Fri, 29 Apr 2016 00:06:10 +0800
Added MIPS 64-bit port.
1 /*
2 * Copyright (c) 1999, 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 /*
26 * This file has been modified by Loongson Technology in 2015. These
27 * modifications are Copyright (c) 2015 Loongson Technology, and are made
28 * available on the same license terms set forth above.
29 */
31 // no precompiled headers
32 #include "classfile/classLoader.hpp"
33 #include "classfile/systemDictionary.hpp"
34 #include "classfile/vmSymbols.hpp"
35 #include "code/icBuffer.hpp"
36 #include "code/vtableStubs.hpp"
37 #include "compiler/compileBroker.hpp"
38 #include "compiler/disassembler.hpp"
39 #include "interpreter/interpreter.hpp"
40 #include "jvm_linux.h"
41 #include "memory/allocation.inline.hpp"
42 #include "memory/filemap.hpp"
43 #include "mutex_linux.inline.hpp"
44 #include "oops/oop.inline.hpp"
45 #include "os_share_linux.hpp"
46 #include "prims/jniFastGetField.hpp"
47 #include "prims/jvm.h"
48 #include "prims/jvm_misc.hpp"
49 #include "runtime/arguments.hpp"
50 #include "runtime/extendedPC.hpp"
51 #include "runtime/globals.hpp"
52 #include "runtime/interfaceSupport.hpp"
53 #include "runtime/init.hpp"
54 #include "runtime/java.hpp"
55 #include "runtime/javaCalls.hpp"
56 #include "runtime/mutexLocker.hpp"
57 #include "runtime/objectMonitor.hpp"
58 #include "runtime/osThread.hpp"
59 #include "runtime/perfMemory.hpp"
60 #include "runtime/sharedRuntime.hpp"
61 #include "runtime/statSampler.hpp"
62 #include "runtime/stubRoutines.hpp"
63 #include "runtime/thread.inline.hpp"
64 #include "runtime/threadCritical.hpp"
65 #include "runtime/timer.hpp"
66 #include "services/attachListener.hpp"
67 #include "services/memTracker.hpp"
68 #include "services/runtimeService.hpp"
69 #include "utilities/decoder.hpp"
70 #include "utilities/defaultStream.hpp"
71 #include "utilities/events.hpp"
72 #include "utilities/elfFile.hpp"
73 #include "utilities/growableArray.hpp"
74 #include "utilities/vmError.hpp"
76 // put OS-includes here
77 # include <sys/types.h>
78 # include <sys/mman.h>
79 # include <sys/stat.h>
80 # include <sys/select.h>
81 # include <pthread.h>
82 # include <signal.h>
83 # include <errno.h>
84 # include <dlfcn.h>
85 # include <stdio.h>
86 # include <unistd.h>
87 # include <sys/resource.h>
88 # include <pthread.h>
89 # include <sys/stat.h>
90 # include <sys/time.h>
91 # include <sys/times.h>
92 # include <sys/utsname.h>
93 # include <sys/socket.h>
94 # include <sys/wait.h>
95 # include <pwd.h>
96 # include <poll.h>
97 # include <semaphore.h>
98 # include <fcntl.h>
99 # include <string.h>
100 # include <syscall.h>
101 # include <sys/sysinfo.h>
102 # include <gnu/libc-version.h>
103 # include <sys/ipc.h>
104 # include <sys/shm.h>
105 # include <link.h>
106 # include <stdint.h>
107 # include <inttypes.h>
108 # include <sys/ioctl.h>
110 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
112 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
113 // getrusage() is prepared to handle the associated failure.
114 #ifndef RUSAGE_THREAD
115 #define RUSAGE_THREAD (1) /* only the calling thread */
116 #endif
118 #define MAX_PATH (2 * K)
120 #define MAX_SECS 100000000
122 // for timer info max values which include all bits
123 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
125 #define LARGEPAGES_BIT (1 << 6)
126 ////////////////////////////////////////////////////////////////////////////////
127 // global variables
128 julong os::Linux::_physical_memory = 0;
130 address os::Linux::_initial_thread_stack_bottom = NULL;
131 uintptr_t os::Linux::_initial_thread_stack_size = 0;
133 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
134 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
135 Mutex* os::Linux::_createThread_lock = NULL;
136 pthread_t os::Linux::_main_thread;
137 int os::Linux::_page_size = -1;
138 const int os::Linux::_vm_default_page_size = (8 * K);
139 bool os::Linux::_is_floating_stack = false;
140 bool os::Linux::_is_NPTL = false;
141 bool os::Linux::_supports_fast_thread_cpu_time = false;
142 const char * os::Linux::_glibc_version = NULL;
143 const char * os::Linux::_libpthread_version = NULL;
144 pthread_condattr_t os::Linux::_condattr[1];
146 static jlong initial_time_count=0;
148 static int clock_tics_per_sec = 100;
150 // For diagnostics to print a message once. see run_periodic_checks
151 static sigset_t check_signal_done;
152 static bool check_signals = true;
154 static pid_t _initial_pid = 0;
156 /* Signal number used to suspend/resume a thread */
158 /* do not use any signal number less than SIGSEGV, see 4355769 */
159 static int SR_signum = SIGUSR2;
160 sigset_t SR_sigset;
162 /* Used to protect dlsym() calls */
163 static pthread_mutex_t dl_mutex;
165 // Declarations
166 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
168 #ifdef JAVASE_EMBEDDED
169 class MemNotifyThread: public Thread {
170 friend class VMStructs;
171 public:
172 virtual void run();
174 private:
175 static MemNotifyThread* _memnotify_thread;
176 int _fd;
178 public:
180 // Constructor
181 MemNotifyThread(int fd);
183 // Tester
184 bool is_memnotify_thread() const { return true; }
186 // Printing
187 char* name() const { return (char*)"Linux MemNotify Thread"; }
189 // Returns the single instance of the MemNotifyThread
190 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
192 // Create and start the single instance of MemNotifyThread
193 static void start();
194 };
195 #endif // JAVASE_EMBEDDED
197 // utility functions
199 static int SR_initialize();
201 julong os::available_memory() {
202 return Linux::available_memory();
203 }
205 julong os::Linux::available_memory() {
206 // values in struct sysinfo are "unsigned long"
207 struct sysinfo si;
208 sysinfo(&si);
210 return (julong)si.freeram * si.mem_unit;
211 }
213 julong os::physical_memory() {
214 return Linux::physical_memory();
215 }
217 ////////////////////////////////////////////////////////////////////////////////
218 // environment support
220 bool os::getenv(const char* name, char* buf, int len) {
221 const char* val = ::getenv(name);
222 if (val != NULL && strlen(val) < (size_t)len) {
223 strcpy(buf, val);
224 return true;
225 }
226 if (len > 0) buf[0] = 0; // return a null string
227 return false;
228 }
231 // Return true if user is running as root.
233 bool os::have_special_privileges() {
234 static bool init = false;
235 static bool privileges = false;
236 if (!init) {
237 privileges = (getuid() != geteuid()) || (getgid() != getegid());
238 init = true;
239 }
240 return privileges;
241 }
244 #ifndef SYS_gettid
245 // i386: 224, ia64: 1105, amd64: 186, sparc 143
246 #ifdef __ia64__
247 #define SYS_gettid 1105
248 #elif __i386__
249 #define SYS_gettid 224
250 #elif __amd64__
251 #define SYS_gettid 186
252 #elif __sparc__
253 #define SYS_gettid 143
254 #elif __mips__
255 #define SYS_gettid 4222
256 #else
257 #error define gettid for the arch
258 #endif
259 #endif
261 // Cpu architecture string
262 #if defined(ZERO)
263 static char cpu_arch[] = ZERO_LIBARCH;
264 #elif defined(IA64)
265 static char cpu_arch[] = "ia64";
266 #elif defined(IA32)
267 static char cpu_arch[] = "i386";
268 #elif defined(AMD64)
269 static char cpu_arch[] = "amd64";
270 #elif defined(ARM)
271 static char cpu_arch[] = "arm";
272 #elif defined(PPC32)
273 static char cpu_arch[] = "ppc";
274 #elif defined(PPC64)
275 static char cpu_arch[] = "ppc64";
276 #elif defined(MIPS64)
277 static char cpu_arch[] = "mipsel";
278 #elif defined(SPARC)
279 # ifdef _LP64
280 static char cpu_arch[] = "sparcv9";
281 # else
282 static char cpu_arch[] = "sparc";
283 # endif
284 #else
285 #error Add appropriate cpu_arch setting
286 #endif
289 // pid_t gettid()
290 //
291 // Returns the kernel thread id of the currently running thread. Kernel
292 // thread id is used to access /proc.
293 //
294 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
295 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
296 //
297 pid_t os::Linux::gettid() {
298 int rslt = syscall(SYS_gettid);
299 if (rslt == -1) {
300 // old kernel, no NPTL support
301 return getpid();
302 } else {
303 return (pid_t)rslt;
304 }
305 }
307 // Most versions of linux have a bug where the number of processors are
308 // determined by looking at the /proc file system. In a chroot environment,
309 // the system call returns 1. This causes the VM to act as if it is
310 // a single processor and elide locking (see is_MP() call).
311 static bool unsafe_chroot_detected = false;
312 static const char *unstable_chroot_error = "/proc file system not found.\n"
313 "Java may be unstable running multithreaded in a chroot "
314 "environment on Linux when /proc filesystem is not mounted.";
316 void os::Linux::initialize_system_info() {
317 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
318 if (processor_count() == 1) {
319 pid_t pid = os::Linux::gettid();
320 char fname[32];
321 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
322 FILE *fp = fopen(fname, "r");
323 if (fp == NULL) {
324 unsafe_chroot_detected = true;
325 } else {
326 fclose(fp);
327 }
328 }
329 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
330 assert(processor_count() > 0, "linux error");
331 }
333 void os::init_system_properties_values() {
334 // The next steps are taken in the product version:
335 //
336 // Obtain the JAVA_HOME value from the location of libjvm.so.
337 // This library should be located at:
338 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
339 //
340 // If "/jre/lib/" appears at the right place in the path, then we
341 // assume libjvm.so is installed in a JDK and we use this path.
342 //
343 // Otherwise exit with message: "Could not create the Java virtual machine."
344 //
345 // The following extra steps are taken in the debugging version:
346 //
347 // If "/jre/lib/" does NOT appear at the right place in the path
348 // instead of exit check for $JAVA_HOME environment variable.
349 //
350 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
351 // then we append a fake suffix "hotspot/libjvm.so" to this path so
352 // it looks like libjvm.so is installed there
353 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
354 //
355 // Otherwise exit.
356 //
357 // Important note: if the location of libjvm.so changes this
358 // code needs to be changed accordingly.
360 // The next few definitions allow the code to be verbatim:
361 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
362 #define getenv(n) ::getenv(n)
364 /*
365 * See ld(1):
366 * The linker uses the following search paths to locate required
367 * shared libraries:
368 * 1: ...
369 * ...
370 * 7: The default directories, normally /lib and /usr/lib.
371 */
372 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390) || defined(MIPS64))
373 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
374 #else
375 #define DEFAULT_LIBPATH "/lib:/usr/lib"
376 #endif
378 #define EXTENSIONS_DIR "/lib/ext"
379 #define ENDORSED_DIR "/lib/endorsed"
380 #define REG_DIR "/usr/java/packages"
382 {
383 /* sysclasspath, java_home, dll_dir */
384 {
385 char *home_path;
386 char *dll_path;
387 char *pslash;
388 char buf[MAXPATHLEN];
389 os::jvm_path(buf, sizeof(buf));
391 // Found the full path to libjvm.so.
392 // Now cut the path to <java_home>/jre if we can.
393 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
394 pslash = strrchr(buf, '/');
395 if (pslash != NULL)
396 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
397 dll_path = malloc(strlen(buf) + 1);
398 if (dll_path == NULL)
399 return;
400 strcpy(dll_path, buf);
401 Arguments::set_dll_dir(dll_path);
403 if (pslash != NULL) {
404 pslash = strrchr(buf, '/');
405 if (pslash != NULL) {
406 *pslash = '\0'; /* get rid of /<arch> */
407 pslash = strrchr(buf, '/');
408 if (pslash != NULL)
409 *pslash = '\0'; /* get rid of /lib */
410 }
411 }
413 home_path = malloc(strlen(buf) + 1);
414 if (home_path == NULL)
415 return;
416 strcpy(home_path, buf);
417 Arguments::set_java_home(home_path);
419 if (!set_boot_path('/', ':'))
420 return;
421 }
423 /*
424 * Where to look for native libraries
425 *
426 * Note: Due to a legacy implementation, most of the library path
427 * is set in the launcher. This was to accomodate linking restrictions
428 * on legacy Linux implementations (which are no longer supported).
429 * Eventually, all the library path setting will be done here.
430 *
431 * However, to prevent the proliferation of improperly built native
432 * libraries, the new path component /usr/java/packages is added here.
433 * Eventually, all the library path setting will be done here.
434 */
435 {
436 char *ld_library_path;
438 /*
439 * Construct the invariant part of ld_library_path. Note that the
440 * space for the colon and the trailing null are provided by the
441 * nulls included by the sizeof operator (so actually we allocate
442 * a byte more than necessary).
443 */
444 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
445 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
446 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
448 /*
449 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
450 * should always exist (until the legacy problem cited above is
451 * addressed).
