestamp and location of symbol
2613 void linux_wrap_code(char* base, size_t size) {
2614 static volatile jint cnt = 0;
2615
2616 if (!UseOprofile) {
2617 return;
2618 }
2619
2620 char buf[PATH_MAX+1];
2621 int num = Atomic::add(1, &cnt);
2622
2623 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2624 os::get_temp_directory(), os::current_process_id(), num);
2625 unlink(buf);
2626
2627 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2628
2629 if (fd != -1) {
2630 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2631 if (rv != (off_t)-1) {
2632 if (::write(fd, "", 1) == 1) {
2633 mmap(base, size,
2634 PROT_READ|PROT_WRITE|PROT_EXEC,
2635 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2636 }
2637 }
2638 ::close(fd);
2639 unlink(buf);
2640 }
2641 }
2642
2643 static bool recoverable_mmap_error(int err) {
2644 // See if the error is one we can let the caller handle. This
2645 // list of errno values comes from JBS-6843484. I can't find a
2646 // Linux man page that documents this specific set of errno
2647 // values so while this list currently matches Solaris, it may
2648 // change as we gain experience with this failure mode.
2649 switch (err) {
2650 case EBADF:
2651 case EINVAL:
2652 case ENOTSUP:
2653 // let the caller deal with these errors
2654 return true;
2655
2656 default:
2657 // Any remaining errors on this OS can cause our reserved mapping
2658 // to be lost. That can cause confusion where different data
2659 // structures think they have the same memory mapped. The worst
2660 // scenario is if both the VM and a library think they have the
2661 // same memory mapped.
2662 return false;
2663 }
2664 }
2665
2666 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2667 int err) {
2668 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2669 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2670 strerror(err), err);
2671 }
2672
2673 static void warn_fail_commit_memory(char* addr, size_t size,
2674 size_t alignment_hint, bool exec,
2675 int err) {
2676 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2677 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2678 alignment_hint, exec, strerror(err), err);
2679 }
2680
2681 // NOTE: Linux kernel does not really reserve the pages for us.
2682 // All it does is to check if there are enough free pages
2683 // left at the time of mmap(). This could be a potential
2684 // problem.
2685 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2686 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2687 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2688 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2689 if (res != (uintptr_t) MAP_FAILED) {
2690 if (UseNUMAInterleaving) {
2691 numa_make_global(addr, size);
2692 }
2693 return 0;
2694 }
2695
2696 int err = errno; // save errno from mmap() call above
2697
2698 if (!recoverable_mmap_error(err)) {
2699 warn_fail_commit_memory(addr, size, exec, err);
2700 vm_exit_out_of_memory(size, "committing reserved memory.");
2701 }
2702
2703 return err;
2704 }
2705
2706 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2707 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2708 }
2709
2710 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2711 const char* mesg) {
2712 assert(mesg != NULL, "mesg must be specified");
2713 int err = os::Linux::commit_memory_impl(addr, size, exec);
2714 if (err != 0) {
2715 // the caller wants all commit errors to exit with the specified mesg:
2716 warn_fail_commit_memory(addr, size, exec, err);
2717 vm_exit_out_of_memory(size, mesg);
2718 }
2719 }
2720
2721 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2722 #ifndef MAP_HUGETLB
2723 #define MAP_HUGETLB 0x40000
2724 #endif
2725
2726 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2727 #ifndef MADV_HUGEPAGE
2728 #define MADV_HUGEPAGE 14
2729 #endif
2730
2731 int os::Linux::commit_memory_impl(char* addr, size_t size,
2732 size_t alignment_hint, bool exec) {
2733 int err = os::Linux::commit_memory_impl(addr, size, exec);
2734 if (err == 0) {
2735 realign_memory(addr, size, alignment_hint);
2736 }
2737 return err;
2738 }
2739
2740 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2741 bool exec) {
2742 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2743 }
2744
2745 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2746 size_t alignment_hint, bool exec,
2747 const char* mesg) {
2748 assert(mesg != NULL, "mesg must be specified");
2749 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2750 if (err != 0) {
2751 // the caller wants all commit errors to exit with the specified mesg:
2752 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2753 vm_exit_out_of_memory(size, mesg);
2754 }
2755 }
2756
2757 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2758 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2759 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2760 // be supported or the memory may already be backed by huge pages.
2761 ::madvise(addr, bytes, MADV_HUGEPAGE);
2762 }
2763 }
2764
2765 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2766 // This method works by doing an mmap over an existing mmaping and effectively discarding
2767 // the existing pages. However it won't work for SHM-based large pages that cannot be
2768 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2769 // small pages on top of the SHM segment. This method always works for small pages, so we
2770 // allow that in any case.
2771 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2772 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2773 }
2774 }
2775
2776 void os::numa_make_global(char *addr, size_t bytes) {
2777 Linux::numa_interleave_memory(addr, bytes);
2778 }
2779
2780 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2781 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2782 }
2783
2784 bool os::numa_topology_changed() { return false; }
2785
2786 size_t os::numa_get_groups_num() {
2787 int max_node = Linux::numa_max_node();
2788 return max_node > 0 ? max_node + 1 : 1;
2789 }
2790
2791 int os::numa_get_group_id() {
2792 int cpu_id = Linux::sched_getcpu();
2793 if (cpu_id != -1) {
2794 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2795 if (lgrp_id != -1) {
2796 return lgrp_id;
2797 }
2798 }
2799 return 0;
2800 }
2801
2802 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2803 for (size_t i = 0; i < size; i++) {
2804 ids[i] = i;
2805 }
2806 return size;
2807 }
2808
2809 bool os::get_page_info(char *start, page_info* info) {
2810 return false;
2811 }
2812
2813 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2814 return end;
2815 }
2816
2817
2818 int os::Linux::sched_getcpu_syscall(void) {
2819 unsigned int cpu;
2820 int retval = -1;
2821
2822 #if defined(IA32)
2823 # ifndef SYS_getcpu
2824 # define SYS_getcpu 318
2825 # endif
2826 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2827 #elif defined(AMD64)
2828 // Unfortunately we have to bring all these macros here from vsyscall.h
2829 // to be able to compile on old linuxes.
2830 # define __NR_vgetcpu 2
2831 # define VSYSCALL_START (-10UL << 20)
2832 # define VSYSCALL_SIZE 1024
2833 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2834 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2835 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2836 retval = vgetcpu(&cpu, NULL, NULL);
2837 #endif
2838
2839 return (retval == -1) ? retval : cpu;
2840 }
2841
2842 // Something to do with the numa-aware allocator needs these symbols
2843 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2844 extern "C" JNIEXPORT void numa_error(char *where) { }
2845 extern "C" JNIEXPORT int fork1() { return fork(); }
2846
2847
2848 // If we are running with libnuma version > 2, then we should
2849 // be trying to use symbols with versions 1.1
2850 // If we are running with earlier version, which did not have symbol versions,
2851 // we should use the base version.
2852 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2853 void *f = dlvsym(handle, name, "libnuma_1.1");
2854 if (f == NULL) {
2855 f = dlsym(handle, name);
2856 }
2857 return f;
2858 }
2859
2860 bool os::Linux::libnuma_init() {
2861 // sched_getcpu() should be in libc.
2862 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2863 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2864
2865 // If it's not, try a direct syscall.
2866 if (sched_getcpu() == -1)
2867 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2868
2869 if (sched_getcpu() != -1) { // Does it work?
2870 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2871 if (handle != NULL) {
2872 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2873 libnuma_dlsym(handle, "numa_node_to_cpus")));
2874 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2875 libnuma_dlsym(handle, "numa_max_node")));
2876 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2877 libnuma_dlsym(handle, "numa_available")));
2878 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2879 libnuma_dlsym(handle, "numa_tonode_memory")));
2880 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2881 libnuma_dlsym(handle, "numa_interleave_memory")));
2882
2883
2884 if (numa_available() != -1) {
2885 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2886 // Create a cpu -> node mapping
2887 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2888 rebuild_cpu_to_node_map();
2889 return true;
2890 }
2891 }
2892 }
2893 return false;
2894 }
2895
2896 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2897 // The table is later used in get_node_by_cpu().
2898 void os::Linux::rebuild_cpu_to_node_map() {
2899 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2900 // in libnuma (possible values are starting from 16,
2901 // and continuing up with every other power of 2, but less
2902 // than the maximum number of CPUs supported by kernel), and
2903 // is a subject to change (in libnuma version 2 the requirements
2904 // are more reasonable) we'll just hardcode the number they use
2905 // in the library.
2906 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2907
2908 size_t cpu_num = os::active_processor_count();
2909 size_t cpu_map_size = NCPUS / BitsPerCLong;
2910 size_t cpu_map_valid_size =
2911 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2912
2913 cpu_to_node()->clear();
2914 cpu_to_node()->at_grow(cpu_num - 1);
2915 size_t node_num = numa_get_groups_num();
2916
2917 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2918 for (size_t i = 0; i < node_num; i++) {
2919 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2920 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2921 if (cpu_map[j] != 0) {
2922 for (size_t k = 0; k < BitsPerCLong; k++) {
2923 if (cpu_map[j] & (1UL << k)) {
2924 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2925 }
2926 }
2927 }
2928 }
2929 }
2930 }
2931 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2932 }
2933
2934 int os::Linux::get_node_by_cpu(int cpu_id) {
2935 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2936 return cpu_to_node()->at(cpu_id);
2937 }
2938 return -1;
2939 }
2940
2941 GrowableArray<int>* os::Linux::_cpu_to_node;
2942 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2943 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2944 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2945 os::Linux::numa_available_func_t os::Linux::_numa_available;
2946 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2947 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2948 unsigned long* os::Linux::_numa_all_nodes;
2949
2950 bool os::pd_uncommit_memory(char* addr, size_t size) {
2951 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2952 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2953 return res != (uintptr_t) MAP_FAILED;
2954 }
2955
2956 // Linux uses a growable mapping for the stack, and if the mapping for
2957 // the stack guard pages is not removed when we detach a thread the
2958 // stack cannot grow beyond the pages where the stack guard was
2959 // mapped. If at some point later in the process the stack expands to
2960 // that point, the Linux kernel cannot expand the stack any further
2961 // because the guard pages are in the way, and a segfault occurs.
2962 //
2963 // However, it's essential not to split the stack region by unmapping
2964 // a region (leaving a hole) that's already part of the stack mapping,
2965 // so if the stack mapping has already grown beyond the guard pages at
2966 // the time we create them, we have to truncate the stack mapping.
2967 // So, we need to know the extent of the stack mapping when
2968 // create_stack_guard_pages() is called.
2969
2970 // Find the bounds of the stack mapping. Return true for success.
2971 //
2972 // We only need this for stacks that are growable: at the time of
2973 // writing thread stacks don't use growable mappings (i.e. those
2974 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2975 // only applies to the main thread.
2976
2977 static
2978 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2979
2980 char buf[128];
2981 int fd, sz;
2982
2983 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2984 return false;
2985 }
2986
2987 const char kw[] = "[stack]";
2988 const int kwlen = sizeof(kw)-1;
2989
2990 // Address part of /proc/self/maps couldn't be more than 128 bytes
2991 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2992 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2993 // Extract addresses
2994 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2995 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2996 if (sp >= *bottom && sp <= *top) {
2997 ::close(fd);
2998 return true;
2999 }
3000 }
3001 }
3002 }
3003
3004 ::close(fd);
3005 return false;
3006 }
3007
3008
3009 // If the (growable) stack mapping already extends beyond the point
3010 // where we're going to put our guard pages, truncate the mapping at
3011 // that point by munmap()ping it. This ensures that when we later
3012 // munmap() the guard pages we don't leave a hole in the stack
3013 // mapping. This only affects the main/initial thread, but guard
3014 // against future OS changes
3015 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3016 uintptr_t stack_extent, stack_base;
3017 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
3018 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
3019 assert(os::Linux::is_initial_thread(),
3020 "growable stack in non-initial thread");
3021 if (stack_extent < (uintptr_t)addr)
3022 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
3023 }
3024
3025 return os::commit_memory(addr, size, !ExecMem);
3026 }
3027
3028 // If this is a growable mapping, remove the guard pages entirely by
3029 // munmap()ping them. If not, just call uncommit_memory(). This only
3030 // affects the main/initial thread, but guard against future OS changes
3031 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3032 uintptr_t stack_extent, stack_base;
3033 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
3034 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
3035 assert(os::Linux::is_initial_thread(),
3036 "growable stack in non-initial thread");
3037
3038 return ::munmap(addr, size) == 0;
3039 }
3040
3041 return os::uncommit_memory(addr, size);
3042 }
3043
3044 static address _highest_vm_reserved_address = NULL;
3045
3046 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3047 // at 'requested_addr'. If there are existing memory mappings at the same
3048 // location, however, they will be overwritten. If 'fixed' is false,
3049 // 'requested_addr' is only treated as a hint, the return value may or
3050 // may not start from the requested address. Unlike Linux mmap(), this
3051 // function returns NULL to indicate failure.
3052 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3053 char * addr;
3054 int flags;
3055
3056 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3057 if (fixed) {
3058 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3059 flags |= MAP_FIXED;
3060 }
3061
3062 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
3063 // to PROT_EXEC if executable when we commit the page.
3064 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
3065 flags, -1, 0);
3066
3067 if (addr != MAP_FAILED) {
3068 // anon_mmap() should only get called during VM initialization,
3069 // don't need lock (actually we can skip locking even it can be called
3070 // from multiple threads, because _highest_vm_reserved_address is just a
3071 // hint about the upper limit of non-stack memory regions.)