452 */
453 char *v = getenv("LD_LIBRARY_PATH");
454 if (v != NULL) {
455 char *t = ld_library_path;
456 /* That's +1 for the colon and +1 for the trailing '\0' */
457 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
458 sprintf(ld_library_path, "%s:%s", v, t);
459 }
460 Arguments::set_library_path(ld_library_path);
461 }
463 /*
464 * Extensions directories.
465 *
466 * Note that the space for the colon and the trailing null are provided
467 * by the nulls included by the sizeof operator (so actually one byte more
468 * than necessary is allocated).
469 */
470 {
471 char *buf = malloc(strlen(Arguments::get_java_home()) +
472 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
473 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
474 Arguments::get_java_home());
475 Arguments::set_ext_dirs(buf);
476 }
478 /* Endorsed standards default directory. */
479 {
480 char * buf;
481 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
482 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
483 Arguments::set_endorsed_dirs(buf);
484 }
485 }
487 #undef malloc
488 #undef getenv
489 #undef EXTENSIONS_DIR
490 #undef ENDORSED_DIR
492 // Done
493 return;
494 }
496 ////////////////////////////////////////////////////////////////////////////////
497 // breakpoint support
499 void os::breakpoint() {
500 BREAKPOINT;
501 }
503 extern "C" void breakpoint() {
504 // use debugger to set breakpoint here
505 }
507 ////////////////////////////////////////////////////////////////////////////////
508 // signal support
510 debug_only(static bool signal_sets_initialized = false);
511 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
513 bool os::Linux::is_sig_ignored(int sig) {
514 struct sigaction oact;
515 sigaction(sig, (struct sigaction*)NULL, &oact);
516 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
517 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
518 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
519 return true;
520 else
521 return false;
522 }
524 void os::Linux::signal_sets_init() {
525 // Should also have an assertion stating we are still single-threaded.
526 assert(!signal_sets_initialized, "Already initialized");
527 // Fill in signals that are necessarily unblocked for all threads in
528 // the VM. Currently, we unblock the following signals:
529 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
530 // by -Xrs (=ReduceSignalUsage));
531 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
532 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
533 // the dispositions or masks wrt these signals.
534 // Programs embedding the VM that want to use the above signals for their
535 // own purposes must, at this time, use the "-Xrs" option to prevent
536 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
537 // (See bug 4345157, and other related bugs).
538 // In reality, though, unblocking these signals is really a nop, since
539 // these signals are not blocked by default.
540 sigemptyset(&unblocked_sigs);
541 sigemptyset(&allowdebug_blocked_sigs);
542 sigaddset(&unblocked_sigs, SIGILL);
543 sigaddset(&unblocked_sigs, SIGSEGV);
544 sigaddset(&unblocked_sigs, SIGBUS);
545 sigaddset(&unblocked_sigs, SIGFPE);
546 #if defined(PPC64)
547 sigaddset(&unblocked_sigs, SIGTRAP);
548 #endif
549 sigaddset(&unblocked_sigs, SR_signum);
551 if (!ReduceSignalUsage) {
552 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
553 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
554 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
555 }
556 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
557 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
558 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
559 }
560 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
561 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
562 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
563 }
564 }
565 // Fill in signals that are blocked by all but the VM thread.
566 sigemptyset(&vm_sigs);
567 if (!ReduceSignalUsage)
568 sigaddset(&vm_sigs, BREAK_SIGNAL);
569 debug_only(signal_sets_initialized = true);
571 }
573 // These are signals that are unblocked while a thread is running Java.
574 // (For some reason, they get blocked by default.)
575 sigset_t* os::Linux::unblocked_signals() {
576 assert(signal_sets_initialized, "Not initialized");
577 return &unblocked_sigs;
578 }
580 // These are the signals that are blocked while a (non-VM) thread is
581 // running Java. Only the VM thread handles these signals.
582 sigset_t* os::Linux::vm_signals() {
583 assert(signal_sets_initialized, "Not initialized");
584 return &vm_sigs;
585 }
587 // These are signals that are blocked during cond_wait to allow debugger in
588 sigset_t* os::Linux::allowdebug_blocked_signals() {
589 assert(signal_sets_initialized, "Not initialized");
590 return &allowdebug_blocked_sigs;
591 }
593 void os::Linux::hotspot_sigmask(Thread* thread) {
595 //Save caller's signal mask before setting VM signal mask
596 sigset_t caller_sigmask;
597 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
599 OSThread* osthread = thread->osthread();
600 osthread->set_caller_sigmask(caller_sigmask);
602 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
604 if (!ReduceSignalUsage) {
605 if (thread->is_VM_thread()) {
606 // Only the VM thread handles BREAK_SIGNAL ...
607 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
608 } else {
609 // ... all other threads block BREAK_SIGNAL
610 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
611 }
612 }
613 }
615 //////////////////////////////////////////////////////////////////////////////
616 // detecting pthread library
618 void os::Linux::libpthread_init() {
619 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
620 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
621 // generic name for earlier versions.
622 // Define macros here so we can build HotSpot on old systems.
623 # ifndef _CS_GNU_LIBC_VERSION
624 # define _CS_GNU_LIBC_VERSION 2
625 # endif
626 # ifndef _CS_GNU_LIBPTHREAD_VERSION
627 # define _CS_GNU_LIBPTHREAD_VERSION 3
628 # endif
630 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
631 if (n > 0) {
632 char *str = (char *)malloc(n, mtInternal);
633 confstr(_CS_GNU_LIBC_VERSION, str, n);
634 os::Linux::set_glibc_version(str);
635 } else {
636 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
637 static char _gnu_libc_version[32];
638 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
639 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
640 os::Linux::set_glibc_version(_gnu_libc_version);
641 }
643 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
644 if (n > 0) {
645 char *str = (char *)malloc(n, mtInternal);
646 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
647 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
648 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
649 // is the case. LinuxThreads has a hard limit on max number of threads.
650 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
651 // On the other hand, NPTL does not have such a limit, sysconf()
652 // will return -1 and errno is not changed. Check if it is really NPTL.
653 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
654 strstr(str, "NPTL") &&
655 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
656 free(str);
657 os::Linux::set_libpthread_version("linuxthreads");
658 } else {
659 os::Linux::set_libpthread_version(str);
660 }
661 } else {
662 // glibc before 2.3.2 only has LinuxThreads.
663 os::Linux::set_libpthread_version("linuxthreads");
664 }
666 if (strstr(libpthread_version(), "NPTL")) {
667 os::Linux::set_is_NPTL();
668 } else {
669 os::Linux::set_is_LinuxThreads();
670 }
672 // LinuxThreads have two flavors: floating-stack mode, which allows variable
673 // stack size; and fixed-stack mode. NPTL is always floating-stack.
674 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
675 os::Linux::set_is_floating_stack();
676 }
677 }
679 /////////////////////////////////////////////////////////////////////////////
680 // thread stack
682 // Force Linux kernel to expand current thread stack. If "bottom" is close
683 // to the stack guard, caller should block all signals.
684 //
685 // MAP_GROWSDOWN:
686 // A special mmap() flag that is used to implement thread stacks. It tells
687 // kernel that the memory region should extend downwards when needed. This
688 // allows early versions of LinuxThreads to only mmap the first few pages
689 // when creating a new thread. Linux kernel will automatically expand thread
690 // stack as needed (on page faults).
691 //
692 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
693 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
694 // region, it's hard to tell if the fault is due to a legitimate stack
695 // access or because of reading/writing non-exist memory (e.g. buffer
696 // overrun). As a rule, if the fault happens below current stack pointer,
697 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
698 // application (see Linux kernel fault.c).
699 //
700 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
701 // stack overflow detection.
702 //
703 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
704 // not use this flag. However, the stack of initial thread is not created
705 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
706 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
707 // and then attach the thread to JVM.
708 //
709 // To get around the problem and allow stack banging on Linux, we need to
710 // manually expand thread stack after receiving the SIGSEGV.
711 //
712 // There are two ways to expand thread stack to address "bottom", we used
713 // both of them in JVM before 1.5:
714 // 1. adjust stack pointer first so that it is below "bottom", and then
715 // touch "bottom"
716 // 2. mmap() the page in question
717 //
718 // Now alternate signal stack is gone, it's harder to use 2. For instance,
719 // if current sp is already near the lower end of page 101, and we need to
720 // call mmap() to map page 100, it is possible that part of the mmap() frame
721 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
722 // That will destroy the mmap() frame and cause VM to crash.
723 //
724 // The following code works by adjusting sp first, then accessing the "bottom"
725 // page to force a page fault. Linux kernel will then automatically expand the
726 // stack mapping.
727 //
728 // _expand_stack_to() assumes its frame size is less than page size, which
729 // should always be true if the function is not inlined.
731 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
732 #define NOINLINE
733 #else
734 #define NOINLINE __attribute__ ((noinline))
735 #endif
737 static void _expand_stack_to(address bottom) NOINLINE;
739 static void _expand_stack_to(address bottom) {
740 address sp;
741 size_t size;
742 volatile char *p;
744 // Adjust bottom to point to the largest address within the same page, it
745 // gives us a one-page buffer if alloca() allocates slightly more memory.
746 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
747 bottom += os::Linux::page_size() - 1;
749 // sp might be slightly above current stack pointer; if that's the case, we
750 // will alloca() a little more space than necessary, which is OK. Don't use
751 // os::current_stack_pointer(), as its result can be slightly below current
752 // stack pointer, causing us to not alloca enough to reach "bottom".
753 sp = (address)&sp;
755 if (sp > bottom) {
756 size = sp - bottom;
757 p = (volatile char *)alloca(size);
758 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
759 p[0] = '\0';
760 }
761 }
763 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
764 assert(t!=NULL, "just checking");
765 assert(t->osthread()->expanding_stack(), "expand should be set");
766 assert(t->stack_base() != NULL, "stack_base was not initialized");
768 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
769 sigset_t mask_all, old_sigset;
770 sigfillset(&mask_all);
771 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
772 _expand_stack_to(addr);
773 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
774 return true;
775 }
776 return false;
777 }
779 //////////////////////////////////////////////////////////////////////////////
780 // create new thread
782 static address highest_vm_reserved_address();
784 // check if it's safe to start a new thread
785 static bool _thread_safety_check(Thread* thread) {
786 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
787 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
788 // Heap is mmap'ed at lower end of memory space. Thread stacks are
789 // allocated (MAP_FIXED) from high address space. Every thread stack
790 // occupies a fixed size slot (usually 2Mbytes, but user can change
791 // it to other values if they rebuild LinuxThreads).
792 //
793 // Problem with MAP_FIXED is that mmap() can still succeed even part of
794 // the memory region has already been mmap'ed. That means if we have too
795 // many threads and/or very large heap, eventually thread stack will
796 // collide with heap.
797 //
798 // Here we try to prevent heap/stack collision by comparing current
799 // stack bottom with the highest address that has been mmap'ed by JVM
800 // plus a safety margin for memory maps created by native code.
801 //
802 // This feature can be disabled by setting ThreadSafetyMargin to 0
803 //
804 if (ThreadSafetyMargin > 0) {
805 address stack_bottom = os::current_stack_base() - os::current_stack_size();
807 // not safe if our stack extends below the safety margin
808 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
809 } else {
810 return true;
811 }
812 } else {
813 // Floating stack LinuxThreads or NPTL:
814 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
815 // there's not enough space left, pthread_create() will fail. If we come
816 // here, that means enough space has been reserved for stack.
817 return true;
818 }
819 }
821 // Thread start routine for all newly created threads
822 static void *java_start(Thread *thread) {
823 // Try to randomize the cache line index of hot stack frames.
824 // This helps when threads of the same stack traces evict each other's
825 // cache lines. The threads can be either from the same JVM instance, or
826 // from different JVM instances. The benefit is especially true for
827 // processors with hyperthreading technology.
828 static int counter = 0;
829 int pid = os::current_process_id();
830 alloca(((pid ^ counter++) & 7) * 128);
832 ThreadLocalStorage::set_thread(thread);
834 OSThread* osthread = thread->osthread();
835 Monitor* sync = osthread->startThread_lock();
837 // non floating stack LinuxThreads needs extra check, see above
838 if (!_thread_safety_check(thread)) {
839 // notify parent thread
840 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
841 osthread->set_state(ZOMBIE);
842 sync->notify_all();
843 return NULL;
844 }
846 // thread_id is kernel thread id (similar to Solaris LWP id)
847 osthread->set_thread_id(os::Linux::gettid());
849 if (UseNUMA) {
850 int lgrp_id = os::numa_get_group_id();
851 if (lgrp_id != -1) {
852 thread->set_lgrp_id(lgrp_id);
853 }
854 }
855 // initialize signal mask for this thread
856 os::Linux::hotspot_sigmask(thread);
858 // initialize floating point control register
859 os::Linux::init_thread_fpu_state();
861 // handshaking with parent thread
862 {
863 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
865 // notify parent thread
866 osthread->set_state(INITIALIZED);
867 sync->notify_all();
869 // wait until os::start_thread()
870 while (osthread->get_state() == INITIALIZED) {
871 sync->wait(Mutex::_no_safepoint_check_flag);
872 }
873 }
875 // call one more level start routine
876 thread->run();
878 return 0;
879 }
881 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
882 assert(thread->osthread() == NULL, "caller responsible");
884 // Allocate the OSThread object
885 OSThread* osthread = new OSThread(NULL, NULL);
886 if (osthread == NULL) {
887 return false;
888 }
890 // set the correct thread state
891 osthread->set_thread_type(thr_type);
893 // Initial state is ALLOCATED but not INITIALIZED
894 osthread->set_state(ALLOCATED);
896 thread->set_osthread(osthread);
898 // init thread attributes
899 pthread_attr_t attr;
900 pthread_attr_init(&attr);
901 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
903 // stack size
904 if (os::Linux::supports_variable_stack_size()) {
905 // calculate stack size if it's not specified by caller
906 if (stack_size == 0) {
907 stack_size = os::Linux::default_stack_size(thr_type);
909 switch (thr_type) {
910 case os::java_thread:
911 // Java threads use ThreadStackSize which default value can be
912 // changed with the flag -Xss
913 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
914 stack_size = JavaThread::stack_size_at_create();
915 break;
916 case os::compiler_thread:
917 if (CompilerThreadStackSize > 0) {
918 stack_size = (size_t)(CompilerThreadStackSize * K);
919 break;
920 } // else fall through:
921 // use VMThreadStackSize if CompilerThreadStackSize is not defined
922 case os::vm_thread:
923 case os::pgc_thread:
924 case os::cgc_thread:
925 case os::watcher_thread:
926 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
927 break;
928 }
929 }
931 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
932 pthread_attr_setstacksize(&attr, stack_size);
933 } else {
934 // let pthread_create() pick the default value.
935 }
937 // glibc guard page
938 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
940 ThreadState state;
942 {
943 // Serialize thread creation if we are running with fixed stack LinuxThreads
944 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
945 if (lock) {
946 os::Linux::createThread_lock()->lock_without_safepoint_check();
947 }
949 pthread_t tid;
950 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
952 pthread_attr_destroy(&attr);
954 if (ret != 0) {
955 if (PrintMiscellaneous && (Verbose || WizardMode)) {
956 perror("pthread_create()");
957 }
958 // Need to clean up stuff we've allocated so far
959 thread->set_osthread(NULL);
960 delete osthread;
961 if (lock) os::Linux::createThread_lock()->unlock();
962 return false;
963 }
965 // Store pthread info into the OSThread
966 osthread->set_pthread_id(tid);
968 // Wait until child thread is either initialized or aborted
969 {
970 Monitor* sync_with_child = osthread->startThread_lock();
971 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
972 while ((state = osthread->get_state()) == ALLOCATED) {
973 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
974 }
975 }
977 if (lock) {
978 os::Linux::createThread_lock()->unlock();
979 }
980 }
982 // Aborted due to thread limit being reached
983 if (state == ZOMBIE) {
984 thread->set_osthread(NULL);
985 delete osthread;
986 return false;
987 }
989 // The thread is returned suspended (in state INITIALIZED),
990 // and is started higher up in the call chain
991 assert(state == INITIALIZED, "race condition");
992 return true;
993 }
995 /////////////////////////////////////////////////////////////////////////////
996 // attach existing thread
998 // bootstrap the main thread
999 bool os::create_main_thread(JavaThread* thread) {
1000 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
1001 return create_attached_thread(thread);
1002 }
1004 bool os::create_attached_thread(JavaThread* thread) {
1005 #ifdef ASSERT
1006 thread->verify_not_published();
1007 #endif
1009 // Allocate the OSThread object
1010 OSThread* osthread = new OSThread(NULL, NULL);
1012 if (osthread == NULL) {
1013 return false;
1014 }
1016 // Store pthread info into the OSThread
1017 osthread->set_thread_id(os::Linux::gettid());
1018 osthread->set_pthread_id(::pthread_self());
1020 // initialize floating point control register
1021 os::Linux::init_thread_fpu_state();
1023 // Initial thread state is RUNNABLE
1024 osthread->set_state(RUNNABLE);
1026 thread->set_osthread(osthread);
1028 if (UseNUMA) {
1029 int lgrp_id = os::numa_get_group_id();
1030 if (lgrp_id != -1) {
1031 thread->set_lgrp_id(lgrp_id);
1032 }
1033 }
1035 if (os::Linux::is_initial_thread()) {
1036 // If current thread is initial thread, its stack is mapped on demand,
1037 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1038 // the entire stack region to avoid SEGV in stack banging.
1039 // It is also useful to get around the heap-stack-gap problem on SuSE
1040 // kernel (see 4821821 for details). We first expand stack to the top
1041 // of yellow zone, then enable stack yellow zone (order is significant,
1042 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1043 // is no gap between the last two virtual memory regions.
1045 JavaThread *jt = (JavaThread *)thread;
1046 address addr = jt->stack_yellow_zone_base();
1047 assert(addr != NULL, "initialization problem?");
1048 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1050 osthread->set_expanding_stack();
1051 os::Linux::manually_expand_stack(jt, addr);
1052 osthread->clear_expanding_stack();
1053 }
1055 // initialize signal mask for this thread
1056 // and save the caller's signal mask
1057 os::Linux::hotspot_sigmask(thread);
1059 return true;
1060 }
1062 void os::pd_start_thread(Thread* thread) {
1063 OSThread * osthread = thread->osthread();
1064 assert(osthread->get_state() != INITIALIZED, "just checking");
1065 Monitor* sync_with_child = osthread->startThread_lock();
1066 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1067 sync_with_child->notify();
1069 #ifdef MIPS64
1070 /* 2013/11/5 Jin: To be accessed in NativeGeneralJump::patch_verified_entry() */
1071 if (thread->is_Java_thread())
1072 {
1073 ((JavaThread*)thread)->set_handle_wrong_method_stub(SharedRuntime::get_handle_wrong_method_stub());
1074 }
1075 #endif
1076 }
1078 // Free Linux resources related to the OSThread
1079 void os::free_thread(OSThread* osthread) {
1080 assert(osthread != NULL, "osthread not set");
1082 if (Thread::current()->osthread() == osthread) {
1083 // Restore caller's signal mask
1084 sigset_t sigmask = osthread->caller_sigmask();
1085 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1086 }
1088 delete osthread;
1089 }
1091 //////////////////////////////////////////////////////////////////////////////
1092 // thread local storage
1094 // Restore the thread pointer if the destructor is called. This is in case
1095 // someone from JNI code sets up a destructor with pthread_key_create to run
1096 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1097 // will hang or crash. When detachCurrentThread is called the key will be set
1098 // to null and we will not be called again. If detachCurrentThread is never
1099 // called we could loop forever depending on the pthread implementation.
1100 static void restore_thread_pointer(void* p) {
1101 Thread* thread = (Thread*) p;
1102 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1103 }
1105 int os::allocate_thread_local_storage() {
1106 pthread_key_t key;
1107 int rslt = pthread_key_create(&key, restore_thread_pointer);
1108 assert(rslt == 0, "cannot allocate thread local storage");
1109 return (int)key;
1110 }
1112 // Note: This is currently not used by VM, as we don't destroy TLS key
1113 // on VM exit.
1114 void os::free_thread_local_storage(int index) {
1115 int rslt = pthread_key_delete((pthread_key_t)index);
1116 assert(rslt == 0, "invalid index");
1117 }
1119 void os::thread_local_storage_at_put(int index, void* value) {
1120 int rslt = pthread_setspecific((pthread_key_t)index, value);
1121 assert(rslt == 0, "pthread_setspecific failed");
1122 }
1124 extern "C" Thread* get_thread() {
1125 return ThreadLocalStorage::thread();
1126 }
1128 //////////////////////////////////////////////////////////////////////////////
1129 // initial thread
1131 // Check if current thread is the initial thread, similar to Solaris thr_main.
1132 bool os::Linux::is_initial_thread(void) {
1133 char dummy;
1134 // If called before init complete, thread stack bottom will be null.
1135 // Can be called if fatal error occurs before initialization.
1136 if (initial_thread_stack_bottom() == NULL) return false;
1137 assert(initial_thread_stack_bottom() != NULL &&
1138 initial_thread_stack_size() != 0,
1139 "os::init did not locate initial thread's stack region");
1140 if ((address)&dummy >= initial_thread_stack_bottom() &&
1141 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1142 return true;
1143 else return false;
1144 }
1146 // Find the virtual memory area that contains addr
1147 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1148 FILE *fp = fopen("/proc/self/maps", "r");
1149 if (fp) {
1150 address low, high;
1151 while (!feof(fp)) {
1152 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1153 if (low <= addr && addr < high) {
1154 if (vma_low) *vma_low = low;
1155 if (vma_high) *vma_high = high;
1156 fclose (fp);
1157 return true;
1158 }
1159 }
1160 for (;;) {
1161 int ch = fgetc(fp);
1162 if (ch == EOF || ch == (int)'\n') break;
1163 }
1164 }
1165 fclose(fp);
1166 }
1167 return false;
1168 }
1170 // Locate initial thread stack. This special handling of initial thread stack
1171 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1172 // bogus value for initial thread.
1173 void os::Linux::capture_initial_stack(size_t max_size) {
1174 // stack size is the easy part, get it from RLIMIT_STACK
1175 size_t stack_size;
1176 struct rlimit rlim;
1177 getrlimit(RLIMIT_STACK, &rlim);
1178 stack_size = rlim.rlim_cur;
1180 // 6308388: a bug in ld.so will relocate its own .data section to the
1181 // lower end of primordial stack; reduce ulimit -s value a little bit
1182 // so we won't install guard page on ld.so's data section.
1183 stack_size -= 2 * page_size();
1185 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1186 // 7.1, in both cases we will get 2G in return value.
1187 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1188 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1189 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1190 // in case other parts in glibc still assumes 2M max stack size.
1191 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1192 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1193 if (stack_size > 2 * K * K IA64_ONLY(*2))
1194 stack_size = 2 * K * K IA64_ONLY(*2);
1195 // Try to figure out where the stack base (top) is. This is harder.
1196 //
1197 // When an application is started, glibc saves the initial stack pointer in
1198 // a global variable "__libc_stack_end", which is then used by system
1199 // libraries. __libc_stack_end should be pretty close to stack top. The
1200 // variable is available since the very early days. However, because it is
1201 // a private interface, it could disappear in the future.
1202 //
1203 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1204 // to __libc_stack_end, it is very close to stack top, but isn't the real
1205 // stack top. Note that /proc may not exist if VM is running as a chroot
1206 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1207 // /proc/<pid>/stat could change in the future (though unlikely).
1208 //
1209 // We try __libc_stack_end first. If that doesn't work, look for
1210 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1211 // as a hint, which should work well in most cases.
1213 uintptr_t stack_start;
1215 // try __libc_stack_end first
1216 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1217 if (p && *p) {
1218 stack_start = *p;
1219 } else {
1220 // see if we can get the start_stack field from /proc/self/stat
1221 FILE *fp;
1222 int pid;
1223 char state;
1224 int ppid;
1225 int pgrp;
1226 int session;
1227 int nr;
1228 int tpgrp;
1229 unsigned long flags;
1230 unsigned long minflt;
1231 unsigned long cminflt;
1232 unsigned long majflt;
1233 unsigned long cmajflt;
1234 unsigned long utime;
1235 unsigned long stime;
1236 long cutime;
1237 long cstime;
1238 long prio;
1239 long nice;
1240 long junk;
1241 long it_real;
1242 uintptr_t start;
1243 uintptr_t vsize;
1244 intptr_t rss;
1245 uintptr_t rsslim;
1246 uintptr_t scodes;
1247 uintptr_t ecode;
1248 int i;
1250 // Figure what the primordial thread stack base is. Code is inspired
1251 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1252 // followed by command name surrounded by parentheses, state, etc.
1253 char stat[2048];
1254 int statlen;
1256 fp = fopen("/proc/self/stat", "r");
1257 if (fp) {
1258 statlen = fread(stat, 1, 2047, fp);
1259 stat[statlen] = '\0';
1260 fclose(fp);
1262 // Skip pid and the command string. Note that we could be dealing with
1263 // weird command names, e.g. user could decide to rename java launcher
1264 // to "java 1.4.2 :)", then the stat file would look like
1265 // 1234 (java 1.4.2 :)) R ... ...
1266 // We don't really need to know the command string, just find the last
1267 // occurrence of ")" and then start parsing from there. See bug 4726580.
1268 char * s = strrchr(stat, ')');
1270 i = 0;
1271 if (s) {
1272 // Skip blank chars
1273 do s++; while (isspace(*s));
1275 #define _UFM UINTX_FORMAT
1276 #define _DFM INTX_FORMAT
1278 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1279 /* 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 */
1280 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,
1281 &state, /* 3 %c */
1282 &ppid, /* 4 %d */
1283 &pgrp, /* 5 %d */
1284 &session, /* 6 %d */
1285 &nr, /* 7 %d */
1286 &tpgrp, /* 8 %d */
1287 &flags, /* 9 %lu */
1288 &minflt, /* 10 %lu */
1289 &cminflt, /* 11 %lu */
1290 &majflt, /* 12 %lu */
1291 &cmajflt, /* 13 %lu */
1292 &utime, /* 14 %lu */
1293 &stime, /* 15 %lu */
1294 &cutime, /* 16 %ld */
1295 &cstime, /* 17 %ld */
1296 &prio, /* 18 %ld */
1297 &nice, /* 19 %ld */
1298 &junk, /* 20 %ld */
1299 &it_real, /* 21 %ld */
1300 &start, /* 22 UINTX_FORMAT */
1301 &vsize, /* 23 UINTX_FORMAT */
1302 &rss, /* 24 INTX_FORMAT */
1303 &rsslim, /* 25 UINTX_FORMAT */
1304 &scodes, /* 26 UINTX_FORMAT */
1305 &ecode, /* 27 UINTX_FORMAT */
1306 &stack_start); /* 28 UINTX_FORMAT */
1307 }
1309 #undef _UFM
1310 #undef _DFM
1312 if (i != 28 - 2) {
1313 assert(false, "Bad conversion from /proc/self/stat");
1314 // product mode - assume we are the initial thread, good luck in the
1315 // embedded case.