3072 if ((address)addr + bytes > _highest_vm_reserved_address) {
3073 _highest_vm_reserved_address = (address)addr + bytes;
3074 }
3075 }
3076
3077 return addr == MAP_FAILED ? NULL : addr;
3078 }
3079
3080 // Don't update _highest_vm_reserved_address, because there might be memory
3081 // regions above addr + size. If so, releasing a memory region only creates
3082 // a hole in the address space, it doesn't help prevent heap-stack collision.
3083 //
3084 static int anon_munmap(char * addr, size_t size) {
3085 return ::munmap(addr, size) == 0;
3086 }
3087
3088 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3089 size_t alignment_hint) {
3090 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3091 }
3092
3093 bool os::pd_release_memory(char* addr, size_t size) {
3094 return anon_munmap(addr, size);
3095 }
3096
3097 static address highest_vm_reserved_address() {
3098 return _highest_vm_reserved_address;
3099 }
3100
3101 static bool linux_mprotect(char* addr, size_t size, int prot) {
3102 // Linux wants the mprotect address argument to be page aligned.
3103 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3104
3105 // According to SUSv3, mprotect() should only be used with mappings
3106 // established by mmap(), and mmap() always maps whole pages. Unaligned
3107 // 'addr' likely indicates problem in the VM (e.g. trying to change
3108 // protection of malloc'ed or statically allocated memory). Check the
3109 // caller if you hit this assert.
3110 assert(addr == bottom, "sanity check");
3111
3112 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3113 return ::mprotect(bottom, size, prot) == 0;
3114 }
3115
3116 // Set protections specified
3117 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3118 bool is_committed) {
3119 unsigned int p = 0;
3120 switch (prot) {
3121 case MEM_PROT_NONE: p = PROT_NONE; break;
3122 case MEM_PROT_READ: p = PROT_READ; break;
3123 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3124 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3125 default:
3126 ShouldNotReachHere();
3127 }
3128 // is_committed is unused.
3129 return linux_mprotect(addr, bytes, p);
3130 }
3131
3132 bool os::guard_memory(char* addr, size_t size) {
3133 return linux_mprotect(addr, size, PROT_NONE);
3134 }
3135
3136 bool os::unguard_memory(char* addr, size_t size) {
3137 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3138 }
3139
3140 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3141 bool result = false;
3142 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3143 MAP_ANONYMOUS|MAP_PRIVATE,
3144 -1, 0);
3145 if (p != MAP_FAILED) {
3146 void *aligned_p = align_ptr_up(p, page_size);
3147
3148 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3149
3150 munmap(p, page_size * 2);
3151 }
3152
3153 if (warn && !result) {
3154 warning("TransparentHugePages is not supported by the operating system.");
3155 }
3156
3157 return result;
3158 }
3159
3160 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3161 bool result = false;
3162 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3163 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3164 -1, 0);
3165
3166 if (p != MAP_FAILED) {
3167 // We don't know if this really is a huge page or not.
3168 FILE *fp = fopen("/proc/self/maps", "r");
3169 if (fp) {
3170 while (!feof(fp)) {
3171 char chars[257];
3172 long x = 0;
3173 if (fgets(chars, sizeof(chars), fp)) {
3174 if (sscanf(chars, "%lx-%*x", &x) == 1
3175 && x == (long)p) {
3176 if (strstr (chars, "hugepage")) {
3177 result = true;
3178 break;
3179 }
3180 }
3181 }
3182 }
3183 fclose(fp);
3184 }
3185 munmap(p, page_size);
3186 }
3187
3188 if (warn && !result) {
3189 warning("HugeTLBFS is not supported by the operating system.");
3190 }
3191
3192 return result;
3193 }
3194
3195 /*
3196 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3197 *
3198 * From the coredump_filter documentation:
3199 *
3200 * - (bit 0) anonymous private memory
3201 * - (bit 1) anonymous shared memory
3202 * - (bit 2) file-backed private memory
3203 * - (bit 3) file-backed shared memory
3204 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3205 * effective only if the bit 2 is cleared)
3206 * - (bit 5) hugetlb private memory
3207 * - (bit 6) hugetlb shared memory
3208 */
3209 static void set_coredump_filter(void) {
3210 FILE *f;
3211 long cdm;
3212
3213 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3214 return;
3215 }
3216
3217 if (fscanf(f, "%lx", &cdm) != 1) {
3218 fclose(f);
3219 return;
3220 }
3221
3222 rewind(f);
3223
3224 if ((cdm & LARGEPAGES_BIT) == 0) {
3225 cdm |= LARGEPAGES_BIT;
3226 fprintf(f, "%#lx", cdm);
3227 }
3228
3229 fclose(f);
3230 }
3231
3232 // Large page support
3233
3234 static size_t _large_page_size = 0;
3235
3236 size_t os::Linux::find_large_page_size() {
3237 size_t large_page_size = 0;
3238
3239 // large_page_size on Linux is used to round up heap size. x86 uses either
3240 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3241 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3242 // page as large as 256M.
3243 //
3244 // Here we try to figure out page size by parsing /proc/meminfo and looking
3245 // for a line with the following format:
3246 // Hugepagesize: 2048 kB
3247 //
3248 // If we can't determine the value (e.g. /proc is not mounted, or the text
3249 // format has been changed), we'll use the largest page size supported by
3250 // the processor.
3251
3252 #ifndef ZERO
3253 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3254 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3255 #endif // ZERO
3256
3257 FILE *fp = fopen("/proc/meminfo", "r");
3258 if (fp) {
3259 while (!feof(fp)) {
3260 int x = 0;
3261 char buf[16];
3262 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3263 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3264 large_page_size = x * K;
3265 break;
3266 }
3267 } else {
3268 // skip to next line
3269 for (;;) {
3270 int ch = fgetc(fp);
3271 if (ch == EOF || ch == (int)'\n') break;
3272 }
3273 }
3274 }
3275 fclose(fp);
3276 }
3277
3278 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3279 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3280 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3281 proper_unit_for_byte_size(large_page_size));
3282 }
3283
3284 return large_page_size;
3285 }
3286
3287 size_t os::Linux::setup_large_page_size() {
3288 _large_page_size = Linux::find_large_page_size();
3289 size_t default_page_size = (size_t)Linux::page_size();
3290 if (_large_page_size > default_page_size) {
3291 _page_sizes[0] = _large_page_size;
3292 _page_sizes[1] = default_page_size;
3293 _page_sizes[2] = 0;
3294 }
3295
3296 return _large_page_size;
3297 }
3298
3299 bool os::Linux::setup_large_page_type(size_t page_size) {
3300 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3301 FLAG_IS_DEFAULT(UseSHM) &&
3302 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3303 // If UseLargePages is specified on the command line try all methods,
3304 // if it's default, then try only UseTransparentHugePages.
3305 if (FLAG_IS_DEFAULT(UseLargePages)) {
3306 UseTransparentHugePages = true;
3307 } else {
3308 UseHugeTLBFS = UseTransparentHugePages = UseSHM = true;
3309 }
3310 }
3311
3312 if (UseTransparentHugePages) {
3313 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3314 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3315 UseHugeTLBFS = false;
3316 UseSHM = false;
3317 return true;
3318 }
3319 UseTransparentHugePages = false;
3320 }
3321
3322 if (UseHugeTLBFS) {
3323 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3324 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3325 UseSHM = false;
3326 return true;
3327 }
3328 UseHugeTLBFS = false;
3329 }
3330
3331 return UseSHM;
3332 }
3333
3334 void os::large_page_init() {
3335 if (!UseLargePages) {
3336 UseHugeTLBFS = false;
3337 UseTransparentHugePages = false;
3338 UseSHM = false;
3339 return;
3340 }
3341
3342 size_t large_page_size = Linux::setup_large_page_size();
3343 UseLargePages = Linux::setup_large_page_type(large_page_size);
3344
3345 set_coredump_filter();
3346 }
3347
3348 #ifndef SHM_HUGETLB
3349 #define SHM_HUGETLB 04000
3350 #endif
3351
3352 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3353 // "exec" is passed in but not used. Creating the shared image for
3354 // the code cache doesn't have an SHM_X executable permission to check.
3355 assert(UseLargePages && UseSHM, "only for SHM large pages");
3356 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3357
3358 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3359 return NULL; // Fallback to small pages.
3360 }
3361
3362 key_t key = IPC_PRIVATE;
3363 char *addr;
3364
3365 bool warn_on_failure = UseLargePages &&
3366 (!FLAG_IS_DEFAULT(UseLargePages) ||
3367 !FLAG_IS_DEFAULT(UseSHM) ||
3368 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3369 );
3370 char msg[128];
3371
3372 // Create a large shared memory region to attach to based on size.
3373 // Currently, size is the total size of the heap
3374 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3375 if (shmid == -1) {
3376 // Possible reasons for shmget failure:
3377 // 1. shmmax is too small for Java heap.
3378 // > check shmmax value: cat /proc/sys/kernel/shmmax
3379 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3380 // 2. not enough large page memory.
3381 // > check available large pages: cat /proc/meminfo
3382 // > increase amount of large pages:
3383 // echo new_value > /proc/sys/vm/nr_hugepages
3384 // Note 1: different Linux may use different name for this property,
3385 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3386 // Note 2: it's possible there's enough physical memory available but
3387 // they are so fragmented after a long run that they can't
3388 // coalesce into large pages. Try to reserve large pages when
3389 // the system is still "fresh".
3390 if (warn_on_failure) {
3391 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3392 warning(msg);
3393 }
3394 return NULL;
3395 }
3396
3397 // attach to the region
3398 addr = (char*)shmat(shmid, req_addr, 0);
3399 int err = errno;
3400
3401 // Remove shmid. If shmat() is successful, the actual shared memory segment
3402 // will be deleted when it's detached by shmdt() or when the process
3403 // terminates. If shmat() is not successful this will remove the shared
3404 // segment immediately.
3405 shmctl(shmid, IPC_RMID, NULL);
3406
3407 if ((intptr_t)addr == -1) {
3408 if (warn_on_failure) {
3409 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3410 warning(msg);
3411 }
3412 return NULL;
3413 }
3414
3415 return addr;
3416 }
3417
3418 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3419 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3420
3421 bool warn_on_failure = UseLargePages &&
3422 (!FLAG_IS_DEFAULT(UseLargePages) ||
3423 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3424 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3425
3426 if (warn_on_failure) {
3427 char msg[128];
3428 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3429 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3430 warning(msg);
3431 }
3432 }
3433
3434 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3435 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3436 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3437 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3438
3439 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3440 char* addr = (char*)::mmap(req_addr, bytes, prot,
3441 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3442 -1, 0);
3443
3444 if (addr == MAP_FAILED) {
3445 warn_on_large_pages_failure(req_addr, bytes, errno);
3446 return NULL;
3447 }
3448
3449 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3450
3451 return addr;
3452 }
3453
3454 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3455 size_t large_page_size = os::large_page_size();
3456
3457 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3458
3459 // Allocate small pages.
3460
3461 char* start;
3462 if (req_addr != NULL) {
3463 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3464 assert(is_size_aligned(bytes, alignment), "Must be");
3465 start = os::reserve_memory(bytes, req_addr);
3466 assert(start == NULL || start == req_addr, "Must be");
3467 } else {
3468 start = os::reserve_memory_aligned(bytes, alignment);
3469 }
3470
3471 if (start == NULL) {
3472 return NULL;
3473 }
3474
3475 assert(is_ptr_aligned(start, alignment), "Must be");
3476
3477 // os::reserve_memory_special will record this memory area.
3478 // Need to release it here to prevent overlapping reservations.
3479 MemTracker::record_virtual_memory_release((address)start, bytes);
3480
3481 char* end = start + bytes;
3482
3483 // Find the regions of the allocated chunk that can be promoted to large pages.
3484 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3485 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3486
3487 size_t lp_bytes = lp_end - lp_start;
3488
3489 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3490
3491 if (lp_bytes == 0) {
3492 // The mapped region doesn't even span the start and the end of a large page.
3493 // Fall back to allocate a non-special area.
3494 ::munmap(start, end - start);
3495 return NULL;
3496 }
3497
3498 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3499
3500
3501 void* result;
3502
3503 if (start != lp_start) {
3504 result = ::mmap(start, lp_start - start, prot,
3505 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3506 -1, 0);
3507 if (result == MAP_FAILED) {
3508 ::munmap(lp_start, end - lp_start);
3509 return NULL;
3510 }
3511 }
3512
3513 result = ::mmap(lp_start, lp_bytes, prot,
3514 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3515 -1, 0);
3516 if (result == MAP_FAILED) {
3517 warn_on_large_pages_failure(req_addr, bytes, errno);
3518 // If the mmap above fails, the large pages region will be unmapped and we
3519 // have regions before and after with small pages. Release these regions.