1316 warning("Can't detect initial thread stack location - bad conversion");
1317 stack_start = (uintptr_t) &rlim;
1318 }
1319 } else {
1320 // For some reason we can't open /proc/self/stat (for example, running on
1321 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1322 // most cases, so don't abort:
1323 warning("Can't detect initial thread stack location - no /proc/self/stat");
1324 stack_start = (uintptr_t) &rlim;
1325 }
1326 }
1328 // Now we have a pointer (stack_start) very close to the stack top, the
1329 // next thing to do is to figure out the exact location of stack top. We
1330 // can find out the virtual memory area that contains stack_start by
1331 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1332 // and its upper limit is the real stack top. (again, this would fail if
1333 // running inside chroot, because /proc may not exist.)
1335 uintptr_t stack_top;
1336 address low, high;
1337 if (find_vma((address)stack_start, &low, &high)) {
1338 // success, "high" is the true stack top. (ignore "low", because initial
1339 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1340 stack_top = (uintptr_t)high;
1341 } else {
1342 // failed, likely because /proc/self/maps does not exist
1343 warning("Can't detect initial thread stack location - find_vma failed");
1344 // best effort: stack_start is normally within a few pages below the real
1345 // stack top, use it as stack top, and reduce stack size so we won't put
1346 // guard page outside stack.
1347 stack_top = stack_start;
1348 stack_size -= 16 * page_size();
1349 }
1351 // stack_top could be partially down the page so align it
1352 stack_top = align_size_up(stack_top, page_size());
1354 if (max_size && stack_size > max_size) {
1355 _initial_thread_stack_size = max_size;
1356 } else {
1357 _initial_thread_stack_size = stack_size;
1358 }
1360 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1361 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1362 }
1364 ////////////////////////////////////////////////////////////////////////////////
1365 // time support
1367 // Time since start-up in seconds to a fine granularity.
1368 // Used by VMSelfDestructTimer and the MemProfiler.
1369 double os::elapsedTime() {
1371 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1372 }
1374 jlong os::elapsed_counter() {
1375 return javaTimeNanos() - initial_time_count;
1376 }
1378 jlong os::elapsed_frequency() {
1379 return NANOSECS_PER_SEC; // nanosecond resolution
1380 }
1382 bool os::supports_vtime() { return true; }
1383 bool os::enable_vtime() { return false; }
1384 bool os::vtime_enabled() { return false; }
1386 double os::elapsedVTime() {
1387 struct rusage usage;
1388 int retval = getrusage(RUSAGE_THREAD, &usage);
1389 if (retval == 0) {
1390 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1391 } else {
1392 // better than nothing, but not much
1393 return elapsedTime();
1394 }
1395 }
1397 jlong os::javaTimeMillis() {
1398 timeval time;
1399 int status = gettimeofday(&time, NULL);
1400 assert(status != -1, "linux error");
1401 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1402 }
1404 #ifndef CLOCK_MONOTONIC
1405 #define CLOCK_MONOTONIC (1)
1406 #endif
1408 void os::Linux::clock_init() {
1409 // we do dlopen's in this particular order due to bug in linux
1410 // dynamical loader (see 6348968) leading to crash on exit
1411 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1412 if (handle == NULL) {
1413 handle = dlopen("librt.so", RTLD_LAZY);
1414 }
1416 if (handle) {
1417 int (*clock_getres_func)(clockid_t, struct timespec*) =
1418 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1419 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1420 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1421 if (clock_getres_func && clock_gettime_func) {
1422 // See if monotonic clock is supported by the kernel. Note that some
1423 // early implementations simply return kernel jiffies (updated every
1424 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1425 // for nano time (though the monotonic property is still nice to have).
1426 // It's fixed in newer kernels, however clock_getres() still returns
1427 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1428 // resolution for now. Hopefully as people move to new kernels, this
1429 // won't be a problem.
1430 struct timespec res;
1431 struct timespec tp;
1432 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1433 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1434 // yes, monotonic clock is supported
1435 _clock_gettime = clock_gettime_func;
1436 return;
1437 } else {
1438 // close librt if there is no monotonic clock
1439 dlclose(handle);
1440 }
1441 }
1442 }
1443 warning("No monotonic clock was available - timed services may " \
1444 "be adversely affected if the time-of-day clock changes");
1445 }
1447 #ifndef SYS_clock_getres
1449 #if defined(IA32) || defined(AMD64)
1450 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1451 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1452 #else
1453 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1454 #define sys_clock_getres(x,y) -1
1455 #endif
1457 #else
1458 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1459 #endif
1461 void os::Linux::fast_thread_clock_init() {
1462 if (!UseLinuxPosixThreadCPUClocks) {
1463 return;
1464 }
1465 clockid_t clockid;
1466 struct timespec tp;
1467 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1468 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1470 // Switch to using fast clocks for thread cpu time if
1471 // the sys_clock_getres() returns 0 error code.
1472 // Note, that some kernels may support the current thread
1473 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1474 // returned by the pthread_getcpuclockid().
1475 // If the fast Posix clocks are supported then the sys_clock_getres()
1476 // must return at least tp.tv_sec == 0 which means a resolution
1477 // better than 1 sec. This is extra check for reliability.
1479 if(pthread_getcpuclockid_func &&
1480 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1481 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1483 _supports_fast_thread_cpu_time = true;
1484 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1485 }
1486 }
1488 jlong os::javaTimeNanos() {
1489 if (Linux::supports_monotonic_clock()) {
1490 struct timespec tp;
1491 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1492 assert(status == 0, "gettime error");
1493 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1494 return result;
1495 } else {
1496 timeval time;
1497 int status = gettimeofday(&time, NULL);
1498 assert(status != -1, "linux error");
1499 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1500 return 1000 * usecs;
1501 }
1502 }
1504 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1505 if (Linux::supports_monotonic_clock()) {
1506 info_ptr->max_value = ALL_64_BITS;
1508 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1509 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1510 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1511 } else {
1512 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1513 info_ptr->max_value = ALL_64_BITS;
1515 // gettimeofday is a real time clock so it skips
1516 info_ptr->may_skip_backward = true;
1517 info_ptr->may_skip_forward = true;
1518 }
1520 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1521 }
1523 // Return the real, user, and system times in seconds from an
1524 // arbitrary fixed point in the past.
1525 bool os::getTimesSecs(double* process_real_time,
1526 double* process_user_time,
1527 double* process_system_time) {
1528 struct tms ticks;
1529 clock_t real_ticks = times(&ticks);
1531 if (real_ticks == (clock_t) (-1)) {
1532 return false;
1533 } else {
1534 double ticks_per_second = (double) clock_tics_per_sec;
1535 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1536 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1537 *process_real_time = ((double) real_ticks) / ticks_per_second;
1539 return true;
1540 }
1541 }
1544 char * os::local_time_string(char *buf, size_t buflen) {
1545 struct tm t;
1546 time_t long_time;
1547 time(&long_time);
1548 localtime_r(&long_time, &t);
1549 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1550 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1551 t.tm_hour, t.tm_min, t.tm_sec);
1552 return buf;
1553 }
1555 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1556 return localtime_r(clock, res);
1557 }
1559 ////////////////////////////////////////////////////////////////////////////////
1560 // runtime exit support
1562 // Note: os::shutdown() might be called very early during initialization, or
1563 // called from signal handler. Before adding something to os::shutdown(), make
1564 // sure it is async-safe and can handle partially initialized VM.
1565 void os::shutdown() {
1567 // allow PerfMemory to attempt cleanup of any persistent resources
1568 perfMemory_exit();
1570 // needs to remove object in file system
1571 AttachListener::abort();
1573 // flush buffered output, finish log files
1574 ostream_abort();
1576 // Check for abort hook
1577 abort_hook_t abort_hook = Arguments::abort_hook();
1578 if (abort_hook != NULL) {
1579 abort_hook();
1580 }
1582 }
1584 // Note: os::abort() might be called very early during initialization, or
1585 // called from signal handler. Before adding something to os::abort(), make
1586 // sure it is async-safe and can handle partially initialized VM.
1587 void os::abort(bool dump_core) {
1588 os::shutdown();
1589 if (dump_core) {
1590 #ifndef PRODUCT
1591 fdStream out(defaultStream::output_fd());
1592 out.print_raw("Current thread is ");
1593 char buf[16];
1594 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1595 out.print_raw_cr(buf);
1596 out.print_raw_cr("Dumping core ...");
1597 #endif
1598 ::abort(); // dump core
1599 }
1601 ::exit(1);
1602 }
1604 // Die immediately, no exit hook, no abort hook, no cleanup.
1605 void os::die() {
1606 // _exit() on LinuxThreads only kills current thread
1607 ::abort();
1608 }
1611 // This method is a copy of JDK's sysGetLastErrorString
1612 // from src/solaris/hpi/src/system_md.c
1614 size_t os::lasterror(char *buf, size_t len) {
1616 if (errno == 0) return 0;
1618 const char *s = ::strerror(errno);
1619 size_t n = ::strlen(s);
1620 if (n >= len) {
1621 n = len - 1;
1622 }
1623 ::strncpy(buf, s, n);
1624 buf[n] = '\0';
1625 return n;
1626 }
1628 intx os::current_thread_id() { return (intx)pthread_self(); }
1629 int os::current_process_id() {
1631 // Under the old linux thread library, linux gives each thread
1632 // its own process id. Because of this each thread will return
1633 // a different pid if this method were to return the result
1634 // of getpid(2). Linux provides no api that returns the pid
1635 // of the launcher thread for the vm. This implementation
1636 // returns a unique pid, the pid of the launcher thread
1637 // that starts the vm 'process'.
1639 // Under the NPTL, getpid() returns the same pid as the
1640 // launcher thread rather than a unique pid per thread.
1641 // Use gettid() if you want the old pre NPTL behaviour.
1643 // if you are looking for the result of a call to getpid() that
1644 // returns a unique pid for the calling thread, then look at the
1645 // OSThread::thread_id() method in osThread_linux.hpp file
1647 return (int)(_initial_pid ? _initial_pid : getpid());
1648 }
1650 // DLL functions
1652 const char* os::dll_file_extension() { return ".so"; }
1654 // This must be hard coded because it's the system's temporary
1655 // directory not the java application's temp directory, ala java.io.tmpdir.
1656 const char* os::get_temp_directory() { return "/tmp"; }
1658 static bool file_exists(const char* filename) {
1659 struct stat statbuf;
1660 if (filename == NULL || strlen(filename) == 0) {
1661 return false;
1662 }
1663 return os::stat(filename, &statbuf) == 0;
1664 }
1666 bool os::dll_build_name(char* buffer, size_t buflen,
1667 const char* pname, const char* fname) {
1668 bool retval = false;
1669 // Copied from libhpi
1670 const size_t pnamelen = pname ? strlen(pname) : 0;
1672 // Return error on buffer overflow.
1673 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1674 return retval;
1675 }
1677 if (pnamelen == 0) {
1678 snprintf(buffer, buflen, "lib%s.so", fname);
1679 retval = true;
1680 } else if (strchr(pname, *os::path_separator()) != NULL) {
1681 int n;
1682 char** pelements = split_path(pname, &n);
1683 if (pelements == NULL) {
1684 return false;
1685 }
1686 for (int i = 0 ; i < n ; i++) {
1687 // Really shouldn't be NULL, but check can't hurt
1688 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1689 continue; // skip the empty path values
1690 }
1691 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1692 if (file_exists(buffer)) {
1693 retval = true;
1694 break;
1695 }
1696 }
1697 // release the storage
1698 for (int i = 0 ; i < n ; i++) {
1699 if (pelements[i] != NULL) {
1700 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1701 }
1702 }
1703 if (pelements != NULL) {
1704 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1705 }
1706 } else {
1707 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1708 retval = true;
1709 }
1710 return retval;
1711 }
1713 // check if addr is inside libjvm.so
1714 bool os::address_is_in_vm(address addr) {
1715 static address libjvm_base_addr;
1716 Dl_info dlinfo;
1718 if (libjvm_base_addr == NULL) {
1719 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1720 libjvm_base_addr = (address)dlinfo.dli_fbase;
1721 }
1722 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1723 }
1725 if (dladdr((void *)addr, &dlinfo) != 0) {
1726 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1727 }
1729 return false;
1730 }
1732 bool os::dll_address_to_function_name(address addr, char *buf,
1733 int buflen, int *offset) {
1734 // buf is not optional, but offset is optional
1735 assert(buf != NULL, "sanity check");
1737 Dl_info dlinfo;
1739 if (dladdr((void*)addr, &dlinfo) != 0) {
1740 // see if we have a matching symbol
1741 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1742 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1743 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1744 }
1745 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1746 return true;
1747 }
1748 // no matching symbol so try for just file info
1749 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1750 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1751 buf, buflen, offset, dlinfo.dli_fname)) {
1752 return true;
1753 }
1754 }
1755 }
1757 buf[0] = '\0';
1758 if (offset != NULL) *offset = -1;
1759 return false;
1760 }
1762 struct _address_to_library_name {
1763 address addr; // input : memory address
1764 size_t buflen; // size of fname
1765 char* fname; // output: library name
1766 address base; // library base addr
1767 };
1769 static int address_to_library_name_callback(struct dl_phdr_info *info,
1770 size_t size, void *data) {
1771 int i;
1772 bool found = false;
1773 address libbase = NULL;
1774 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1776 // iterate through all loadable segments
1777 for (i = 0; i < info->dlpi_phnum; i++) {
1778 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1779 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1780 // base address of a library is the lowest address of its loaded
1781 // segments.
1782 if (libbase == NULL || libbase > segbase) {
1783 libbase = segbase;
1784 }
1785 // see if 'addr' is within current segment
1786 if (segbase <= d->addr &&
1787 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1788 found = true;
1789 }
1790 }
1791 }
1793 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1794 // so dll_address_to_library_name() can fall through to use dladdr() which
1795 // can figure out executable name from argv[0].
1796 if (found && info->dlpi_name && info->dlpi_name[0]) {
1797 d->base = libbase;
1798 if (d->fname) {
1799 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1800 }
1801 return 1;
1802 }
1803 return 0;
1804 }
1806 bool os::dll_address_to_library_name(address addr, char* buf,
1807 int buflen, int* offset) {
1808 // buf is not optional, but offset is optional
1809 assert(buf != NULL, "sanity check");
1811 Dl_info dlinfo;
1812 struct _address_to_library_name data;
1814 // There is a bug in old glibc dladdr() implementation that it could resolve
1815 // to wrong library name if the .so file has a base address != NULL. Here
1816 // we iterate through the program headers of all loaded libraries to find
1817 // out which library 'addr' really belongs to. This workaround can be
1818 // removed once the minimum requirement for glibc is moved to 2.3.x.
1819 data.addr = addr;
1820 data.fname = buf;
1821 data.buflen = buflen;
1822 data.base = NULL;
1823 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1825 if (rslt) {
1826 // buf already contains library name
1827 if (offset) *offset = addr - data.base;
1828 return true;
1829 }
1830 if (dladdr((void*)addr, &dlinfo) != 0) {
1831 if (dlinfo.dli_fname != NULL) {
1832 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1833 }
1834 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1835 *offset = addr - (address)dlinfo.dli_fbase;
1836 }
1837 return true;
1838 }
1840 buf[0] = '\0';
1841 if (offset) *offset = -1;
1842 return false;
1843 }
1845 // Loads .dll/.so and
1846 // in case of error it checks if .dll/.so was built for the
1847 // same architecture as Hotspot is running on
1850 // Remember the stack's state. The Linux dynamic linker will change
1851 // the stack to 'executable' at most once, so we must safepoint only once.
1852 bool os::Linux::_stack_is_executable = false;
1854 // VM operation that loads a library. This is necessary if stack protection
1855 // of the Java stacks can be lost during loading the library. If we
1856 // do not stop the Java threads, they can stack overflow before the stacks
1857 // are protected again.
1858 class VM_LinuxDllLoad: public VM_Operation {
1859 private:
1860 const char *_filename;
1861 char *_ebuf;
1862 int _ebuflen;
1863 void *_lib;
1864 public:
1865 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1866 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1867 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1868 void doit() {
1869 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1870 os::Linux::_stack_is_executable = true;
1871 }
1872 void* loaded_library() { return _lib; }
1873 };
1875 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1876 {
1877 void * result = NULL;
1878 bool load_attempted = false;
1880 // Check whether the library to load might change execution rights
1881 // of the stack. If they are changed, the protection of the stack
1882 // guard pages will be lost. We need a safepoint to fix this.
1883 //
1884 // See Linux man page execstack(8) for more info.
1885 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1886 ElfFile ef(filename);
1887 if (!ef.specifies_noexecstack()) {
1888 if (!is_init_completed()) {
1889 os::Linux::_stack_is_executable = true;
1890 // This is OK - No Java threads have been created yet, and hence no
1891 // stack guard pages to fix.
1892 //
1893 // This should happen only when you are building JDK7 using a very
1894 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1895 //
1896 // Dynamic loader will make all stacks executable after
1897 // this function returns, and will not do that again.
1898 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1899 } else {
1900 warning("You have loaded library %s which might have disabled stack guard. "
1901 "The VM will try to fix the stack guard now.\n"
1902 "It's highly recommended that you fix the library with "
1903 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1904 filename);
1906 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1907 JavaThread *jt = JavaThread::current();
1908 if (jt->thread_state() != _thread_in_native) {
1909 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1910 // that requires ExecStack. Cannot enter safe point. Let's give up.
1911 warning("Unable to fix stack guard. Giving up.");
1912 } else {
1913 if (!LoadExecStackDllInVMThread) {
1914 // This is for the case where the DLL has an static
1915 // constructor function that executes JNI code. We cannot
1916 // load such DLLs in the VMThread.
1917 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1918 }
1920 ThreadInVMfromNative tiv(jt);
1921 debug_only(VMNativeEntryWrapper vew;)
1923 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1924 VMThread::execute(&op);
1925 if (LoadExecStackDllInVMThread) {
1926 result = op.loaded_library();
1927 }
1928 load_attempted = true;
1929 }
1930 }
1931 }
1932 }
1934 if (!load_attempted) {
1935 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1936 }
1938 if (result != NULL) {
1939 // Successful loading
1940 return result;
1941 }
1943 Elf32_Ehdr elf_head;
1944 int diag_msg_max_length=ebuflen-strlen(ebuf);
1945 char* diag_msg_buf=ebuf+strlen(ebuf);
1947 if (diag_msg_max_length==0) {
1948 // No more space in ebuf for additional diagnostics message
1949 return NULL;
1950 }
1953 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1955 if (file_descriptor < 0) {
1956 // Can't open library, report dlerror() message
1957 return NULL;
1958 }
1960 bool failed_to_read_elf_head=
1961 (sizeof(elf_head)!=
1962 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1964 ::close(file_descriptor);
1965 if (failed_to_read_elf_head) {
1966 // file i/o error - report dlerror() msg
1967 return NULL;
1968 }
1970 typedef struct {
1971 Elf32_Half code; // Actual value as defined in elf.h
1972 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1973 char elf_class; // 32 or 64 bit
1974 char endianess; // MSB or LSB
1975 char* name; // String representation
1976 } arch_t;
1978 #ifndef EM_486
1979 #define EM_486 6 /* Intel 80486 */
1980 #endif
1982 static const arch_t arch_array[]={
1983 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1984 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1985 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1986 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1987 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1988 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1989 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1990 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1991 #if defined(VM_LITTLE_ENDIAN)
1992 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1993 #else
1994 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1995 #endif
1996 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1997 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1998 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1999 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
2000 {EM_MIPS, EM_MIPS, ELFCLASS64, ELFDATA2LSB, (char*)"MIPS64 LE"},
2001 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
2002 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
2003 };
2005 #if (defined IA32)
2006 static Elf32_Half running_arch_code=EM_386;
2007 #elif (defined AMD64)
2008 static Elf32_Half running_arch_code=EM_X86_64;
2009 #elif (defined IA64)
2010 static Elf32_Half running_arch_code=EM_IA_64;
2011 #elif (defined __sparc) && (defined _LP64)
2012 static Elf32_Half running_arch_code=EM_SPARCV9;
2013 #elif (defined __sparc) && (!defined _LP64)
2014 static Elf32_Half running_arch_code=EM_SPARC;
2015 #elif (defined MIPS64)
2016 static Elf32_Half running_arch_code=EM_MIPS;
2017 #elif (defined __powerpc64__)
2018 static Elf32_Half running_arch_code=EM_PPC64;
2019 #elif (defined __powerpc__)
2020 static Elf32_Half running_arch_code=EM_PPC;
2021 #elif (defined ARM)
2022 static Elf32_Half running_arch_code=EM_ARM;
2023 #elif (defined S390)
2024 static Elf32_Half running_arch_code=EM_S390;
2025 #elif (defined ALPHA)
2026 static Elf32_Half running_arch_code=EM_ALPHA;
2027 #elif (defined MIPSEL)
2028 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2029 #elif (defined PARISC)
2030 static Elf32_Half running_arch_code=EM_PARISC;
2031 #elif (defined MIPS)
2032 static Elf32_Half running_arch_code=EM_MIPS;
2033 #elif (defined M68K)
2034 static Elf32_Half running_arch_code=EM_68K;
2035 #else
2036 #error Method os::dll_load requires that one of following is defined:\
2037 IA32, AMD64, IA64, __mips64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2038 #endif
2040 // Identify compatability class for VM's architecture and library's architecture
2041 // Obtain string descriptions for architectures
2043 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2044 int running_arch_index=-1;
2046 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2047 if (running_arch_code == arch_array[i].code) {
2048 running_arch_index = i;
2049 }
2050 if (lib_arch.code == arch_array[i].code) {
2051 lib_arch.compat_class = arch_array[i].compat_class;
2052 lib_arch.name = arch_array[i].name;
2053 }
2054 }
2056 assert(running_arch_index != -1,
2057 "Didn't find running architecture code (running_arch_code) in arch_array");
2058 if (running_arch_index == -1) {
2059 // Even though running architecture detection failed
2060 // we may still continue with reporting dlerror() message
2061 return NULL;
2062 }
2064 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2065 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2066 return NULL;
2067 }
2069 #ifndef S390
2070 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2071 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2072 return NULL;
2073 }
2074 #endif // !S390
2076 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2077 if ( lib_arch.name!=NULL ) {
2078 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2079 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2080 lib_arch.name, arch_array[running_arch_index].name);
2081 } else {
2082 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2083 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2084 lib_arch.code,
2085 arch_array[running_arch_index].name);
2086 }
2087 }
2089 return NULL;
2090 }
2092 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2093 void * result = ::dlopen(filename, RTLD_LAZY);
2094 if (result == NULL) {
2095 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2096 ebuf[ebuflen-1] = '\0';
2097 }
2098 return result;
2099 }
2101 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2102 void * result = NULL;
2103 if (LoadExecStackDllInVMThread) {
2104 result = dlopen_helper(filename, ebuf, ebuflen);
2105 }
2107 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2108 // library that requires an executable stack, or which does not have this
2109 // stack attribute set, dlopen changes the stack attribute to executable. The
2110 // read protection of the guard pages gets lost.
2111 //
2112 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2113 // may have been queued at the same time.
2115 if (!_stack_is_executable) {
2116 JavaThread *jt = Threads::first();
2118 while (jt) {
2119 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2120 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2121 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2122 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2123 warning("Attempt to reguard stack yellow zone failed.");
2124 }
2125 }
2126 jt = jt->next();
2127 }
2128 }
2130 return result;
2131 }
2133 /*
2134 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2135 * chances are you might want to run the generated bits against glibc-2.0
2136 * libdl.so, so always use locking for any version of glibc.
2137 */
2138 void* os::dll_lookup(void* handle, const char* name) {
2139 pthread_mutex_lock(&dl_mutex);
2140 void* res = dlsym(handle, name);
2141 pthread_mutex_unlock(&dl_mutex);
2142 return res;
2143 }
2145 void* os::get_default_process_handle() {
2146 return (void*)::dlopen(NULL, RTLD_LAZY);
2147 }
2149 static bool _print_ascii_file(const char* filename, outputStream* st) {
2150 int fd = ::open(filename, O_RDONLY);
2151 if (fd == -1) {
2152 return false;
2153 }
2155 char buf[32];
2156 int bytes;
2157 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2158 st->print_raw(buf, bytes);
2159 }
2161 ::close(fd);
2163 return true;
2164 }
2166 void os::print_dll_info(outputStream *st) {
2167 st->print_cr("Dynamic libraries:");
2169 char fname[32];
2170 pid_t pid = os::Linux::gettid();
2172 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2174 if (!_print_ascii_file(fname, st)) {
2175 st->print("Can not get library information for pid = %d\n", pid);
2176 }
2177 }
2179 void os::print_os_info_brief(outputStream* st) {
2180 os::Linux::print_distro_info(st);
2182 os::Posix::print_uname_info(st);
2184 os::Linux::print_libversion_info(st);
2186 }
2188 void os::print_os_info(outputStream* st) {
2189 st->print("OS:");
2191 os::Linux::print_distro_info(st);
2193 os::Posix::print_uname_info(st);
2195 // Print warning if unsafe chroot environment detected
2196 if (unsafe_chroot_detected) {
2197 st->print("WARNING!! ");
2198 st->print_cr("%s", unstable_chroot_error);
2199 }
2201 os::Linux::print_libversion_info(st);
2203 os::Posix::print_rlimit_info(st);
2205 os::Posix::print_load_average(st);
2207 os::Linux::print_full_memory_info(st);
2208 }
2210 // Try to identify popular distros.
2211 // Most Linux distributions have a /etc/XXX-release file, which contains
2212 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2213 // file that also contains the OS version string. Some have more than one
2214 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2215 // /etc/redhat-release.), so the order is important.
2216 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2217 // their own specific XXX-release file as well as a redhat-release file.
2218 // Because of this the XXX-release file needs to be searched for before the
2219 // redhat-release file.
2220 // Since Red Hat has a lsb-release file that is not very descriptive the
2221 // search for redhat-release needs to be before lsb-release.