3520 //
3521 // | mapped | unmapped | mapped |
3522 // ^ ^ ^ ^
3523 // start lp_start lp_end end
3524 //
3525 ::munmap(start, lp_start - start);
3526 ::munmap(lp_end, end - lp_end);
3527 return NULL;
3528 }
3529
3530 if (lp_end != end) {
3531 result = ::mmap(lp_end, end - lp_end, prot,
3532 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3533 -1, 0);
3534 if (result == MAP_FAILED) {
3535 ::munmap(start, lp_end - start);
3536 return NULL;
3537 }
3538 }
3539
3540 return start;
3541 }
3542
3543 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3544 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3545 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3546 assert(is_power_of_2(alignment), "Must be");
3547 assert(is_power_of_2(os::large_page_size()), "Must be");
3548 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3549
3550 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3551 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3552 } else {
3553 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3554 }
3555 }
3556
3557 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3558 assert(UseLargePages, "only for large pages");
3559
3560 char* addr;
3561 if (UseSHM) {
3562 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3563 } else {
3564 assert(UseHugeTLBFS, "must be");
3565 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3566 }
3567
3568 if (addr != NULL) {
3569 if (UseNUMAInterleaving) {
3570 numa_make_global(addr, bytes);
3571 }
3572
3573 // The memory is committed
3574 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3575 }
3576
3577 return addr;
3578 }
3579
3580 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3581 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3582 return shmdt(base) == 0;
3583 }
3584
3585 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3586 return pd_release_memory(base, bytes);
3587 }
3588
3589 bool os::release_memory_special(char* base, size_t bytes) {
3590 assert(UseLargePages, "only for large pages");
3591
3592 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3593
3594 bool res;
3595 if (UseSHM) {
3596 res = os::Linux::release_memory_special_shm(base, bytes);
3597 } else {
3598 assert(UseHugeTLBFS, "must be");
3599 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3600 }
3601
3602 if (res) {
3603 tkr.record((address)base, bytes);
3604 } else {
3605 tkr.discard();
3606 }
3607
3608 return res;
3609 }
3610
3611
3612 size_t os::large_page_size() {
3613 return _large_page_size;
3614 }
3615
3616 // With SysV SHM the entire memory region must be allocated as shared
3617 // memory.
3618 // HugeTLBFS allows application to commit large page memory on demand.
3619 // However, when committing memory with HugeTLBFS fails, the region
3620 // that was supposed to be committed will loose the old reservation
3621 // and allow other threads to steal that memory region. Because of this
3622 // behavior we can't commit HugeTLBFS memory.
3623 bool os::can_commit_large_page_memory() {
3624 return UseTransparentHugePages;
3625 }
3626
3627 bool os::can_execute_large_page_memory() {
3628 return UseTransparentHugePages || UseHugeTLBFS;
3629 }
3630
3631 // Reserve memory at an arbitrary address, only if that area is
3632 // available (and not reserved for something else).
3633
3634 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3635 const int max_tries = 10;
3636 char* base[max_tries];
3637 size_t size[max_tries];
3638 const size_t gap = 0x000000;
3639
3640 // Assert only that the size is a multiple of the page size, since
3641 // that's all that mmap requires, and since that's all we really know
3642 // about at this low abstraction level. If we need higher alignment,
3643 // we can either pass an alignment to this method or verify alignment
3644 // in one of the methods further up the call chain. See bug 5044738.
3645 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3646
3647 // Repeatedly allocate blocks until the block is allocated at the
3648 // right spot. Give up after max_tries. Note that reserve_memory() will
3649 // automatically update _highest_vm_reserved_address if the call is
3650 // successful. The variable tracks the highest memory address every reserved
3651 // by JVM. It is used to detect heap-stack collision if running with
3652 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3653 // space than needed, it could confuse the collision detecting code. To
3654 // solve the problem, save current _highest_vm_reserved_address and
3655 // calculate the correct value before return.
3656 address old_highest = _highest_vm_reserved_address;
3657
3658 // Linux mmap allows caller to pass an address as hint; give it a try first,
3659 // if kernel honors the hint then we can return immediately.
3660 char * addr = anon_mmap(requested_addr, bytes, false);
3661 if (addr == requested_addr) {
3662 return requested_addr;
3663 }
3664
3665 if (addr != NULL) {
3666 // mmap() is successful but it fails to reserve at the requested address
3667 anon_munmap(addr, bytes);
3668 }
3669
3670 int i;
3671 for (i = 0; i < max_tries; ++i) {
3672 base[i] = reserve_memory(bytes);
3673
3674 if (base[i] != NULL) {
3675 // Is this the block we wanted?
3676 if (base[i] == requested_addr) {
3677 size[i] = bytes;
3678 break;
3679 }
3680
3681 // Does this overlap the block we wanted? Give back the overlapped
3682 // parts and try again.
3683
3684 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3685 if (top_overlap >= 0 && top_overlap < bytes) {
3686 unmap_memory(base[i], top_overlap);
3687 base[i] += top_overlap;
3688 size[i] = bytes - top_overlap;
3689 } else {
3690 size_t bottom_overlap = base[i] + bytes - requested_addr;
3691 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3692 unmap_memory(requested_addr, bottom_overlap);
3693 size[i] = bytes - bottom_overlap;
3694 } else {
3695 size[i] = bytes;
3696 }
3697 }
3698 }
3699 }
3700
3701 // Give back the unused reserved pieces.
3702
3703 for (int j = 0; j < i; ++j) {
3704 if (base[j] != NULL) {
3705 unmap_memory(base[j], size[j]);
3706 }
3707 }
3708
3709 if (i < max_tries) {
3710 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3711 return requested_addr;
3712 } else {
3713 _highest_vm_reserved_address = old_highest;
3714 return NULL;
3715 }
3716 }
3717
3718 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3719 return ::read(fd, buf, nBytes);
3720 }
3721
3722 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3723 // Solaris uses poll(), linux uses park().
3724 // Poll() is likely a better choice, assuming that Thread.interrupt()
3725 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3726 // SIGSEGV, see 4355769.
3727
3728 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3729 assert(thread == Thread::current(), "thread consistency check");
3730
3731 ParkEvent * const slp = thread->_SleepEvent ;
3732 slp->reset() ;
3733 OrderAccess::fence() ;
3734
3735 if (interruptible) {
3736 jlong prevtime = javaTimeNanos();
3737
3738 for (;;) {
3739 if (os::is_interrupted(thread, true)) {
3740 return OS_INTRPT;
3741 }
3742
3743 jlong newtime = javaTimeNanos();
3744
3745 if (newtime - prevtime < 0) {
3746 // time moving backwards, should only happen if no monotonic clock
3747 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3748 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3749 } else {
3750 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3751 }
3752
3753 if(millis <= 0) {
3754 return OS_OK;
3755 }
3756
3757 prevtime = newtime;
3758
3759 {
3760 assert(thread->is_Java_thread(), "sanity check");
3761 JavaThread *jt = (JavaThread *) thread;
3762 ThreadBlockInVM tbivm(jt);
3763 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3764
3765 jt->set_suspend_equivalent();
3766 // cleared by handle_special_suspend_equivalent_condition() or
3767 // java_suspend_self() via check_and_wait_while_suspended()
3768
3769 slp->park(millis);
3770
3771 // were we externally suspended while we were waiting?
3772 jt->check_and_wait_while_suspended();
3773 }
3774 }
3775 } else {
3776 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3777 jlong prevtime = javaTimeNanos();
3778
3779 for (;;) {
3780 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3781 // the 1st iteration ...
3782 jlong newtime = javaTimeNanos();
3783
3784 if (newtime - prevtime < 0) {
3785 // time moving backwards, should only happen if no monotonic clock
3786 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3787 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3788 } else {
3789 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3790 }
3791
3792 if(millis <= 0) break ;
3793
3794 prevtime = newtime;
3795 slp->park(millis);
3796 }
3797 return OS_OK ;
3798 }
3799 }
3800
3801 int os::naked_sleep() {
3802 // %% make the sleep time an integer flag. for now use 1 millisec.
3803 return os::sleep(Thread::current(), 1, false);
3804 }
3805
3806 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3807 void os::infinite_sleep() {
3808 while (true) { // sleep forever ...
3809 ::sleep(100); // ... 100 seconds at a time
3810 }
3811 }
3812
3813 // Used to convert frequent JVM_Yield() to nops
3814 bool os::dont_yield() {
3815 return DontYieldALot;
3816 }
3817
3818 void os::yield() {
3819 sched_yield();
3820 }
3821
3822 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3823
3824 void os::yield_all(int attempts) {
3825 // Yields to all threads, including threads with lower priorities
3826 // Threads on Linux are all with same priority. The Solaris style
3827 // os::yield_all() with nanosleep(1ms) is not necessary.
3828 sched_yield();
3829 }
3830
3831 // Called from the tight loops to possibly influence time-sharing heuristics
3832 void os::loop_breaker(int attempts) {
3833 os::yield_all(attempts);
3834 }
3835
3836 ////////////////////////////////////////////////////////////////////////////////
3837 // thread priority support
3838
3839 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3840 // only supports dynamic priority, static priority must be zero. For real-time
3841 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3842 // However, for large multi-threaded applications, SCHED_RR is not only slower
3843 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3844 // of 5 runs - Sep 2005).
3845 //
3846 // The following code actually changes the niceness of kernel-thread/LWP. It
3847 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3848 // not the entire user process, and user level threads are 1:1 mapped to kernel
3849 // threads. It has always been the case, but could change in the future. For
3850 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3851 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3852
3853 int os::java_to_os_priority[CriticalPriority + 1] = {
3854 19, // 0 Entry should never be used
3855
3856 4, // 1 MinPriority
3857 3, // 2
3858 2, // 3
3859
3860 1, // 4
3861 0, // 5 NormPriority
3862 -1, // 6
3863
3864 -2, // 7
3865 -3, // 8
3866 -4, // 9 NearMaxPriority
3867
3868 -5, // 10 MaxPriority
3869
3870 -5 // 11 CriticalPriority
3871 };
3872
3873 static int prio_init() {
3874 if (ThreadPriorityPolicy == 1) {
3875 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3876 // if effective uid is not root. Perhaps, a more elegant way of doing
3877 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3878 if (geteuid() != 0) {
3879 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3880 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3881 }
3882 ThreadPriorityPolicy = 0;
3883 }
3884 }
3885 if (UseCriticalJavaThreadPriority) {
3886 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3887 }
3888 return 0;
3889 }
3890
3891 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3892 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3893
3894 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3895 return (ret == 0) ? OS_OK : OS_ERR;
3896 }
3897
3898 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3899 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3900 *priority_ptr = java_to_os_priority[NormPriority];
3901 return OS_OK;
3902 }
3903
3904 errno = 0;
3905 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3906 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3907 }
3908
3909 // Hint to the underlying OS that a task switch would not be good.
3910 // Void return because it's a hint and can fail.
3911 void os::hint_no_preempt() {}
3912
3913 ////////////////////////////////////////////////////////////////////////////////
3914 // suspend/resume support
3915
3916 // the low-level signal-based suspend/resume support is a remnant from the
3917 // old VM-suspension that used to be for java-suspension, safepoints etc,
3918 // within hotspot. Now there is a single use-case for this:
3919 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3920 // that runs in the watcher thread.
3921 // The remaining code is greatly simplified from the more general suspension
3922 // code that used to be used.
3923 //
3924 // The protocol is quite simple:
3925 // - suspend:
3926 // - sends a signal to the target thread
3927 // - polls the suspend state of the osthread using a yield loop
3928 // - target thread signal handler (SR_handler) sets suspend state
3929 // and blocks in sigsuspend until continued
3930 // - resume:
3931 // - sets target osthread state to continue
3932 // - sends signal to end the sigsuspend loop in the SR_handler
3933 //
3934 // Note that the SR_lock plays no role in this suspend/resume protocol.
3935 //
3936
3937 static void resume_clear_context(OSThread *osthread) {
3938 osthread->set_ucontext(NULL);
3939 osthread->set_siginfo(NULL);
3940 }
3941
3942 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3943 osthread->set_ucontext(context);
3944 osthread->set_siginfo(siginfo);
3945 }
3946
3947 //
3948 // Handler function invoked when a thread's execution is suspended or
3949 // resumed. We have to be careful that only async-safe functions are
3950 // called here (Note: most pthread functions are not async safe and
3951 // should be avoided.)
3952 //
3953 // Note: sigwait() is a more natural fit than sigsuspend() from an
3954 // interface point of view, but sigwait() prevents the signal hander
3955 // from being run. libpthread would get very confused by not having
3956 // its signal handlers run and prevents sigwait()'s use with the
3957 // mutex granting granting signal.
3958 //
3959 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3960 //
3961 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3962 // Save and restore errno to avoid confusing native code with EINTR
3963 // after sigsuspend.
3964 int old_errno = errno;
3965
3966 Thread* thread = Thread::current();
3967 OSThread* osthread = thread->osthread();
3968 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3969
3970 os::SuspendResume::State current = osthread->sr.state();
3971 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3972 suspend_save_context(osthread, siginfo, context);
3973
3974 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3975 os::SuspendResume::State state = osthread->sr.suspended();
3976 if (state == os::SuspendResume::SR_SUSPENDED) {
3977 sigset_t suspend_set; // signals for sigsuspend()
3978
3979 // get current set of blocked signals and unblock resume signal
3980 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3981 sigdelset(&suspend_set, SR_signum);
3982
3983 sr_semaphore.signal();
3984 // wait here until we are resumed
3985 while (1) {
3986 sigsuspend(&suspend_set);
3987
3988 os::SuspendResume::State result = osthread->sr.running();
3989 if (result == os::SuspendResume::SR_RUNNING) {
3990 sr_semaphore.signal();
3991 break;
3992 }
3993 }
3994
3995 } else if (state == os::SuspendResume::SR_RUNNING) {
3996 // request was cancelled, continue
3997 } else {
3998 ShouldNotReachHere();
3999 }
4000
4001 resume_clear_context(osthread);
4002 } else if (current == os::SuspendResume::SR_RUNNING) {
4003 // request was cancelled, continue
4004 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4005 // ignore
4006 } else {
4007 // ignore
4008 }
4009
4010 errno = old_errno;
4011 }
4012
4013
4014 static int SR_initialize() {
4015 struct sigaction act;
4016 char *s;
4017 /* Get signal number to use for suspend/resume */
4018 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4019 int sig = ::strtol(s, 0, 10);
4020 if (sig > 0 || sig < _NSIG) {
4021 SR_signum = sig;
4022 }
4023 }
4024
4025 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4026 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4027
4028 sigemptyset(&SR_sigset);
4029 sigaddset(&SR_sigset, SR_signum);
4030
4031 /* Set up signal handler for suspend/resume */
4032 act.sa_flags = SA_RESTART|SA_SIGINFO;
4033 act.sa_handler = (void (*)(int)) SR_handler;
4034
4035 // SR_signum is blocked by default.