2222 // Since the lsb-release file is the new standard it needs to be searched
2223 // before the older style release files.
2224 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2225 // next to last resort. The os-release file is a new standard that contains
2226 // distribution information and the system-release file seems to be an old
2227 // standard that has been replaced by the lsb-release and os-release files.
2228 // Searching for the debian_version file is the last resort. It contains
2229 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2230 // "Debian " is printed before the contents of the debian_version file.
2231 void os::Linux::print_distro_info(outputStream* st) {
2232 if (!_print_ascii_file("/etc/oracle-release", st) &&
2233 !_print_ascii_file("/etc/mandriva-release", st) &&
2234 !_print_ascii_file("/etc/mandrake-release", st) &&
2235 !_print_ascii_file("/etc/sun-release", st) &&
2236 !_print_ascii_file("/etc/redhat-release", st) &&
2237 !_print_ascii_file("/etc/lsb-release", st) &&
2238 !_print_ascii_file("/etc/SuSE-release", st) &&
2239 !_print_ascii_file("/etc/turbolinux-release", st) &&
2240 !_print_ascii_file("/etc/gentoo-release", st) &&
2241 !_print_ascii_file("/etc/ltib-release", st) &&
2242 !_print_ascii_file("/etc/angstrom-version", st) &&
2243 !_print_ascii_file("/etc/system-release", st) &&
2244 !_print_ascii_file("/etc/os-release", st)) {
2246 if (file_exists("/etc/debian_version")) {
2247 st->print("Debian ");
2248 _print_ascii_file("/etc/debian_version", st);
2249 } else {
2250 st->print("Linux");
2251 }
2252 }
2253 st->cr();
2254 }
2256 void os::Linux::print_libversion_info(outputStream* st) {
2257 // libc, pthread
2258 st->print("libc:");
2259 st->print("%s ", os::Linux::glibc_version());
2260 st->print("%s ", os::Linux::libpthread_version());
2261 if (os::Linux::is_LinuxThreads()) {
2262 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2263 }
2264 st->cr();
2265 }
2267 void os::Linux::print_full_memory_info(outputStream* st) {
2268 st->print("\n/proc/meminfo:\n");
2269 _print_ascii_file("/proc/meminfo", st);
2270 st->cr();
2271 }
2273 void os::print_memory_info(outputStream* st) {
2275 st->print("Memory:");
2276 st->print(" %dk page", os::vm_page_size()>>10);
2278 // values in struct sysinfo are "unsigned long"
2279 struct sysinfo si;
2280 sysinfo(&si);
2282 st->print(", physical " UINT64_FORMAT "k",
2283 os::physical_memory() >> 10);
2284 st->print("(" UINT64_FORMAT "k free)",
2285 os::available_memory() >> 10);
2286 st->print(", swap " UINT64_FORMAT "k",
2287 ((jlong)si.totalswap * si.mem_unit) >> 10);
2288 st->print("(" UINT64_FORMAT "k free)",
2289 ((jlong)si.freeswap * si.mem_unit) >> 10);
2290 st->cr();
2291 }
2293 void os::pd_print_cpu_info(outputStream* st) {
2294 st->print("\n/proc/cpuinfo:\n");
2295 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2296 st->print(" <Not Available>");
2297 }
2298 st->cr();
2299 }
2301 void os::print_siginfo(outputStream* st, void* siginfo) {
2302 const siginfo_t* si = (const siginfo_t*)siginfo;
2304 os::Posix::print_siginfo_brief(st, si);
2306 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2307 UseSharedSpaces) {
2308 FileMapInfo* mapinfo = FileMapInfo::current_info();
2309 if (mapinfo->is_in_shared_space(si->si_addr)) {
2310 st->print("\n\nError accessing class data sharing archive." \
2311 " Mapped file inaccessible during execution, " \
2312 " possible disk/network problem.");
2313 }
2314 }
2315 st->cr();
2316 }
2319 static void print_signal_handler(outputStream* st, int sig,
2320 char* buf, size_t buflen);
2322 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2323 st->print_cr("Signal Handlers:");
2324 print_signal_handler(st, SIGSEGV, buf, buflen);
2325 print_signal_handler(st, SIGBUS , buf, buflen);
2326 print_signal_handler(st, SIGFPE , buf, buflen);
2327 print_signal_handler(st, SIGPIPE, buf, buflen);
2328 print_signal_handler(st, SIGXFSZ, buf, buflen);
2329 print_signal_handler(st, SIGILL , buf, buflen);
2330 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2331 print_signal_handler(st, SR_signum, buf, buflen);
2332 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2333 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2334 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2335 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2336 #if defined(PPC64)
2337 print_signal_handler(st, SIGTRAP, buf, buflen);
2338 #endif
2339 }
2341 static char saved_jvm_path[MAXPATHLEN] = {0};
2343 // Find the full path to the current module, libjvm.so
2344 void os::jvm_path(char *buf, jint buflen) {
2345 // Error checking.
2346 if (buflen < MAXPATHLEN) {
2347 assert(false, "must use a large-enough buffer");
2348 buf[0] = '\0';
2349 return;
2350 }
2351 // Lazy resolve the path to current module.
2352 if (saved_jvm_path[0] != 0) {
2353 strcpy(buf, saved_jvm_path);
2354 return;
2355 }
2357 char dli_fname[MAXPATHLEN];
2358 bool ret = dll_address_to_library_name(
2359 CAST_FROM_FN_PTR(address, os::jvm_path),
2360 dli_fname, sizeof(dli_fname), NULL);
2361 assert(ret, "cannot locate libjvm");
2362 char *rp = NULL;
2363 if (ret && dli_fname[0] != '\0') {
2364 rp = realpath(dli_fname, buf);
2365 }
2366 if (rp == NULL)
2367 return;
2369 if (Arguments::created_by_gamma_launcher()) {
2370 // Support for the gamma launcher. Typical value for buf is
2371 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2372 // the right place in the string, then assume we are installed in a JDK and
2373 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2374 // up the path so it looks like libjvm.so is installed there (append a
2375 // fake suffix hotspot/libjvm.so).
2376 const char *p = buf + strlen(buf) - 1;
2377 for (int count = 0; p > buf && count < 5; ++count) {
2378 for (--p; p > buf && *p != '/'; --p)
2379 /* empty */ ;
2380 }
2382 if (strncmp(p, "/jre/lib/", 9) != 0) {
2383 // Look for JAVA_HOME in the environment.
2384 char* java_home_var = ::getenv("JAVA_HOME");
2385 if (java_home_var != NULL && java_home_var[0] != 0) {
2386 char* jrelib_p;
2387 int len;
2389 // Check the current module name "libjvm.so".
2390 p = strrchr(buf, '/');
2391 assert(strstr(p, "/libjvm") == p, "invalid library name");
2393 rp = realpath(java_home_var, buf);
2394 if (rp == NULL)
2395 return;
2397 // determine if this is a legacy image or modules image
2398 // modules image doesn't have "jre" subdirectory
2399 len = strlen(buf);
2400 assert(len < buflen, "Ran out of buffer room");
2401 jrelib_p = buf + len;
2402 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2403 if (0 != access(buf, F_OK)) {
2404 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2405 }
2407 if (0 == access(buf, F_OK)) {
2408 // Use current module name "libjvm.so"
2409 len = strlen(buf);
2410 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2411 } else {
2412 // Go back to path of .so
2413 rp = realpath(dli_fname, buf);
2414 if (rp == NULL)
2415 return;
2416 }
2417 }
2418 }
2419 }
2421 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2422 }
2424 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2425 // no prefix required, not even "_"
2426 }
2428 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2429 // no suffix required
2430 }
2432 ////////////////////////////////////////////////////////////////////////////////
2433 // sun.misc.Signal support
2435 static volatile jint sigint_count = 0;
2437 static void
2438 UserHandler(int sig, void *siginfo, void *context) {
2439 // 4511530 - sem_post is serialized and handled by the manager thread. When
2440 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2441 // don't want to flood the manager thread with sem_post requests.
2442 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2443 return;
2445 // Ctrl-C is pressed during error reporting, likely because the error
2446 // handler fails to abort. Let VM die immediately.
2447 if (sig == SIGINT && is_error_reported()) {
2448 os::die();
2449 }
2451 os::signal_notify(sig);
2452 }
2454 void* os::user_handler() {
2455 return CAST_FROM_FN_PTR(void*, UserHandler);
2456 }
2458 class Semaphore : public StackObj {
2459 public:
2460 Semaphore();
2461 ~Semaphore();
2462 void signal();
2463 void wait();
2464 bool trywait();
2465 bool timedwait(unsigned int sec, int nsec);
2466 private:
2467 sem_t _semaphore;
2468 };
2470 Semaphore::Semaphore() {
2471 sem_init(&_semaphore, 0, 0);
2472 }
2474 Semaphore::~Semaphore() {
2475 sem_destroy(&_semaphore);
2476 }
2478 void Semaphore::signal() {
2479 sem_post(&_semaphore);
2480 }
2482 void Semaphore::wait() {
2483 sem_wait(&_semaphore);
2484 }
2486 bool Semaphore::trywait() {
2487 return sem_trywait(&_semaphore) == 0;
2488 }
2490 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2492 struct timespec ts;
2493 // Semaphore's are always associated with CLOCK_REALTIME
2494 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2495 // see unpackTime for discussion on overflow checking
2496 if (sec >= MAX_SECS) {
2497 ts.tv_sec += MAX_SECS;
2498 ts.tv_nsec = 0;
2499 } else {
2500 ts.tv_sec += sec;
2501 ts.tv_nsec += nsec;
2502 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2503 ts.tv_nsec -= NANOSECS_PER_SEC;
2504 ++ts.tv_sec; // note: this must be <= max_secs
2505 }
2506 }
2508 while (1) {
2509 int result = sem_timedwait(&_semaphore, &ts);
2510 if (result == 0) {
2511 return true;
2512 } else if (errno == EINTR) {
2513 continue;
2514 } else if (errno == ETIMEDOUT) {
2515 return false;
2516 } else {
2517 return false;
2518 }
2519 }
2520 }
2522 extern "C" {
2523 typedef void (*sa_handler_t)(int);
2524 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2525 }
2527 void* os::signal(int signal_number, void* handler) {
2528 struct sigaction sigAct, oldSigAct;
2530 sigfillset(&(sigAct.sa_mask));
2531 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2532 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2534 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2535 // -1 means registration failed
2536 return (void *)-1;
2537 }
2539 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2540 }
2542 void os::signal_raise(int signal_number) {
2543 ::raise(signal_number);
2544 }
2546 /*
2547 * The following code is moved from os.cpp for making this
2548 * code platform specific, which it is by its very nature.
2549 */
2551 // Will be modified when max signal is changed to be dynamic
2552 int os::sigexitnum_pd() {
2553 return NSIG;
2554 }
2556 // a counter for each possible signal value
2557 static volatile jint pending_signals[NSIG+1] = { 0 };
2559 // Linux(POSIX) specific hand shaking semaphore.
2560 static sem_t sig_sem;
2561 static Semaphore sr_semaphore;
2563 void os::signal_init_pd() {
2564 // Initialize signal structures
2565 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2567 // Initialize signal semaphore
2568 ::sem_init(&sig_sem, 0, 0);
2569 }
2571 void os::signal_notify(int sig) {
2572 Atomic::inc(&pending_signals[sig]);
2573 ::sem_post(&sig_sem);
2574 }
2576 static int check_pending_signals(bool wait) {
2577 Atomic::store(0, &sigint_count);
2578 for (;;) {
2579 for (int i = 0; i < NSIG + 1; i++) {
2580 jint n = pending_signals[i];
2581 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2582 return i;
2583 }
2584 }
2585 if (!wait) {
2586 return -1;
2587 }
2588 JavaThread *thread = JavaThread::current();
2589 ThreadBlockInVM tbivm(thread);
2591 bool threadIsSuspended;
2592 do {
2593 thread->set_suspend_equivalent();
2594 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2595 ::sem_wait(&sig_sem);
2597 // were we externally suspended while we were waiting?
2598 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2599 if (threadIsSuspended) {
2600 //
2601 // The semaphore has been incremented, but while we were waiting
2602 // another thread suspended us. We don't want to continue running
2603 // while suspended because that would surprise the thread that
2604 // suspended us.
2605 //
2606 ::sem_post(&sig_sem);
2608 thread->java_suspend_self();
2609 }
2610 } while (threadIsSuspended);
2611 }
2612 }
2614 int os::signal_lookup() {
2615 return check_pending_signals(false);
2616 }
2618 int os::signal_wait() {
2619 return check_pending_signals(true);
2620 }
2622 ////////////////////////////////////////////////////////////////////////////////
2623 // Virtual Memory
2625 int os::vm_page_size() {
2626 // Seems redundant as all get out
2627 assert(os::Linux::page_size() != -1, "must call os::init");
2628 return os::Linux::page_size();
2629 }
2631 // Solaris allocates memory by pages.
2632 int os::vm_allocation_granularity() {
2633 assert(os::Linux::page_size() != -1, "must call os::init");
2634 return os::Linux::page_size();
2635 }
2637 // Rationale behind this function:
2638 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2639 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2640 // samples for JITted code. Here we create private executable mapping over the code cache
2641 // and then we can use standard (well, almost, as mapping can change) way to provide
2642 // info for the reporting script by storing timestamp and location of symbol
2643 void linux_wrap_code(char* base, size_t size) {
2644 static volatile jint cnt = 0;
2646 if (!UseOprofile) {
2647 return;
2648 }
2650 char buf[PATH_MAX+1];
2651 int num = Atomic::add(1, &cnt);
2653 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2654 os::get_temp_directory(), os::current_process_id(), num);
2655 unlink(buf);
2657 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2659 if (fd != -1) {
2660 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2661 if (rv != (off_t)-1) {
2662 if (::write(fd, "", 1) == 1) {
2663 mmap(base, size,
2664 PROT_READ|PROT_WRITE|PROT_EXEC,
2665 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2666 }
2667 }
2668 ::close(fd);
2669 unlink(buf);
2670 }
2671 }
2673 static bool recoverable_mmap_error(int err) {
2674 // See if the error is one we can let the caller handle. This
2675 // list of errno values comes from JBS-6843484. I can't find a
2676 // Linux man page that documents this specific set of errno
2677 // values so while this list currently matches Solaris, it may
2678 // change as we gain experience with this failure mode.
2679 switch (err) {
2680 case EBADF:
2681 case EINVAL:
2682 case ENOTSUP:
2683 // let the caller deal with these errors
2684 return true;
2686 default:
2687 // Any remaining errors on this OS can cause our reserved mapping
2688 // to be lost. That can cause confusion where different data
2689 // structures think they have the same memory mapped. The worst
2690 // scenario is if both the VM and a library think they have the
2691 // same memory mapped.
2692 return false;
2693 }
2694 }
2696 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2697 int err) {
2698 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2699 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2700 strerror(err), err);
2701 }
2703 static void warn_fail_commit_memory(char* addr, size_t size,
2704 size_t alignment_hint, bool exec,
2705 int err) {
2706 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2707 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2708 alignment_hint, exec, strerror(err), err);
2709 }
2711 // NOTE: Linux kernel does not really reserve the pages for us.
2712 // All it does is to check if there are enough free pages
2713 // left at the time of mmap(). This could be a potential
2714 // problem.
2715 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2716 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2717 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2718 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2719 if (res != (uintptr_t) MAP_FAILED) {
2720 if (UseNUMAInterleaving) {
2721 numa_make_global(addr, size);
2722 }
2723 return 0;
2724 }
2726 int err = errno; // save errno from mmap() call above
2728 if (!recoverable_mmap_error(err)) {
2729 warn_fail_commit_memory(addr, size, exec, err);
2730 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2731 }
2733 return err;
2734 }
2736 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2737 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2738 }
2740 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2741 const char* mesg) {
2742 assert(mesg != NULL, "mesg must be specified");
2743 int err = os::Linux::commit_memory_impl(addr, size, exec);
2744 if (err != 0) {
2745 // the caller wants all commit errors to exit with the specified mesg:
2746 warn_fail_commit_memory(addr, size, exec, err);
2747 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2748 }
2749 }
2751 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2752 #ifndef MAP_HUGETLB
2753 #define MAP_HUGETLB 0x40000
2754 #endif
2756 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2757 #ifndef MADV_HUGEPAGE
2758 #define MADV_HUGEPAGE 14
2759 #endif
2761 int os::Linux::commit_memory_impl(char* addr, size_t size,
2762 size_t alignment_hint, bool exec) {
2763 int err = os::Linux::commit_memory_impl(addr, size, exec);
2764 if (err == 0) {
2765 realign_memory(addr, size, alignment_hint);
2766 }
2767 return err;
2768 }
2770 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2771 bool exec) {
2772 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2773 }
2775 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2776 size_t alignment_hint, bool exec,
2777 const char* mesg) {
2778 assert(mesg != NULL, "mesg must be specified");
2779 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2780 if (err != 0) {
2781 // the caller wants all commit errors to exit with the specified mesg:
2782 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2783 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2784 }
2785 }
2787 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2788 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2789 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2790 // be supported or the memory may already be backed by huge pages.
2791 ::madvise(addr, bytes, MADV_HUGEPAGE);
2792 }
2793 }
2795 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2796 // This method works by doing an mmap over an existing mmaping and effectively discarding
2797 // the existing pages. However it won't work for SHM-based large pages that cannot be
2798 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2799 // small pages on top of the SHM segment. This method always works for small pages, so we
2800 // allow that in any case.
2801 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2802 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2803 }
2804 }
2806 void os::numa_make_global(char *addr, size_t bytes) {
2807 Linux::numa_interleave_memory(addr, bytes);
2808 }
2810 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2811 // bind policy to MPOL_PREFERRED for the current thread.
2812 #define USE_MPOL_PREFERRED 0
2814 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2815 // To make NUMA and large pages more robust when both enabled, we need to ease
2816 // the requirements on where the memory should be allocated. MPOL_BIND is the
2817 // default policy and it will force memory to be allocated on the specified
2818 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2819 // the specified node, but will not force it. Using this policy will prevent
2820 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2821 // free large pages.
2822 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2823 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2824 }
2826 bool os::numa_topology_changed() { return false; }
2828 size_t os::numa_get_groups_num() {
2829 int max_node = Linux::numa_max_node();
2830 return max_node > 0 ? max_node + 1 : 1;
2831 }
2833 int os::numa_get_group_id() {
2834 int cpu_id = Linux::sched_getcpu();
2835 if (cpu_id != -1) {
2836 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2837 if (lgrp_id != -1) {
2838 return lgrp_id;
2839 }
2840 }
2841 return 0;
2842 }
2844 int os::numa_get_cpu_id() {
2845 int cpu_id = Linux::sched_getcpu();
2846 if(cpu_id != -1)
2847 return cpu_id;
2848 else {
2849 tty->print_cr("cpu_id got a unacceptable value");
2850 return 0;
2851 }
2852 }
2854 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2855 for (size_t i = 0; i < size; i++) {
2856 ids[i] = i;
2857 }
2858 return size;
2859 }
2861 bool os::get_page_info(char *start, page_info* info) {
2862 return false;
2863 }
2865 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2866 return end;
2867 }
2870 int os::Linux::sched_getcpu_syscall(void) {
2871 unsigned int cpu;
2872 int retval = -1;
2874 #if defined(IA32)
2875 # ifndef SYS_getcpu
2876 # define SYS_getcpu 318
2877 # endif
2878 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2879 #elif defined(AMD64)
2880 // Unfortunately we have to bring all these macros here from vsyscall.h
2881 // to be able to compile on old linuxes.
2882 # define __NR_vgetcpu 2
2883 # define VSYSCALL_START (-10UL << 20)
2884 # define VSYSCALL_SIZE 1024
2885 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2886 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2887 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2888 retval = vgetcpu(&cpu, NULL, NULL);
2889 #endif
2891 return (retval == -1) ? retval : cpu;
2892 }
2894 // Something to do with the numa-aware allocator needs these symbols
2895 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2896 extern "C" JNIEXPORT void numa_error(char *where) { }
2897 extern "C" JNIEXPORT int fork1() { return fork(); }
2900 // If we are running with libnuma version > 2, then we should
2901 // be trying to use symbols with versions 1.1
2902 // If we are running with earlier version, which did not have symbol versions,
2903 // we should use the base version.
2904 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2905 void *f = dlvsym(handle, name, "libnuma_1.1");
2906 if (f == NULL) {
2907 f = dlsym(handle, name);
2908 }
2909 return f;
2910 }
2912 bool os::Linux::libnuma_init() {
2913 // sched_getcpu() should be in libc.
2914 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2915 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2917 // If it's not, try a direct syscall.
2918 if (sched_getcpu() == -1)
2919 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2921 if (sched_getcpu() != -1) { // Does it work?
2922 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2923 if (handle != NULL) {
2924 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2925 libnuma_dlsym(handle, "numa_node_to_cpus")));
2926 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2927 libnuma_dlsym(handle, "numa_max_node")));
2928 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2929 libnuma_dlsym(handle, "numa_available")));
2930 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2931 libnuma_dlsym(handle, "numa_tonode_memory")));
2932 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2933 libnuma_dlsym(handle, "numa_interleave_memory")));
2934 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2935 libnuma_dlsym(handle, "numa_set_bind_policy")));
2938 if (numa_available() != -1) {
2939 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2940 // Create a cpu -> node mapping
2941 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2942 rebuild_cpu_to_node_map();
2943 return true;
2944 }
2945 }
2946 }
2947 return false;
2948 }
2950 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2951 // The table is later used in get_node_by_cpu().
2952 void os::Linux::rebuild_cpu_to_node_map() {
2953 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2954 // in libnuma (possible values are starting from 16,
2955 // and continuing up with every other power of 2, but less
2956 // than the maximum number of CPUs supported by kernel), and
2957 // is a subject to change (in libnuma version 2 the requirements
2958 // are more reasonable) we'll just hardcode the number they use
2959 // in the library.
2960 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2962 size_t cpu_num = os::active_processor_count();
2963 size_t cpu_map_size = NCPUS / BitsPerCLong;
2964 size_t cpu_map_valid_size =
2965 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2967 cpu_to_node()->clear();
2968 cpu_to_node()->at_grow(cpu_num - 1);
2969 size_t node_num = numa_get_groups_num();
2971 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2972 for (size_t i = 0; i < node_num; i++) {
2973 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2974 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2975 if (cpu_map[j] != 0) {
2976 for (size_t k = 0; k < BitsPerCLong; k++) {
2977 if (cpu_map[j] & (1UL << k)) {
2978 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2979 }
2980 }
2981 }
2982 }
2983 }
2984 }
2985 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2986 }
2988 int os::Linux::get_node_by_cpu(int cpu_id) {
2989 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2990 return cpu_to_node()->at(cpu_id);
2991 }
2992 return -1;
2993 }
2995 GrowableArray<int>* os::Linux::_cpu_to_node;
2996 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2997 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2998 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2999 os::Linux::numa_available_func_t os::Linux::_numa_available;
3000 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3001 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3002 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3003 unsigned long* os::Linux::_numa_all_nodes;
3005 bool os::pd_uncommit_memory(char* addr, size_t size) {
3006 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3007 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3008 return res != (uintptr_t) MAP_FAILED;
3009 }
3011 static
3012 address get_stack_commited_bottom(address bottom, size_t size) {
3013 address nbot = bottom;
3014 address ntop = bottom + size;
3016 size_t page_sz = os::vm_page_size();
3017 unsigned pages = size / page_sz;
3019 unsigned char vec[1];
3020 unsigned imin = 1, imax = pages + 1, imid;
3021 int mincore_return_value = 0;
3023 assert(imin <= imax, "Unexpected page size");
3025 while (imin < imax) {
3026 imid = (imax + imin) / 2;
3027 nbot = ntop - (imid * page_sz);
3029 // Use a trick with mincore to check whether the page is mapped or not.
3030 // mincore sets vec to 1 if page resides in memory and to 0 if page
3031 // is swapped output but if page we are asking for is unmapped
3032 // it returns -1,ENOMEM
3033 mincore_return_value = mincore(nbot, page_sz, vec);
3035 if (mincore_return_value == -1) {
3036 // Page is not mapped go up
3037 // to find first mapped page
3038 if (errno != EAGAIN) {
3039 assert(errno == ENOMEM, "Unexpected mincore errno");
3040 imax = imid;
3041 }
3042 } else {
3043 // Page is mapped go down
3044 // to find first not mapped page
3045 imin = imid + 1;
3046 }
3047 }
3049 nbot = nbot + page_sz;
3051 // Adjust stack bottom one page up if last checked page is not mapped
3052 if (mincore_return_value == -1) {
3053 nbot = nbot + page_sz;
3054 }
3056 return nbot;
3057 }
3060 // Linux uses a growable mapping for the stack, and if the mapping for
3061 // the stack guard pages is not removed when we detach a thread the
3062 // stack cannot grow beyond the pages where the stack guard was
3063 // mapped. If at some point later in the process the stack expands to
3064 // that point, the Linux kernel cannot expand the stack any further
3065 // because the guard pages are in the way, and a segfault occurs.
3066 //
3067 // However, it's essential not to split the stack region by unmapping
3068 // a region (leaving a hole) that's already part of the stack mapping,
3069 // so if the stack mapping has already grown beyond the guard pages at
3070 // the time we create them, we have to truncate the stack mapping.
3071 // So, we need to know the extent of the stack mapping when
3072 // create_stack_guard_pages() is called.
3074 // We only need this for stacks that are growable: at the time of
3075 // writing thread stacks don't use growable mappings (i.e. those
3076 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3077 // only applies to the main thread.
3079 // If the (growable) stack mapping already extends beyond the point
3080 // where we're going to put our guard pages, truncate the mapping at
3081 // that point by munmap()ping it. This ensures that when we later
3082 // munmap() the guard pages we don't leave a hole in the stack
3083 // mapping. This only affects the main/initial thread
3085 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3087 if (os::Linux::is_initial_thread()) {
3088 // As we manually grow stack up to bottom inside create_attached_thread(),
3089 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3090 // we don't need to do anything special.
3091 // Check it first, before calling heavy function.