4036 // 4528190 - We also need to block pthread restart signal (32 on all
4037 // supported Linux platforms). Note that LinuxThreads need to block
4038 // this signal for all threads to work properly. So we don't have
4039 // to use hard-coded signal number when setting up the mask.
4040 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4041
4042 if (sigaction(SR_signum, &act, 0) == -1) {
4043 return -1;
4044 }
4045
4046 // Save signal flag
4047 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4048 return 0;
4049 }
4050
4051 static int SR_finalize() {
4052 return 0;
4053 }
4054
4055 static int sr_notify(OSThread* osthread) {
4056 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4057 assert_status(status == 0, status, "pthread_kill");
4058 return status;
4059 }
4060
4061 // "Randomly" selected value for how long we want to spin
4062 // before bailing out on suspending a thread, also how often
4063 // we send a signal to a thread we want to resume
4064 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4065 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4066
4067 // returns true on success and false on error - really an error is fatal
4068 // but this seems the normal response to library errors
4069 static bool do_suspend(OSThread* osthread) {
4070 assert(osthread->sr.is_running(), "thread should be running");
4071 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4072
4073 // mark as suspended and send signal
4074 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4075 // failed to switch, state wasn't running?
4076 ShouldNotReachHere();
4077 return false;
4078 }
4079
4080 if (sr_notify(osthread) != 0) {
4081 ShouldNotReachHere();
4082 }
4083
4084 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4085 while (true) {
4086 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4087 break;
4088 } else {
4089 // timeout
4090 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4091 if (cancelled == os::SuspendResume::SR_RUNNING) {
4092 return false;
4093 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4094 // make sure that we consume the signal on the semaphore as well
4095 sr_semaphore.wait();
4096 break;
4097 } else {
4098 ShouldNotReachHere();
4099 return false;
4100 }
4101 }
4102 }
4103
4104 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4105 return true;
4106 }
4107
4108 static void do_resume(OSThread* osthread) {
4109 assert(osthread->sr.is_suspended(), "thread should be suspended");
4110 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4111
4112 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4113 // failed to switch to WAKEUP_REQUEST
4114 ShouldNotReachHere();
4115 return;
4116 }
4117
4118 while (true) {
4119 if (sr_notify(osthread) == 0) {
4120 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4121 if (osthread->sr.is_running()) {
4122 return;
4123 }
4124 }
4125 } else {
4126 ShouldNotReachHere();
4127 }
4128 }
4129
4130 guarantee(osthread->sr.is_running(), "Must be running!");
4131 }
4132
4133 ////////////////////////////////////////////////////////////////////////////////
4134 // interrupt support
4135
4136 void os::interrupt(Thread* thread) {
4137 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4138 "possibility of dangling Thread pointer");
4139
4140 OSThread* osthread = thread->osthread();
4141
4142 if (!osthread->interrupted()) {
4143 osthread->set_interrupted(true);
4144 // More than one thread can get here with the same value of osthread,
4145 // resulting in multiple notifications. We do, however, want the store
4146 // to interrupted() to be visible to other threads before we execute unpark().
4147 OrderAccess::fence();
4148 ParkEvent * const slp = thread->_SleepEvent ;
4149 if (slp != NULL) slp->unpark() ;
4150 }
4151
4152 // For JSR166. Unpark even if interrupt status already was set
4153 if (thread->is_Java_thread())
4154 ((JavaThread*)thread)->parker()->unpark();
4155
4156 ParkEvent * ev = thread->_ParkEvent ;
4157 if (ev != NULL) ev->unpark() ;
4158
4159 }
4160
4161 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4162 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4163 "possibility of dangling Thread pointer");
4164
4165 OSThread* osthread = thread->osthread();
4166
4167 bool interrupted = osthread->interrupted();
4168
4169 if (interrupted && clear_interrupted) {
4170 osthread->set_interrupted(false);
4171 // consider thread->_SleepEvent->reset() ... optional optimization
4172 }
4173
4174 return interrupted;
4175 }
4176
4177 ///////////////////////////////////////////////////////////////////////////////////
4178 // signal handling (except suspend/resume)
4179
4180 // This routine may be used by user applications as a "hook" to catch signals.
4181 // The user-defined signal handler must pass unrecognized signals to this
4182 // routine, and if it returns true (non-zero), then the signal handler must
4183 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4184 // routine will never retun false (zero), but instead will execute a VM panic
4185 // routine kill the process.
4186 //
4187 // If this routine returns false, it is OK to call it again. This allows
4188 // the user-defined signal handler to perform checks either before or after
4189 // the VM performs its own checks. Naturally, the user code would be making
4190 // a serious error if it tried to handle an exception (such as a null check
4191 // or breakpoint) that the VM was generating for its own correct operation.
4192 //
4193 // This routine may recognize any of the following kinds of signals:
4194 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4195 // It should be consulted by handlers for any of those signals.
4196 //
4197 // The caller of this routine must pass in the three arguments supplied
4198 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4199 // field of the structure passed to sigaction(). This routine assumes that
4200 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4201 //
4202 // Note that the VM will print warnings if it detects conflicting signal
4203 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4204 //
4205 extern "C" JNIEXPORT int
4206 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4207 void* ucontext, int abort_if_unrecognized);
4208
4209 void signalHandler(int sig, siginfo_t* info, void* uc) {
4210 assert(info != NULL && uc != NULL, "it must be old kernel");
4211 JVM_handle_linux_signal(sig, info, uc, true);
4212 }
4213
4214
4215 // This boolean allows users to forward their own non-matching signals
4216 // to JVM_handle_linux_signal, harmlessly.
4217 bool os::Linux::signal_handlers_are_installed = false;
4218
4219 // For signal-chaining
4220 struct sigaction os::Linux::sigact[MAXSIGNUM];
4221 unsigned int os::Linux::sigs = 0;
4222 bool os::Linux::libjsig_is_loaded = false;
4223 typedef struct sigaction *(*get_signal_t)(int);
4224 get_signal_t os::Linux::get_signal_action = NULL;
4225
4226 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4227 struct sigaction *actp = NULL;
4228
4229 if (libjsig_is_loaded) {
4230 // Retrieve the old signal handler from libjsig
4231 actp = (*get_signal_action)(sig);
4232 }
4233 if (actp == NULL) {
4234 // Retrieve the preinstalled signal handler from jvm
4235 actp = get_preinstalled_handler(sig);
4236 }
4237
4238 return actp;
4239 }
4240
4241 static bool call_chained_handler(struct sigaction *actp, int sig,
4242 siginfo_t *siginfo, void *context) {
4243 // Call the old signal handler
4244 if (actp->sa_handler == SIG_DFL) {
4245 // It's more reasonable to let jvm treat it as an unexpected exception
4246 // instead of taking the default action.
4247 return false;
4248 } else if (actp->sa_handler != SIG_IGN) {
4249 if ((actp->sa_flags & SA_NODEFER) == 0) {
4250 // automaticlly block the signal
4251 sigaddset(&(actp->sa_mask), sig);
4252 }
4253
4254 sa_handler_t hand;
4255 sa_sigaction_t sa;
4256 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4257 // retrieve the chained handler
4258 if (siginfo_flag_set) {
4259 sa = actp->sa_sigaction;
4260 } else {
4261 hand = actp->sa_handler;
4262 }
4263
4264 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4265 actp->sa_handler = SIG_DFL;
4266 }
4267
4268 // try to honor the signal mask
4269 sigset_t oset;
4270 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4271
4272 // call into the chained handler
4273 if (siginfo_flag_set) {
4274 (*sa)(sig, siginfo, context);
4275 } else {
4276 (*hand)(sig);
4277 }
4278
4279 // restore the signal mask
4280 pthread_sigmask(SIG_SETMASK, &oset, 0);
4281 }
4282 // Tell jvm's signal handler the signal is taken care of.
4283 return true;
4284 }
4285
4286 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4287 bool chained = false;
4288 // signal-chaining
4289 if (UseSignalChaining) {
4290 struct sigaction *actp = get_chained_signal_action(sig);
4291 if (actp != NULL) {
4292 chained = call_chained_handler(actp, sig, siginfo, context);
4293 }
4294 }
4295 return chained;
4296 }
4297
4298 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4299 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4300 return &sigact[sig];
4301 }
4302 return NULL;
4303 }
4304
4305 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4306 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4307 sigact[sig] = oldAct;
4308 sigs |= (unsigned int)1 << sig;
4309 }
4310
4311 // for diagnostic
4312 int os::Linux::sigflags[MAXSIGNUM];
4313
4314 int os::Linux::get_our_sigflags(int sig) {
4315 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4316 return sigflags[sig];
4317 }
4318
4319 void os::Linux::set_our_sigflags(int sig, int flags) {
4320 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4321 sigflags[sig] = flags;
4322 }
4323
4324 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4325 // Check for overwrite.
4326 struct sigaction oldAct;
4327 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4328
4329 void* oldhand = oldAct.sa_sigaction
4330 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4331 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4332 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4333 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4334 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4335 if (AllowUserSignalHandlers || !set_installed) {
4336 // Do not overwrite; user takes responsibility to forward to us.
4337 return;
4338 } else if (UseSignalChaining) {
4339 // save the old handler in jvm
4340 save_preinstalled_handler(sig, oldAct);
4341 // libjsig also interposes the sigaction() call below and saves the
4342 // old sigaction on it own.
4343 } else {
4344 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4345 "%#lx for signal %d.", (long)oldhand, sig));
4346 }
4347 }
4348
4349 struct sigaction sigAct;
4350 sigfillset(&(sigAct.sa_mask));
4351 sigAct.sa_handler = SIG_DFL;
4352 if (!set_installed) {
4353 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4354 } else {
4355 sigAct.sa_sigaction = signalHandler;
4356 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4357 }
4358 // Save flags, which are set by ours
4359 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4360 sigflags[sig] = sigAct.sa_flags;
4361
4362 int ret = sigaction(sig, &sigAct, &oldAct);
4363 assert(ret == 0, "check");
4364
4365 void* oldhand2 = oldAct.sa_sigaction
4366 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4367 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4368 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4369 }
4370
4371 // install signal handlers for signals that HotSpot needs to
4372 // handle in order to support Java-level exception handling.
4373
4374 void os::Linux::install_signal_handlers() {
4375 if (!signal_handlers_are_installed) {
4376 signal_handlers_are_installed = true;
4377
4378 // signal-chaining
4379 typedef void (*signal_setting_t)();
4380 signal_setting_t begin_signal_setting = NULL;
4381 signal_setting_t end_signal_setting = NULL;
4382 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4383 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4384 if (begin_signal_setting != NULL) {
4385 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4386 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4387 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4388 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4389 libjsig_is_loaded = true;
4390 assert(UseSignalChaining, "should enable signal-chaining");
4391 }
4392 if (libjsig_is_loaded) {
4393 // Tell libjsig jvm is setting signal handlers
4394 (*begin_signal_setting)();
4395 }
4396
4397 set_signal_handler(SIGSEGV, true);
4398 set_signal_handler(SIGPIPE, true);
4399 set_signal_handler(SIGBUS, true);
4400 set_signal_handler(SIGILL, true);
4401 set_signal_handler(SIGFPE, true);
4402 set_signal_handler(SIGXFSZ, true);
4403
4404 if (libjsig_is_loaded) {
4405 // Tell libjsig jvm finishes setting signal handlers
4406 (*end_signal_setting)();
4407 }
4408
4409 // We don't activate signal checker if libjsig is in place, we trust ourselves
4410 // and if UserSignalHandler is installed all bets are off.
4411 // Log that signal checking is off only if -verbose:jni is specified.
4412 if (CheckJNICalls) {
4413 if (libjsig_is_loaded) {
4414 if (PrintJNIResolving) {
4415 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4416 }
4417 check_signals = false;
4418 }
4419 if (AllowUserSignalHandlers) {
4420 if (PrintJNIResolving) {
4421 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4422 }
4423 check_signals = false;
4424 }
4425 }
4426 }
4427 }
4428
4429 // This is the fastest way to get thread cpu time on Linux.
4430 // Returns cpu time (user+sys) for any thread, not only for current.
4431 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4432 // It might work on 2.6.10+ with a special kernel/glibc patch.
4433 // For reference, please, see IEEE Std 1003.1-2004:
4434 // http://www.unix.org/single_unix_specification
4435
4436 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4437 struct timespec tp;
4438 int rc = os::Linux::clock_gettime(clockid, &tp);
4439 assert(rc == 0, "clock_gettime is expected to return 0 code");
4440
4441 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4442 }
4443
4444 /////
4445 // glibc on Linux platform uses non-documented flag
4446 // to indicate, that some special sort of signal
4447 // trampoline is used.
4448 // We will never set this flag, and we should
4449 // ignore this flag in our diagnostic
4450 #ifdef SIGNIFICANT_SIGNAL_MASK
4451 #undef SIGNIFICANT_SIGNAL_MASK
4452 #endif
4453 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4454
4455 static const char* get_signal_handler_name(address handler,
4456 char* buf, int buflen) {
4457 int offset;
4458 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4459 if (found) {
4460 // skip directory names
4461 const char *p1, *p2;
4462 p1 = buf;
4463 size_t len = strlen(os::file_separator());
4464 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4465 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4466 } else {
4467 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4468 }
4469 return buf;
4470 }
4471
4472 static void print_signal_handler(outputStream* st, int sig,
4473 char* buf, size_t buflen) {
4474 struct sigaction sa;
4475
4476 sigaction(sig, NULL, &sa);
4477
4478 // See comment for SIGNIFICANT_SIGNAL_MASK define
4479 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4480
4481 st->print("%s: ", os::exception_name(sig, buf, buflen));
4482
4483 address handler = (sa.sa_flags & SA_SIGINFO)
4484 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4485 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4486
4487 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4488 st->print("SIG_DFL");
4489 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4490 st->print("SIG_IGN");
4491 } else {
4492 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4493 }
4494
4495 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4496
4497 address rh = VMError::get_resetted_sighandler(sig);
4498 // May be, handler was resetted by VMError?
4499 if(rh != NULL) {
4500 handler = rh;
4501 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4502 }
4503
4504 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4505
4506 // Check: is it our handler?
4507 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4508 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4509 // It is our signal handler
4510 // check for flags, reset system-used one!
4511 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4512 st->print(
4513 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4514 os::Linux::get_our_sigflags(sig));
4515 }
4516 }
4517 st->cr();
4518 }
4519
4520
4521 #define DO_SIGNAL_CHECK(sig) \
4522 if (!sigismember(&check_signal_done, sig)) \
4523 os::Linux::check_signal_handler(sig)
4524
4525 // This method is a periodic task to check for misbehaving JNI applications
4526 // under CheckJNI, we can add any periodic checks here
4527
4528 void os::run_periodic_checks() {
4529
4530 if (check_signals == false) return;
4531
4532 // SEGV and BUS if overridden could potentially prevent
4533 // generation of hs*.log in the event of a crash, debugging
4534 // such a case can be very challenging, so we absolutely
4535 // check the following for a good measure:
4536 DO_SIGNAL_CHECK(SIGSEGV);
4537 DO_SIGNAL_CHECK(SIGILL);
4538 DO_SIGNAL_CHECK(SIGFPE);
4539 DO_SIGNAL_CHECK(SIGBUS);
4540 DO_SIGNAL_CHECK(SIGPIPE);
4541 DO_SIGNAL_CHECK(SIGXFSZ);
4542
4543
4544 // ReduceSignalUsage allows the user to override these handlers
4545 // see comments at the very top and jvm_solaris.h
4546 if (!ReduceSignalUsage) {
4547 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4548 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4549 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4550 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4551 }
4552
4553 DO_SIGNAL_CHECK(SR_signum);
4554 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4555 }
4556
4557 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4558
4559 static os_sigaction_t os_sigaction = NULL;
4560
4561 void os::Linux::check_signal_handler(int sig) {
4562 char buf[O_BUFLEN];
4563 address jvmHandler = NULL;
4564
4565
4566 struct sigaction act;
4567 if (os_sigaction == NULL) {
4568 // only trust the default sigaction, in case it has been interposed
4569 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4570 if (os_sigaction == NULL) return;
4571 }
4572
4573 os_sigaction(sig, (struct sigaction*)NULL, &act);
4574
4575
4576 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4577
4578 address thisHandler = (act.sa_flags & SA_SIGINFO)
4579 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4580 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4581
4582
4583 switch(sig) {
4584 case SIGSEGV:
4585 case SIGBUS:
4586 case SIGFPE:
4587 case SIGPIPE:
4588 case SIGILL:
4589 case SIGXFSZ:
4590 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4591 break;
4592
4593 case SHUTDOWN1_SIGNAL:
4594 case SHUTDOWN2_SIGNAL:
4595 case SHUTDOWN3_SIGNAL:
4596 case BREAK_SIGNAL:
4597 jvmHandler = (address)user_handler();
4598 break;
4599
4600 case INTERRUPT_SIGNAL:
4601 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4602 break;
4603
4604 default:
4605 if (sig == SR_signum) {
4606 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4607 } else {
4608 return;
4609 }
4610 break;
4611 }
4612
4613 if (thisHandler != jvmHandler) {
4614 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4615 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4616 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4617 // No need to check this sig any longer
4618 sigaddset(&check_signal_done, sig);
4619 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4620 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4621 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4622 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4623 // No need to check this sig any longer
4624 sigaddset(&check_signal_done, sig);
4625 }
4626
4627 // Dump all the signal
4628 if (sigismember(&check_signal_done, sig)) {
4629 print_signal_handlers(tty, buf, O_BUFLEN);
4630 }
4631 }
4632
4633 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4634
4635 extern bool signal_name(int signo, char* buf, size_t len);
4636
4637 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4638 if (0 < exception_code && exception_code <= SIGRTMAX) {
4639 // signal
4640 if (!signal_name(exception_code, buf, size)) {
4641 jio_snprintf(buf, size, "SIG%d", exception_code);
4642 }
4643 return buf;
4644 } else {
4645 return NULL;
4646 }
4647 }
4648
4649 // this is called _before_ the most of global arguments have been parsed
4650 void os::init(void) {
4651 char dummy; /* used to get a guess on initial stack address */
4652 // first_hrtime = gethrtime();
4653
4654 // With LinuxThreads the JavaMain thread pid (primordial thread)
4655 // is different than the pid of the java launcher thread.
4656 // So, on Linux, the launcher thread pid is passed to the VM
4657 // via the sun.java.launcher.pid property.
4658 // Use this property instead of getpid() if it was correctly passed.
4659 // See bug 6351349.
4660 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4661
4662 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4663
4664 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4665
4666 init_random(1234567);
4667
4668 ThreadCritical::initialize();
4669
4670 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4671 if (Linux::page_size() == -1) {
4672 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4673 strerror(errno)));
4674 }
4675 init_page_sizes((size_t) Linux::page_size());
4676
4677 Linux::initialize_system_info();
4678
4679 // main_thread points to the aboriginal thread
4680 Linux::_main_thread = pthread_self();
4681
4682 Linux::clock_init();
4683 initial_time_count = os::elapsed_counter();
4684 pthread_mutex_init(&dl_mutex, NULL);
4685
4686 // If the pagesize of the VM is greater than 8K determine the appropriate
4687 // number of initial guard pages. The user can change this with the
4688 // command line arguments, if needed.
4689 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4690 StackYellowPages = 1;
4691 StackRedPages = 1;
4692 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4693 }
4694 }
4695
4696 // To install functions for atexit system call
4697 extern "C" {
4698 static void perfMemory_exit_helper() {
4699 perfMemory_exit();
4700 }
4701 }
4702
4703 // this is called _after_ the global arguments have been parsed
4704 jint os::init_2(void)
4705 {
4706 Linux::fast_thread_clock_init();
4707
4708 // Allocate a single page and mark it as readable for safepoint polling
4709 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4710 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4711
4712 os::set_polling_page( polling_page );
4713
4714 #ifndef PRODUCT
4715 if(Verbose && PrintMiscellaneous)
4716 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4717 #endif
4718
4719 if (!UseMembar) {
4720 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4721 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4722 os::set_memory_serialize_page( mem_serialize_page );
4723
4724 #ifndef PRODUCT
4725 if(Verbose && PrintMiscellaneous)
4726 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4727 #endif
4728 }
4729
4730 os::large_page_init();
4731
4732 // initialize suspend/resume support - must do this before signal_sets_init()
4733 if (SR_initialize() != 0) {
4734 perror("SR_initialize failed");
4735 return JNI_ERR;
4736 }
4737
4738 Linux::signal_sets_init();
4739 Linux::install_signal_handlers();
4740
4741 // Check minimum allowable stack size for thread creation and to initialize
4742 // the java system classes, including StackOverflowError - depends on page
4743 // size. Add a page for compiler2 recursion in main thread.
4744 // Add in 2*BytesPerWord times page size to account for VM stack during
4745 // class initialization depending on 32 or 64 bit VM.
4746 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4747 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4748 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4749
4750 size_t threadStackSizeInBytes = ThreadStackSize * K;
4751 if (threadStackSizeInBytes != 0 &&
4752 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4753 tty->print_cr("\nThe stack size specified is too small, "
4754 "Specify at least %dk",
4755 os::Linux::min_stack_allowed/ K);
4756 return JNI_ERR;
4757 }
4758
4759 // Make the stack size a multiple of the page size so that
4760 // the yellow/red zones can be guarded.
4761 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4762 vm_page_size()));
4763
4764 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4765
4766 Linux::libpthread_init();
4767 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4768 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4769 Linux::glibc_version(), Linux::libpthread_version(),
4770 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4771 }
4772
4773 if (UseNUMA) {
4774 if (!Linux::libnuma_init()) {
4775 UseNUMA = false;
4776 } else {
4777 if ((Linux::numa_max_node() < 1)) {
4778 // There's only one node(they start from 0), disable NUMA.
4779 UseNUMA = false;
4780 }
4781 }
4782 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4783 // we can make the adaptive lgrp chunk resizing work. If the user specified
4784 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4785 // disable adaptive resizing.
4786 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4787 if (FLAG_IS_DEFAULT(UseNUMA)) {
4788 UseNUMA = false;
4789 } else {
4790 if (FLAG_IS_DEFAULT(UseLargePages) &&
4791 FLAG_IS_DEFAULT(UseSHM) &&
4792 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4793 UseLargePages = false;
4794 } else {
4795 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4796 UseAdaptiveSizePolicy = false;
4797 UseAdaptiveNUMAChunkSizing = false;
4798 }
4799 }
4800 }
4801 if (!UseNUMA && ForceNUMA) {
4802 UseNUMA = true;
4803 }
4804 }
4805
4806 if (MaxFDLimit) {
4807 // set the number of file descriptors to max. print out error
4808 // if getrlimit/setrlimit fails but continue regardless.
4809 struct rlimit nbr_files;
4810 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4811 if (status != 0) {
4812 if (PrintMiscellaneous && (Verbose || WizardMode))
4813 perror("os::init_2 getrlimit failed");
4814 } else {
4815 nbr_files.rlim_cur = nbr_files.rlim_max;
4816 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4817 if (status != 0) {
4818 if (PrintMiscellaneous && (Verbose || WizardMode))
4819 perror("os::init_2 setrlimit failed");
4820 }
4821 }
4822 }
4823
4824 // Initialize lock used to serialize thread creation (see os::create_thread)
4825 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4826
4827 // at-exit methods are called in the reverse order of their registration.
4828 // atexit functions are called on return from main or as a result of a
4829 // call to exit(3C). There can be only 32 of these functions registered
4830 // and atexit() does not set errno.
4831
4832 if (PerfAllowAtExitRegistration) {
4833 // only register atexit functions if PerfAllowAtExitRegistration is set.
4834 // atexit functions can be delayed until process exit time, which
4835 // can be problematic for embedded VM situations. Embedded VMs should
4836 // call DestroyJavaVM() to assure that VM resources are released.
4837
4838 // note: perfMemory_exit_helper atexit function may be removed in
4839 // the future if the appropriate cleanup code can be added to the
4840 // VM_Exit VMOperation's doit method.
4841 if (atexit(perfMemory_exit_helper) != 0) {
4842 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4843 }
4844 }
4845
4846 // initialize thread priority policy
4847 prio_init();
4848
4849 return JNI_OK;
4850 }
4851
4852 // this is called at the end of vm_initialization
4853 void os::init_3(void)
4854 {
4855 #ifdef JAVASE_EMBEDDED
4856 // Start the MemNotifyThread
4857 if (LowMemoryProtection) {
4858 MemNotifyThread::start();
4859 }
4860 return;
4861 #endif
4862 }
4863
4864 // Mark the polling page as unreadable
4865 void os::make_polling_page_unreadable(void) {
4866 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4867 fatal("Could not disable polling page");
4868 };
4869
4870 // Mark the polling page as readable
4871 void os::make_polling_page_readable(void) {
4872 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4873 fatal("Could not enable polling page");
4874 }
4875 };
4876
4877 int os::active_processor_count() {
4878 // Linux doesn't yet have a (official) notion of processor sets,
4879 // so just return the number of online processors.
4880 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4881 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4882 return online_cpus;
4883 }
4884
4885 void os::set_native_thread_name(const char *name) {
4886 // Not yet implemented.
4887 return;
4888 }
4889
4890 bool os::distribute_processes(uint length, uint* distribution) {
4891 // Not yet implemented.
4892 return false;
4893 }
4894
4895 bool os::bind_to_processor(uint processor_id) {
4896 // Not yet implemented.
4897 return false;
4898 }
4899
4900 ///
4901
4902 void os::SuspendedThreadTask::internal_do_task() {
4903 if (do_suspend(_thread->osthread())) {
4904 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4905 do_task(context);
4906 do_resume(_thread->osthread());
4907 }
4908 }
4909
4910 class PcFetcher : public os::SuspendedThreadTask {
4911 public:
4912 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4913 ExtendedPC result();
4914 protected:
4915 void do_task(const os::SuspendedThreadTaskContext& context);
4916 private:
4917 ExtendedPC _epc;
4918 };
4919
4920 ExtendedPC PcFetcher::result() {
4921 guarantee(is_done(), "task is not done yet.");
4922 return _epc;
4923 }
4924
4925 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4926 Thread* thread = context.thread();
4927 OSThread* osthread = thread->osthread();
4928 if (osthread->ucontext() != NULL) {
4929 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4930 } else {
4931 // NULL context is unexpected, double-check this is the VMThread
4932 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4933 }
4934 }
4935
4936 // Suspends the target using the signal mechanism and then grabs the PC before
4937 // resuming the target. Used by the flat-profiler only
4938 ExtendedPC os::get_thread_pc(Thread* thread) {
4939 // Make sure that it is called by the watcher for the VMThread
4940 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4941 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4942
4943 PcFetcher fetcher(thread);
4944 fetcher.run();
4945 return fetcher.result();
4946 }
4947
4948 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4949 {
4950 if (is_NPTL()) {
4951 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4952 } else {
4953 #ifndef IA64
4954 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4955 // word back to default 64bit precision if condvar is signaled. Java
4956 // wants 53bit precision. Save and restore current value.
4957 int fpu = get_fpu_control_word();
4958 #endif // IA64
4959 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4960 #ifndef IA64
4961 set_fpu_control_word(fpu);
4962 #endif // IA64
4963 return status;
4964 }
4965 }
4966
4967 ////////////////////////////////////////////////////////////////////////////////
4968 // debug support
4969
4970 static address same_page(address x, address y) {
4971 int page_bits = -os::vm_page_size();
4972 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4973 return x;
4974 else if (x > y)
4975 return (address)(intptr_t(y) | ~page_bits) + 1;
4976 else
4977 return (address)(intptr_t(y) & page_bits);
4978 }
4979
4980 bool os::find(address addr, outputStream* st) {
4981 Dl_info dlinfo;
4982 memset(&dlinfo, 0, sizeof(dlinfo));
4983 if (dladdr(addr, &dlinfo)) {
4984 st->print(PTR_FORMAT ": ", addr);
4985 if (dlinfo.dli_sname != NULL) {
4986 st->print("%s+%#x", dlinfo.dli_sname,
4987 addr - (intptr_t)dlinfo.dli_saddr);
4988 } else if (dlinfo.dli_fname) {
4989 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4990 } else {
4991 st->print("<absolute address>");
4992 }
4993 if (dlinfo.dli_fname) {
4994 st->print(" in %s", dlinfo.dli_fname);
4995 }
4996 if (dlinfo.dli_fbase) {
4997 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4998 }
4999 st->cr();
5000
5001 if (Verbose) {
5002 // decode some bytes around the PC
5003 address begin = same_page(addr-40, addr);
5004 address end = same_page(addr+40, addr);
5005 address lowest = (address) dlinfo.dli_sname;
5006 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5007 if (begin < lowest) begin = lowest;
5008 Dl_info dlinfo2;
5009 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
5010 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5011 end = (address) dlinfo2.dli_saddr;
5012 Disassembler::decode(begin, end, st);
5013 }
5014 return true;
5015 }
5016 return false;
5017 }
5018
5019 ////////////////////////////////////////////////////////////////////////////////
5020 // misc
5021
5022 // This does not do anything on Linux. This is basically a hook for being
5023 // able to use structured exception handling (thread-local exception filters)
5024 // on, e.g., Win32.
5025 void
5026 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5027 JavaCallArguments* args, Thread* thread) {
5028 f(value, method, args, thread);
5029 }
5030
5031 void os::print_statistics() {
5032 }
5033
5034 int os::message_box(const char* title, const char* message) {
5035 int i;
5036 fdStream err(defaultStream::error_fd());
5037 for (i = 0; i < 78; i++) err.print_raw("=");
5038 err.cr();
5039 err.print_raw_cr(title);
5040 for (i = 0; i < 78; i++) err.print_raw("-");
5041 err.cr();
5042 err.print_raw_cr(message);
5043 for (i = 0; i < 78; i++) err.print_raw("=");
5044 err.cr();
5045
5046 char buf[16];
5047 // Prevent process from exiting upon "read error" without consuming all CPU
5048 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5049
5050 return buf[0] == 'y' || buf[0] == 'Y';
5051 }
5052
5053 int os::stat(const char *path, struct stat *sbuf) {
5054 char pathbuf[MAX_PATH];
5055 if (strlen(path) > MAX_PATH - 1) {
5056 errno = ENAMETOOLONG;
5057 return -1;
5058 }
5059 os::native_path(strcpy(pathbuf, path));
5060 return ::stat(pathbuf, sbuf);
5061 }
5062
5063 bool os::check_heap(bool force) {
5064 return true;
5065 }
5066
5067 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5068 return ::vsnprintf(buf, count, format, args);
5069 }
5070
5071 // Is a (classpath) directory empty?
5072 bool os::dir_is_empty(const char* path) {
5073 DIR *dir = NULL;
5074 struct dirent *ptr;
5075
5076 dir = opendir(path);
5077 if (dir == NULL) return true;
5078
5079 /* Scan the directory */
5080 bool result = true;
5081 char buf[sizeof(struct dirent) + MAX_PATH];
5082 while (result && (ptr = ::readdir(dir)) != NULL) {
5083 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5084 result = false;
5085 }
5086 }
5087 closedir(dir);
5088 return result;
5089 }
5090
5091 // This code originates from JDK's sysOpen and open64_w
5092 // from src/solaris/hpi/src/system_md.c
5093
5094 #ifndef O_DELETE
5095 #define O_DELETE 0x10000
5096 #endif
5097
5098 // Open a file. Unlink the file immediately after open returns
5099 // if the specified oflag has the O_DELETE flag set.
5100 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5101
5102 int os::open(const char *path, int oflag, int mode) {
5103
5104 if (strlen(path) > MAX_PATH - 1) {
5105 errno = ENAMETOOLONG;
5106 return -1;
5107 }
5108 int fd;
5109 int o_delete = (oflag & O_DELETE);
5110 oflag = oflag & ~O_DELETE;
5111
5112 fd = ::open64(path, oflag, mode);
5113 if (fd == -1) return -1;
5114
5115 //If the open succeeded, the file might still be a directory
5116 {
5117 struct stat64 buf64;
5118 int ret = ::fstat64(fd, &buf64);
5119 int st_mode = buf64.st_mode;
5120
5121 if (ret != -1) {
5122 if ((st_mode & S_IFMT) == S_IFDIR) {
5123 errno = EISDIR;
5124 ::close(fd);
5125 return -1;
5126 }
5127 } else {
5128 ::close(fd);
5129 return -1;
5130 }
5131 }
5132
5133 /*
5134 * All file descriptors that are opened in the JVM and not
5135 * specifically destined for a subprocess should have the
5136 * close-on-exec flag set. If we don't set it, then careless 3rd
5137 * party native code might fork and exec without closing all
5138 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5139 * UNIXProcess.c), and this in turn might:
5140 *
5141 * - cause end-of-file to fail to be detected on some file
5142 * descriptors, resulting in mysterious hangs, or
5143 *
5144 * - might cause an fopen in the subprocess to fail on a system
5145 * suffering from bug 1085341.
5146 *
5147 * (Yes, the default setting of the close-on-exec flag is a Unix
5148 * design flaw)
5149 *
5150 * See:
5151 * 1085341: 32-bit stdio routines should support file descriptors >255
5152 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5153 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5154 */
5155 #ifdef FD_CLOEXEC
5156 {
5157 int flags = ::fcntl(fd, F_GETFD);
5158 if (flags != -1)
5159 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5160 }
5161 #endif
5162
5163 if (o_delete != 0) {
5164 ::unlink(path);
5165 }
5166 return fd;
5167 }
5168
5169
5170 // create binary file, rewriting existing file if required
5171 int os::create_binary_file(const char* path, bool rewrite_existing) {
5172 int oflags = O_WRONLY | O_CREAT;
5173 if (!rewrite_existing) {
5174 oflags |= O_EXCL;
5175 }
5176 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5177 }
5178
5179 // return current position of file pointer
5180 jlong os::current_file_offset(int fd) {
5181 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5182 }
5183
5184 // move file pointer to the specified offset
5185 jlong os::seek_to_file_offset(int fd, jlong offset) {
5186 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5187 }
5188
5189 // This code originates from JDK's sysAvailable
5190 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5191
5192 int os::available(int fd, jlong *bytes) {
5193 jlong cur, end;
5194 int mode;
5195 struct stat64 buf64;
5196
5197 if (::fstat64(fd, &buf64) >= 0) {
5198 mode = buf64.st_mode;
5199 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5200 /*
5201 * XXX: is the following call interruptible? If so, this might
5202 * need to go through the INTERRUPT_IO() wrapper as for other
5203 * blocking, interruptible calls in this file.
5204 */
5205 int n;
5206 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5207 *bytes = n;
5208 return 1;
5209 }
5210 }
5211 }
5212 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5213 return 0;
5214 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5215 return 0;
5216 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5217 return 0;
5218 }
5219 *bytes = end - cur;
5220 return 1;
5221 }
5222
5223 int os::socket_available(int fd, jint *pbytes) {
5224 // Linux doc says EINTR not returned, unlike Solaris
5225 int ret = ::ioctl(fd, FIONREAD, pbytes);
5226
5227 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5228 // is expected to return 0 on failure and 1 on success to the jdk.
5229 return (ret < 0) ? 0 : 1;
5230 }
5231
5232 // Map a block of memory.
5233 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5234 char *addr, size_t bytes, bool read_only,
5235 bool allow_exec) {
5236 int prot;
5237 int flags = MAP_PRIVATE;
5238
5239 if (read_only) {
5240 prot = PROT_READ;
5241 } else {
5242 prot = PROT_READ | PROT_WRITE;
5243 }
5244
5245 if (allow_exec) {
5246 prot |= PROT_EXEC;
5247 }
5248
5249 if (addr != NULL) {
5250 flags |= MAP_FIXED;
5251 }
5252
5253 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5254 fd, file_offset);
5255 if (mapped_address == MAP_FAILED) {
5256 return NULL;
5257 }
5258 return mapped_address;
5259 }
5260
5261
5262 // Remap a block of memory.
5263 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5264 char *addr, size_t bytes, bool read_only,
5265 bool allow_exec) {
5266 // same as map_memory() on this OS
5267 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5268 allow_exec);
5269 }
5270
5271
5272 // Unmap a block of memory.
5273 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5274 return munmap(addr, bytes) == 0;
5275 }
5276
5277 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5278
5279 static clockid_t thread_cpu_clockid(Thread* thread) {
5280 pthread_t tid = thread->osthread()->pthread_id();
5281 clockid_t clockid;
5282
5283 // Get thread clockid
5284 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5285 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5286 return clockid;
5287 }
5288
5289 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5290 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5291 // of a thread.
5292 //
5293 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5294 // the fast estimate available on the platform.
5295
5296 jlong os::current_thread_cpu_time() {
5297 if (os::Linux::supports_fast_thread_cpu_time()) {
5298 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5299 } else {
5300 // return user + sys since the cost is the same
5301 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5302 }
5303 }
5304
5305 jlong os::thread_cpu_time(Thread* thread) {
5306 // consistent with what current_thread_cpu_time() returns
5307 if (os::Linux::supports_fast_thread_cpu_time()) {
5308 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5309 } else {
5310 return slow_thread_cpu_time(thread, true /* user + sys */);
5311 }
5312 }
5313
5314 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5315 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5316 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5317 } else {
5318 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5319 }
5320 }
5321
5322 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5323 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5324 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5325 } else {
5326 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5327 }
5328 }
5329
5330 //
5331 // -1 on error.
5332 //
5333
5334 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5335 static bool proc_pid_cpu_avail = true;
5336 static bool proc_task_unchecked = true;
5337 static const char *proc_stat_path = "/proc/%d/stat";
5338 pid_t tid = thread->osthread()->thread_id();
5339 int i;
5340 char *s;
5341 char stat[2048];
5342 int statlen;
5343 char proc_name[64];
5344 int count;
5345 long sys_time, user_time;
5346 char string[64];
5347 char cdummy;
5348 int idummy;
5349 long ldummy;
5350 FILE *fp;
5351
5352 // We first try accessing /proc/<pid>/cpu since this is faster to
5353 // process. If this file is not present (linux kernels 2.5 and above)
5354 // then we open /proc/<pid>/stat.
5355 if ( proc_pid_cpu_avail ) {
5356 sprintf(proc_name, "/proc/%d/cpu", tid);
5357 fp = fopen(proc_name, "r");
5358 if ( fp != NULL ) {
5359 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
5360 fclose(fp);
5361 if ( count != 3 ) return -1;
5362
5363 if (user_sys_cpu_time) {
5364 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5365 } else {
5366 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5367 }
5368 }
5369 else proc_pid_cpu_avail = false;
5370 }
5371
5372 // The /proc/<tid>/stat aggregates per-process usage on
5373 // new Linux kernels 2.6+ where NPTL is supported.
5374 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5375 // See bug 6328462.
5376 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
5377 // and possibly in some other cases, so we check its availability.
5378 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5379 // This is executed only once
5380 proc_task_unchecked = false;
5381 fp = fopen("/proc/self/task", "r");
5382 if (fp != NULL) {
5383 proc_stat_path = "/proc/self/task/%d/stat";
5384 fclose(fp);
5385 }
5386 }
5387
5388 sprintf(proc_name, proc_stat_path, tid);
5389 fp = fopen(proc_name, "r");
5390 if ( fp == NULL ) return -1;
5391 statlen = fread(stat, 1, 2047, fp);
5392 stat[statlen] = '\0';
5393 fclose(fp);
5394
5395 // Skip pid and the command string. Note that we could be dealing with
5396 // weird command names, e.g. user could decide to rename java launcher
5397 // to "java 1.4.2 :)", then the stat file would look like
5398 // 1234 (java 1.4.2 :)) R ... ...
5399 // We don't really need to know the command string, just find the last
5400 // occurrence of ")" and then start parsing from there. See bug 4726580.
5401 s = strrchr(stat, ')');
5402 i = 0;
5403 if (s == NULL ) return -1;
5404
5405 // Skip blank chars
5406 do s++; while (isspace(*s));
5407
5408 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5409 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5410 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5411 &user_time, &sys_time);
5412 if ( count != 13 ) return -1;
5413 if (user_sys_cpu_time) {
5414 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5415 } else {
5416 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5417 }
5418 }
5419
5420 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5421 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5422 info_ptr->may_skip_backward = false; // elapsed time not wall time
5423 info_ptr->may_skip_forward = false; // elapsed time not wall time
5424 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5425 }
5426
5427 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5428 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5429 info_ptr->may_skip_backward = false; // elapsed time not wall time
5430 info_ptr->may_skip_forward = false; // elapsed time not wall time
5431 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5432 }
5433
5434 bool os::is_thread_cpu_time_supported() {
5435 return true;
5436 }
5437
5438 // System loadavg support. Returns -1 if load average cannot be obtained.
5439 // Linux doesn't yet have a (official) notion of processor sets,
5440 // so just return the system wide load average.
5441 int os::loadavg(double loadavg[], int nelem) {
5442 return ::getloadavg(loadavg, nelem);
5443 }
5444
5445 void os::pause() {
5446 char filename[MAX_PATH];
5447 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5448 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5449 } else {
5450 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5451 }
5452
5453 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5454 if (fd != -1) {
5455 struct stat buf;
5456 ::close(fd);
5457 while (::stat(filename, &buf) == 0) {
5458 (void)::poll(NULL, 0, 100);
5459 }
5460 } else {
5461 jio_fprintf(stderr,
5462 "Could not open pause file '%s', continuing immediately.\n", filename);
5463 }
5464 }
5465
5466
5467 // Refer to the comments in os_solaris.cpp park-unpark.
5468 //
5469 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5470 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5471 // For specifics regarding the bug see GLIBC BUGID 261237 :
5472 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5473 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5474 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5475 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5476 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5477 // and monitorenter when we're using 1-0 locking. All those operations may result in
5478 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5479 // of libpthread avoids the problem, but isn't practical.
5480 //
5481 // Possible remedies:
5482 //
5483 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5484 // This is palliative and probabilistic, however. If the thread is preempted
5485 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5486 // than the minimum period may have passed, and the abstime may be stale (in the
5487 // past) resultin in a hang. Using this technique reduces the odds of a hang
5488 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5489 //
5490 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5491 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5492 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5493 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5494 // thread.
5495 //
5496 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5497 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5498 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5499 // This also works well. In fact it avoids kernel-level scalability impediments
5500 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5501 // timers in a graceful fashion.
5502 //
5503 // 4. When the abstime value is in the past it appears that control returns
5504 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5505 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5506 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5507 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5508 // It may be possible to avoid reinitialization by checking the return
5509 // value from pthread_cond_timedwait(). In addition to reinitializing the
5510 // condvar we must establish the invariant that cond_signal() is only called
5511 // within critical sections protected by the adjunct mutex. This prevents
5512 // cond_signal() from "seeing" a condvar that's in the midst of being
5513 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5514 // desirable signal-after-unlock optimization that avoids futile context switching.
5515 //
5516 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5517 // structure when a condvar is used or initialized. cond_destroy() would
5518 // release the helper structure. Our reinitialize-after-timedwait fix
5519 // put excessive stress on malloc/free and locks protecting the c-heap.
5520 //
5521 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5522 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5523 // and only enabling the work-around for vulnerable environments.
5524
5525 // utility to compute the abstime argument to timedwait:
5526 // millis is the relative timeout time
5527 // abstime will be the absolute timeout time
5528 // TODO: replace compute_abstime() with unpackTime()
5529
5530 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5531 if (millis < 0) millis = 0;
5532 struct timeval now;
5533 int status = gettimeofday(&now, NULL);
5534 assert(status == 0, "gettimeofday");
5535 jlong seconds = millis / 1000;
5536 millis %= 1000;
5537 if (seconds > 50000000) { // see man cond_timedwait(3T)
5538 seconds = 50000000;
5539 }
5540 abstime->tv_sec = now.tv_sec + seconds;
5541 long usec = now.tv_usec + millis * 1000;
5542 if (usec >= 1000000) {
5543 abstime->tv_sec += 1;
5544 usec -= 1000000;
5545 }
5546 abstime->tv_nsec = usec * 1000;
5547 return abstime;
5548 }
5549
5550
5551 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5552 // Conceptually TryPark() should be equivalent to park(0).
5553
5554 int os::PlatformEvent::TryPark() {
5555 for (;;) {
5556 const int v = _Event ;
5557 guarantee ((v == 0) || (v == 1), "invariant") ;
5558 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5559 }
5560 }
5561
5562 void os::PlatformEvent::park() { // AKA "down()"
5563 // Invariant: Only the thread associated with the Event/PlatformEvent
5564 // may call park().
5565 // TODO: assert that _Assoc != NULL or _Assoc == Self
5566 int v ;
5567 for (;;) {
5568 v = _Event ;
5569 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5570 }
5571 guarantee (v >= 0, "invariant") ;
5572 if (v == 0) {
5573 // Do this the hard way by blocking ...
5574 int status = pthread_mutex_lock(_mutex);
5575 assert_status(status == 0, status, "mutex_lock");
5576 guarantee (_nParked == 0, "invariant") ;
5577 ++ _nParked ;
5578 while (_Event < 0) {
5579 status = pthread_cond_wait(_cond, _mutex);
5580 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5581 // Treat this the same as if the wait was interrupted
5582 if (status == ETIME) { status = EINTR; }
5583 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5584 }
5585 -- _nParked ;
5586
5587 _Event = 0 ;
5588 status = pthread_mutex_unlock(_mutex);
5589 assert_status(status == 0, status, "mutex_unlock");
5590 // Paranoia to ensure our locked and lock-free paths interact
5591 // correctly with each other.
5592 OrderAccess::fence();
5593 }
5594 guarantee (_Event >= 0, "invariant") ;
5595 }
5596
5597 int os::PlatformEvent::park(jlong millis) {
5598 guarantee (_nParked == 0, "invariant") ;
5599
5600 int v ;
5601 for (;;) {
5602 v = _Event ;
5603 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5604 }
5605 guarantee (v >= 0, "invariant") ;
5606 if (v != 0) return OS_OK ;
5607
5608 // We do this the hard way, by blocking the thread.
5609 // Consider enforcing a minimum timeout value.
5610 struct timespec abst;
5611 compute_abstime(&abst, millis);
5612
5613 int ret = OS_TIMEOUT;
5614 int status = pthread_mutex_lock(_mutex);
5615 assert_status(status == 0, status, "mutex_lock");
5616 guarantee (_nParked == 0, "invariant") ;
5617 ++_nParked ;
5618
5619 // Object.wait(timo) will return because of
5620 // (a) notification
5621 // (b) timeout
5622 // (c) thread.interrupt
5623 //
5624 // Thread.interrupt and object.notify{All} both call Event::set.
5625 // That is, we treat thread.interrupt as a special case of notification.
5626 // The underlying Solaris implementation, cond_timedwait, admits
5627 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5628 // JVM from making those visible to Java code. As such, we must
5629 // filter out spurious wakeups. We assume all ETIME returns are valid.
5630 //
5631 // TODO: properly differentiate simultaneous notify+interrupt.
5632 // In that case, we should propagate the notify to another waiter.
5633
5634 while (_Event < 0) {
5635 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5636 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5637 pthread_cond_destroy (_cond);
5638 pthread_cond_init (_cond, NULL) ;
5639 }
5640 assert_status(status == 0 || status == EINTR ||
5641 status == ETIME || status == ETIMEDOUT,
5642 status, "cond_timedwait");
5643 if (!FilterSpuriousWakeups) break ; // previous semantics
5644 if (status == ETIME || status == ETIMEDOUT) break ;
5645 // We consume and ignore EINTR and spurious wakeups.
5646 }
5647 --_nParked ;
5648 if (_Event >= 0) {
5649 ret = OS_OK;
5650 }
5651 _Event = 0 ;
5652 status = pthread_mutex_unlock(_mutex);
5653 assert_status(status == 0, status, "mutex_unlock");
5654 assert (_nParked == 0, "invariant") ;
5655 // Paranoia to ensure our locked and lock-free paths interact
5656 // correctly with each other.
5657 OrderAccess::fence();
5658 return ret;
5659 }
5660
5661 void os::PlatformEvent::unpark() {
5662 // Transitions for _Event:
5663 // 0 :=> 1
5664 // 1 :=> 1
5665 // -1 :=> either 0 or 1; must signal target thread
5666 // That is, we can safely transition _Event from -1 to either
5667 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5668 // unpark() calls.
5669 // See also: "Semaphores in Plan 9" by Mullender & Cox
5670 //
5671 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5672 // that it will take two back-to-back park() calls for the owning
5673 // thread to block. This has the benefit of forcing a spurious return
5674 // from the first park() call after an unpark() call which will help
5675 // shake out uses of park() and unpark() without condition variables.
5676
5677 if (Atomic::xchg(1, &_Event) >= 0) return;
5678
5679 // Wait for the thread associated with the event to vacate
5680 int status = pthread_mutex_lock(_mutex);
5681 assert_status(status == 0, status, "mutex_lock");
5682 int AnyWaiters = _nParked;
5683 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5684 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5685 AnyWaiters = 0;
5686 pthread_cond_signal(_cond);
5687 }
5688 status = pthread_mutex_unlock(_mutex);
5689 assert_status(status == 0, status, "mutex_unlock");
5690 if (AnyWaiters != 0) {
5691 status = pthread_cond_signal(_cond);
5692 assert_status(status == 0, status, "cond_signal");
5693 }
5694
5695 // Note that we signal() _after dropping the lock for "immortal" Events.
5696 // This is safe and avoids a common class of futile wakeups. In rare
5697 // circumstances this can cause a thread to return prematurely from
5698 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5699 // simply re-test the condition and re-park itself.
5700 }
5701
5702
5703 // JSR166
5704 // -------------------------------------------------------
5705
5706 /*
5707 * The solaris and linux implementations of park/unpark are fairly
5708 * conservative for now, but can be improved. They currently use a
5709 * mutex/condvar pair, plus a a count.
5710 * Park decrements count if > 0, else does a condvar wait. Unpark
5711 * sets count to 1 and signals condvar. Only one thread ever waits
5712 * on the condvar. Contention seen when trying to park implies that someone
5713 * is unparking you, so don't wait. And spurious returns are fine, so there
5714 * is no need to track notifications.
5715 */
5716
5717 #define MAX_SECS 100000000
5718 /*
5719 * This code is common to linux and solaris and will be moved to a
5720 * common place in dolphin.
5721 *
5722 * The passed in time value is either a relative time in nanoseconds
5723 * or an absolute time in milliseconds. Either way it has to be unpacked
5724 * into suitable seconds and nanoseconds components and stored in the
5725 * given timespec structure.
5726 * Given time is a 64-bit value and the time_t used in the timespec is only
5727 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5728 * overflow if times way in the future are given. Further on Solaris versions
5729 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5730 * number of seconds, in abstime, is less than current_time + 100,000,000.
5731 * As it will be 28 years before "now + 100000000" will overflow we can
5732 * ignore overflow and just impose a hard-limit on seconds using the value
5733 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5734 * years from "now".
5735 */
5736
5737 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5738 assert (time > 0, "convertTime");
5739
5740 struct timeval now;
5741 int status = gettimeofday(&now, NULL);
5742 assert(status == 0, "gettimeofday");
5743
5744 time_t max_secs = now.tv_sec + MAX_SECS;
5745
5746 if (isAbsolute) {
5747 jlong secs = time / 1000;
5748 if (secs > max_secs) {
5749 absTime->tv_sec = max_secs;
5750 }
5751 else {
5752 absTime->tv_sec = secs;
5753 }
5754 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5755 }
5756 else {
5757 jlong secs = time / NANOSECS_PER_SEC;
5758 if (secs >= MAX_SECS) {
5759 absTime->tv_sec = max_secs;
5760 absTime->tv_nsec = 0;
5761 }
5762 else {
5763 absTime->tv_sec = now.tv_sec + secs;
5764 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5765 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5766 absTime->tv_nsec -= NANOSECS_PER_SEC;
5767 ++absTime->tv_sec; // note: this must be <= max_secs
5768 }
5769 }
5770 }
5771 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5772 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5773 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5774 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5775 }
5776
5777 void Parker::park(bool isAbsolute, jlong time) {
5778 // Ideally we'd do something useful while spinning, such
5779 // as calling unpackTime().
5780
5781 // Optional fast-path check:
5782 // Return immediately if a permit is available.
5783 // We depend on Atomic::xchg() having full barrier semantics
5784 // since we are doing a lock-free update to _counter.
5785 if (Atomic::xchg(0, &_counter) > 0) return;
5786
5787 Thread* thread = Thread::current();
5788 assert(thread->is_Java_thread(), "Must be JavaThread");
5789 JavaThread *jt = (JavaThread *)thread;
5790
5791 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5792 // Check interrupt before trying to wait
5793 if (Thread::is_interrupted(thread, false)) {
5794 return;
5795 }
5796
5797 // Next, demultiplex/decode time arguments
5798 timespec absTime;
5799 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5800 return;
5801 }
5802 if (time > 0) {
5803 unpackTime(&absTime, isAbsolute, time);
5804 }
5805
5806
5807 // Enter safepoint region
5808 // Beware of deadlocks such as 6317397.
5809 // The per-thread Parker:: mutex is a classic leaf-lock.
5810 // In particular a thread must never block on the Threads_lock while
5811 // holding the Parker:: mutex. If safepoints are pending both the
5812 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5813 ThreadBlockInVM tbivm(jt);
5814
5815 // Don't wait if cannot get lock since interference arises from
5816 // unblocking. Also. check interrupt before trying wait
5817 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5818 return;
5819 }
5820
5821 int status ;
5822 if (_counter > 0) { // no wait needed
5823 _counter = 0;
5824 status = pthread_mutex_unlock(_mutex);
5825 assert (status == 0, "invariant") ;
5826 // Paranoia to ensure our locked and lock-free paths interact
5827 // correctly with each other and Java-level accesses.
5828 OrderAccess::fence();
5829 return;
5830 }
5831
5832 #ifdef ASSERT
5833 // Don't catch signals while blocked; let the running threads have the signals.
5834 // (This allows a debugger to break into the running thread.)
5835 sigset_t oldsigs;
5836 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5837 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5838 #endif
5839
5840 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5841 jt->set_suspend_equivalent();
5842 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5843
5844 if (time == 0) {
5845 status = pthread_cond_wait (_cond, _mutex) ;
5846 } else {
5847 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5848 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5849 pthread_cond_destroy (_cond) ;
5850 pthread_cond_init (_cond, NULL);
5851 }
5852 }
5853 assert_status(status == 0 || status == EINTR ||
5854 status == ETIME || status == ETIMEDOUT,
5855 status, "cond_timedwait");
5856
5857 #ifdef ASSERT
5858 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5859 #endif
5860
5861 _counter = 0 ;
5862 status = pthread_mutex_unlock(_mutex) ;
5863 assert_status(status == 0, status, "invariant") ;
5864 // Paranoia to ensure our locked and lock-free paths interact
5865 // correctly with each other and Java-level accesses.
5866 OrderAccess::fence();
5867
5868 // If externally suspended while waiting, re-suspend
5869 if (jt->handle_special_suspend_equivalent_condition()) {
5870 jt->java_suspend_self();
5871 }
5872 }
5873
5874 void Parker::unpark() {
5875 int s, status ;
5876 status = pthread_mutex_lock(_mutex);
5877 assert (status == 0, "invariant") ;
5878 s = _counter;
5879 _counter = 1;
5880 if (s < 1) {
5881 if (WorkAroundNPTLTimedWaitHang) {
5882 status = pthread_cond_signal (_cond) ;
5883 assert (status == 0, "invariant") ;
5884 status = pthread_mutex_unlock(_mutex);
5885 assert (status == 0, "invariant") ;
5886 } else {
5887 status = pthread_mutex_unlock(_mutex);
5888 assert (status == 0, "invariant") ;
5889 status = pthread_cond_signal (_cond) ;
5890 assert (status == 0, "invariant") ;
5891 }
5892 } else {
5893 pthread_mutex_unlock(_mutex);
5894 assert (status == 0, "invariant") ;
5895 }
5896 }
5897
5898
5899 extern char** environ;
5900
5901 #ifndef __NR_fork
5902 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5903 #endif
5904
5905 #ifndef __NR_execve
5906 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5907 #endif
5908
5909 // Run the specified command in a separate process. Return its exit value,
5910 // or -1 on failure (e.g. can't fork a new process).
5911 // Unlike system(), this function can be called from signal handler. It
5912 // doesn't block SIGINT et al.
5913 int os::fork_and_exec(char* cmd) {
5914 const char * argv[4] = {"sh", "-c", cmd, NULL};
5915
5916 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5917 // pthread_atfork handlers and reset pthread library. All we need is a
5918 // separate process to execve. Make a direct syscall to fork process.
5919 // On IA64 there's no fork syscall, we have to use fork() and hope for
5920 // the best...
5921 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5922 IA64_ONLY(fork();)
5923
5924 if (pid < 0) {
5925 // fork failed
5926 return -1;
5927
5928 } else if (pid == 0) {
5929 // child process
5930
5931 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5932 // first to kill every thread on the thread list. Because this list is
5933 // not reset by fork() (see notes above), execve() will instead kill
5934 // every thread in the parent process. We know this is the only thread
5935 // in the new process, so make a system call directly.
5936 // IA64 should use normal execve() from glibc to match the glibc fork()
5937 // above.
5938 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5939 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5940
5941 // execve failed
5942 _exit(-1);
5943
5944 } else {
5945 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5946 // care about the actual exit code, for now.
5947
5948 int status;
5949
5950 // Wait for the child process to exit. This returns immediately if
5951 // the child has already exited. */
5952 while (waitpid(pid, &status, 0) < 0) {
5953 switch (errno) {
5954 case ECHILD: return 0;
5955 case EINTR: break;
5956 default: return -1;
5957 }
5958 }
5959
5960 if (WIFEXITED(status)) {
5961 // The child exited normally; get its exit code.
5962 return WEXITSTATUS(status);
5963 } else if (WIFSIGNALED(status)) {
5964 // The child exited because of a signal
5965 // The best value to return is 0x80 + signal number,
5966 // because that is what all Unix shells do, and because
5967 // it allows callers to distinguish between process exit and
5968 // process death by signal.
5969 return 0x80 + WTERMSIG(status);
5970 } else {
5971 // Unknown exit code; pass it through
5972 return status;
5973 }
5974 }
5975 }
5976
5977 // is_headless_jre()
5978 //
5979 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5980 // in order to report if we are running in a headless jre
5981 //
5982 // Since JDK8 xawt/libmawt.so was moved into the same directory
5983 // as libawt.so, and renamed libawt_xawt.so
5984 //
5985 bool os::is_headless_jre() {
5986 struct stat statbuf;
5987 char buf[MAXPATHLEN];
5988 char libmawtpath[MAXPATHLEN];
5989 const char *xawtstr = "/xawt/libmawt.so";
5990 const char *new_xawtstr = "/libawt_xawt.so";
5991 char *p;
5992
5993 // Get path to libjvm.so
5994 os::jvm_path(buf, sizeof(buf));
5995
5996 // Get rid of libjvm.so
5997 p = strrchr(buf, '/');
5998 if (p == NULL) return false;
5999 else *p = '\0';
6000
6001 // Get rid of client or server
6002 p = strrchr(buf, '/');
6003 if (p == NULL) return false;
6004 else *p = '\0';
6005
6006 // check xawt/libmawt.so
6007 strcpy(libmawtpath, buf);
6008 strcat(libmawtpath, xawtstr);
6009 if (::stat(libmawtpath, &statbuf) == 0) return false;
6010
6011 // check libawt_xawt.so
6012 strcpy(libmawtpath, buf);
6013 strcat(libmawtpath, new_xawtstr);
6014 if (::stat(libmawtpath, &statbuf) == 0) return false;
6015
6016 return true;
6017 }
6018
6019 // Get the default path to the core file
6020 // Returns the length of the string
6021 int os::get_core_path(char* buffer, size_t bufferSize) {
6022 const char* p = get_current_directory(buffer, bufferSize);
6023
6024 if (p == NULL) {
6025 assert(p != NULL, "failed to get current directory");
6026 return 0;
6027 }
6028
6029 return strlen(buffer);
6030 }
6031
6032 #ifdef JAVASE_EMBEDDED
6033 //
6034 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6035 //
6036 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6037
6038 // ctor
6039 //
6040 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6041 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6042 _fd = fd;
6043
6044 if (os::create_thread(this, os::os_thread)) {
6045 _memnotify_thread = this;
6046 os::set_priority(this, NearMaxPriority);
6047 os::start_thread(this);
6048 }
6049 }
6050
6051 // Where all the work gets done
6052 //
6053 void MemNotifyThread::run() {
6054 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6055
6056 // Set up the select arguments
6057 fd_set rfds;
6058 if (_fd != -1) {
6059 FD_ZERO(&rfds);
6060 FD_SET(_fd, &rfds);
6061 }
6062
6063 // Now wait for the mem_notify device to wake up
6064 while (1) {
6065 // Wait for the mem_notify device to signal us..
6066 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6067 if (rc == -1) {
6068 perror("select!\n");
6069 break;
6070 } else if (rc) {
6071 //ssize_t free_before = os::available_memory();
6072 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6073
6074 // The kernel is telling us there is not much memory left...
6075 // try to do something about that
6076
6077 // If we are not already in a GC, try one.
6078 if (!Universe::heap()->is_gc_active()) {
6079 Universe::heap()->collect(GCCause::_allocation_failure);
6080
6081 //ssize_t free_after = os::available_memory();
6082 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6083 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6084 }
6085 // We might want to do something like the following if we find the GC's are not helping...
6086 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6087 }
6088 }
6089 }
6090
6091 //
6092 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6093 //
6094 void MemNotifyThread::start() {
6095 int fd;
6096 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6097 if (fd < 0) {
6098 return;
6099 }
6100
6101 if (memnotify_thread() == NULL) {
6102 new MemNotifyThread(fd);
6103 }
6104 }
6105
6106 #endif // JAVASE_EMBEDDED
6107
6108
6109 /////////////// Unit tests ///////////////
6110
6111 #ifndef PRODUCT
6112
6113 #define test_log(...) \
6114 do {\
6115 if (VerboseInternalVMTests) { \
6116 tty->print_cr(__VA_ARGS__); \
6117 tty->flush(); \
6118 }\
6119 } while (false)
6120
6121 class TestReserveMemorySpecial : AllStatic {
6122 public:
6123 static void small_page_write(void* addr, size_t size) {
6124 size_t page_size = os::vm_page_size();
6125
6126 char* end = (char*)addr + size;
6127 for (char* p = (char*)addr; p < end; p += page_size) {
6128 *p = 1;
6129 }
6130 }
6131
6132 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6133 if (!UseHugeTLBFS) {
6134 return;
6135 }
6136
6137 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6138
6139 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6140
6141 if (addr != NULL) {
6142 small_page_write(addr, size);
6143
6144 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6145 }
6146 }
6147
6148 static void test_reserve_memory_special_huge_tlbfs_only() {
6149 if (!UseHugeTLBFS) {
6150 return;
6151 }
6152
6153 size_t lp = os::large_page_size();
6154
6155 for (size_t size = lp; size <= lp * 10; size += lp) {
6156 test_reserve_memory_special_huge_tlbfs_only(size);
6157 }
6158 }
6159
6160 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6161 if (!UseHugeTLBFS) {
6162 return;
6163 }
6164
6165 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6166 size, alignment);
6167
6168 assert(size >= os::large_page_size(), "Incorrect input to test");
6169
6170 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6171
6172 if (addr != NULL) {
6173 small_page_write(addr, size);
6174
6175 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6176 }
6177 }
6178
6179 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6180 size_t lp = os::large_page_size();
6181 size_t ag = os::vm_allocation_granularity();
6182
6183 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6184 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6185 }
6186 }
6187
6188 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6189 size_t lp = os::large_page_size();
6190 size_t ag = os::vm_allocation_granularity();
6191
6192 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6193 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6194 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6195 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6196 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6197 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6198 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6199 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6200 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6201 }
6202
6203 static void test_reserve_memory_special_huge_tlbfs() {
6204 if (!UseHugeTLBFS) {
6205 return;
6206 }
6207
6208 test_reserve_memory_special_huge_tlbfs_only();
6209 test_reserve_memory_special_huge_tlbfs_mixed();
6210 }
6211
6212 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6213 if (!UseSHM) {
6214 return;
6215 }
6216
6217 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6218
6219 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6220
6221 if (addr != NULL) {
6222 assert(is_ptr_aligned(addr, alignment), "Check");
6223 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6224
6225 small_page_write(addr, size);
6226
6227 os::Linux::release_memory_special_shm(addr, size);
6228 }
6229 }
6230
6231 static void test_reserve_memory_special_shm() {
6232 size_t lp = os::large_page_size();
6233 size_t ag = os::vm_allocation_granularity();
6234
6235 for (size_t size = ag; size < lp * 3; size += ag) {
6236 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6237 test_reserve_memory_special_shm(size, alignment);
6238 }
6239 }
6240 }
6241
6242 static void test() {
6243 test_reserve_memory_special_huge_tlbfs();
6244 test_reserve_memory_special_shm();
6245 }
6246 };
6247
6248 void TestReserveMemorySpecial_test() {
6249 TestReserveMemorySpecial::test();
6250 }
6251
6252 #endif