3092 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3093 unsigned char vec[1];
3095 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3096 // Fallback to slow path on all errors, including EAGAIN
3097 stack_extent = (uintptr_t) get_stack_commited_bottom(
3098 os::Linux::initial_thread_stack_bottom(),
3099 (size_t)addr - stack_extent);
3100 }
3102 if (stack_extent < (uintptr_t)addr) {
3103 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3104 }
3105 }
3107 return os::commit_memory(addr, size, !ExecMem);
3108 }
3110 // If this is a growable mapping, remove the guard pages entirely by
3111 // munmap()ping them. If not, just call uncommit_memory(). This only
3112 // affects the main/initial thread, but guard against future OS changes
3113 // It's safe to always unmap guard pages for initial thread because we
3114 // always place it right after end of the mapped region
3116 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3117 uintptr_t stack_extent, stack_base;
3119 if (os::Linux::is_initial_thread()) {
3120 return ::munmap(addr, size) == 0;
3121 }
3123 return os::uncommit_memory(addr, size);
3124 }
3126 static address _highest_vm_reserved_address = NULL;
3128 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3129 // at 'requested_addr'. If there are existing memory mappings at the same
3130 // location, however, they will be overwritten. If 'fixed' is false,
3131 // 'requested_addr' is only treated as a hint, the return value may or
3132 // may not start from the requested address. Unlike Linux mmap(), this
3133 // function returns NULL to indicate failure.
3134 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3135 char * addr;
3136 int flags;
3138 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3139 if (fixed) {
3140 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3141 flags |= MAP_FIXED;
3142 }
3144 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3145 // touch an uncommitted page. Otherwise, the read/write might
3146 // succeed if we have enough swap space to back the physical page.
3147 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3148 flags, -1, 0);
3150 if (addr != MAP_FAILED) {
3151 // anon_mmap() should only get called during VM initialization,
3152 // don't need lock (actually we can skip locking even it can be called
3153 // from multiple threads, because _highest_vm_reserved_address is just a
3154 // hint about the upper limit of non-stack memory regions.)
3155 if ((address)addr + bytes > _highest_vm_reserved_address) {
3156 _highest_vm_reserved_address = (address)addr + bytes;
3157 }
3158 }
3160 return addr == MAP_FAILED ? NULL : addr;
3161 }
3163 // Don't update _highest_vm_reserved_address, because there might be memory
3164 // regions above addr + size. If so, releasing a memory region only creates
3165 // a hole in the address space, it doesn't help prevent heap-stack collision.
3166 //
3167 static int anon_munmap(char * addr, size_t size) {
3168 return ::munmap(addr, size) == 0;
3169 }
3171 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3172 size_t alignment_hint) {
3173 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3174 }
3176 bool os::pd_release_memory(char* addr, size_t size) {
3177 return anon_munmap(addr, size);
3178 }
3180 static address highest_vm_reserved_address() {
3181 return _highest_vm_reserved_address;
3182 }
3184 static bool linux_mprotect(char* addr, size_t size, int prot) {
3185 // Linux wants the mprotect address argument to be page aligned.
3186 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3188 // According to SUSv3, mprotect() should only be used with mappings
3189 // established by mmap(), and mmap() always maps whole pages. Unaligned
3190 // 'addr' likely indicates problem in the VM (e.g. trying to change
3191 // protection of malloc'ed or statically allocated memory). Check the
3192 // caller if you hit this assert.
3193 assert(addr == bottom, "sanity check");
3195 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3196 return ::mprotect(bottom, size, prot) == 0;
3197 }
3199 // Set protections specified
3200 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3201 bool is_committed) {
3202 unsigned int p = 0;
3203 switch (prot) {
3204 case MEM_PROT_NONE: p = PROT_NONE; break;
3205 case MEM_PROT_READ: p = PROT_READ; break;
3206 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3207 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3208 default:
3209 ShouldNotReachHere();
3210 }
3211 // is_committed is unused.
3212 return linux_mprotect(addr, bytes, p);
3213 }
3215 bool os::guard_memory(char* addr, size_t size) {
3216 return linux_mprotect(addr, size, PROT_NONE);
3217 }
3219 bool os::unguard_memory(char* addr, size_t size) {
3220 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3221 }
3223 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3224 bool result = false;
3225 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3226 MAP_ANONYMOUS|MAP_PRIVATE,
3227 -1, 0);
3228 if (p != MAP_FAILED) {
3229 void *aligned_p = align_ptr_up(p, page_size);
3231 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3233 munmap(p, page_size * 2);
3234 }
3236 if (warn && !result) {
3237 warning("TransparentHugePages is not supported by the operating system.");
3238 }
3240 return result;
3241 }
3243 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3244 bool result = false;
3245 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3246 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3247 -1, 0);
3249 if (p != MAP_FAILED) {
3250 // We don't know if this really is a huge page or not.
3251 FILE *fp = fopen("/proc/self/maps", "r");
3252 if (fp) {
3253 while (!feof(fp)) {
3254 char chars[257];
3255 long x = 0;
3256 if (fgets(chars, sizeof(chars), fp)) {
3257 if (sscanf(chars, "%lx-%*x", &x) == 1
3258 && x == (long)p) {
3259 if (strstr (chars, "hugepage")) {
3260 result = true;
3261 break;
3262 }
3263 }
3264 }
3265 }
3266 fclose(fp);
3267 }
3268 munmap(p, page_size);
3269 }
3271 if (warn && !result) {
3272 warning("HugeTLBFS is not supported by the operating system.");
3273 }
3275 return result;
3276 }
3278 /*
3279 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3280 *
3281 * From the coredump_filter documentation:
3282 *
3283 * - (bit 0) anonymous private memory
3284 * - (bit 1) anonymous shared memory
3285 * - (bit 2) file-backed private memory
3286 * - (bit 3) file-backed shared memory
3287 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3288 * effective only if the bit 2 is cleared)
3289 * - (bit 5) hugetlb private memory
3290 * - (bit 6) hugetlb shared memory
3291 */
3292 static void set_coredump_filter(void) {
3293 FILE *f;
3294 long cdm;
3296 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3297 return;
3298 }
3300 if (fscanf(f, "%lx", &cdm) != 1) {
3301 fclose(f);
3302 return;
3303 }
3305 rewind(f);
3307 if ((cdm & LARGEPAGES_BIT) == 0) {
3308 cdm |= LARGEPAGES_BIT;
3309 fprintf(f, "%#lx", cdm);
3310 }
3312 fclose(f);
3313 }
3315 // Large page support
3317 static size_t _large_page_size = 0;
3319 void os::large_page_init() {
3320 if (!UseLargePages) {
3321 UseHugeTLBFS = false;
3322 UseSHM = false;
3323 return;
3324 }
3326 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
3327 // If UseLargePages is specified on the command line try both methods,
3328 // if it's default, then try only HugeTLBFS.
3329 if (FLAG_IS_DEFAULT(UseLargePages)) {
3330 UseHugeTLBFS = true;
3331 } else {
3332 UseHugeTLBFS = UseSHM = true;
3333 }
3334 }
3336 if (LargePageSizeInBytes) {
3337 _large_page_size = LargePageSizeInBytes;
3338 } else {
3339 // large_page_size on Linux is used to round up heap size. x86 uses either
3340 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3341 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3342 // page as large as 256M.
3343 //
3344 // Here we try to figure out page size by parsing /proc/meminfo and looking
3345 // for a line with the following format:
3346 // Hugepagesize: 2048 kB
3347 //
3348 // If we can't determine the value (e.g. /proc is not mounted, or the text
3349 // format has been changed), we'll use the largest page size supported by
3350 // the processor.
3352 #ifndef ZERO
3353 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3354 ARM_ONLY(2 * M) PPC_ONLY(4 * M) MIPS64_ONLY(4 * M); //In MIPS _large_page_size is seted 4*M.
3355 #endif // ZERO
3357 FILE *fp = fopen("/proc/meminfo", "r");
3358 if (fp) {
3359 while (!feof(fp)) {
3360 int x = 0;
3361 char buf[16];
3362 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3363 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3364 _large_page_size = x * K;
3365 break;
3366 }
3367 } else {
3368 // skip to next line
3369 for (;;) {
3370 int ch = fgetc(fp);
3371 if (ch == EOF || ch == (int)'\n') break;
3372 }
3373 }
3374 }
3375 fclose(fp);
3376 }
3377 }
3379 // print a warning if any large page related flag is specified on command line
3380 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3382 const size_t default_page_size = (size_t)Linux::page_size();
3383 if (_large_page_size > default_page_size) {
3384 _page_sizes[0] = _large_page_size;
3385 _page_sizes[1] = default_page_size;
3386 _page_sizes[2] = 0;
3387 }
3388 UseHugeTLBFS = UseHugeTLBFS &&
3389 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3391 if (UseHugeTLBFS)
3392 UseSHM = false;
3394 UseLargePages = UseHugeTLBFS || UseSHM;
3396 set_coredump_filter();
3397 }
3399 #ifndef SHM_HUGETLB
3400 #define SHM_HUGETLB 04000
3401 #endif
3403 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3404 // "exec" is passed in but not used. Creating the shared image for
3405 // the code cache doesn't have an SHM_X executable permission to check.
3406 assert(UseLargePages && UseSHM, "only for SHM large pages");
3407 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3409 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3410 return NULL; // Fallback to small pages.
3411 }
3413 key_t key = IPC_PRIVATE;
3414 char *addr;
3416 bool warn_on_failure = UseLargePages &&
3417 (!FLAG_IS_DEFAULT(UseLargePages) ||
3418 !FLAG_IS_DEFAULT(UseSHM) ||
3419 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3420 );
3421 char msg[128];
3423 // Create a large shared memory region to attach to based on size.
3424 // Currently, size is the total size of the heap
3425 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3426 if (shmid == -1) {
3427 // Possible reasons for shmget failure:
3428 // 1. shmmax is too small for Java heap.
3429 // > check shmmax value: cat /proc/sys/kernel/shmmax
3430 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3431 // 2. not enough large page memory.
3432 // > check available large pages: cat /proc/meminfo
3433 // > increase amount of large pages:
3434 // echo new_value > /proc/sys/vm/nr_hugepages
3435 // Note 1: different Linux may use different name for this property,
3436 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3437 // Note 2: it's possible there's enough physical memory available but
3438 // they are so fragmented after a long run that they can't
3439 // coalesce into large pages. Try to reserve large pages when
3440 // the system is still "fresh".
3441 if (warn_on_failure) {
3442 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3443 warning("%s", msg);
3444 }
3445 return NULL;
3446 }
3448 // attach to the region
3449 addr = (char*)shmat(shmid, req_addr, 0);
3450 int err = errno;
3452 // Remove shmid. If shmat() is successful, the actual shared memory segment
3453 // will be deleted when it's detached by shmdt() or when the process
3454 // terminates. If shmat() is not successful this will remove the shared
3455 // segment immediately.
3456 shmctl(shmid, IPC_RMID, NULL);
3458 if ((intptr_t)addr == -1) {
3459 if (warn_on_failure) {
3460 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3461 warning("%s", msg);
3462 }
3463 return NULL;
3464 }
3466 return addr;
3467 }
3469 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3470 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3472 bool warn_on_failure = UseLargePages &&
3473 (!FLAG_IS_DEFAULT(UseLargePages) ||
3474 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3475 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3477 if (warn_on_failure) {
3478 char msg[128];
3479 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3480 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3481 warning("%s", msg);
3482 }
3483 }
3485 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3486 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3487 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3488 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3490 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3491 char* addr = (char*)::mmap(req_addr, bytes, prot,
3492 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3493 -1, 0);
3495 if (addr == MAP_FAILED) {
3496 warn_on_large_pages_failure(req_addr, bytes, errno);
3497 return NULL;
3498 }
3500 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3502 return addr;
3503 }
3505 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3506 size_t large_page_size = os::large_page_size();
3508 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3510 // Allocate small pages.
3512 char* start;
3513 if (req_addr != NULL) {
3514 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3515 assert(is_size_aligned(bytes, alignment), "Must be");
3516 start = os::reserve_memory(bytes, req_addr);
3517 assert(start == NULL || start == req_addr, "Must be");
3518 } else {
3519 start = os::reserve_memory_aligned(bytes, alignment);
3520 }
3522 if (start == NULL) {
3523 return NULL;
3524 }
3526 assert(is_ptr_aligned(start, alignment), "Must be");
3528 // os::reserve_memory_special will record this memory area.
3529 // Need to release it here to prevent overlapping reservations.
3530 MemTracker::record_virtual_memory_release((address)start, bytes);
3532 char* end = start + bytes;
3534 // Find the regions of the allocated chunk that can be promoted to large pages.
3535 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3536 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3538 size_t lp_bytes = lp_end - lp_start;
3540 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3542 if (lp_bytes == 0) {
3543 // The mapped region doesn't even span the start and the end of a large page.
3544 // Fall back to allocate a non-special area.
3545 ::munmap(start, end - start);
3546 return NULL;
3547 }
3549 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3552 void* result;
3554 if (start != lp_start) {
3555 result = ::mmap(start, lp_start - start, prot,
3556 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3557 -1, 0);
3558 if (result == MAP_FAILED) {
3559 ::munmap(lp_start, end - lp_start);
3560 return NULL;
3561 }
3562 }
3564 result = ::mmap(lp_start, lp_bytes, prot,
3565 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3566 -1, 0);
3567 if (result == MAP_FAILED) {
3568 warn_on_large_pages_failure(req_addr, bytes, errno);
3569 // If the mmap above fails, the large pages region will be unmapped and we
3570 // have regions before and after with small pages. Release these regions.
3571 //
3572 // | mapped | unmapped | mapped |
3573 // ^ ^ ^ ^
3574 // start lp_start lp_end end
3575 //
3576 ::munmap(start, lp_start - start);
3577 ::munmap(lp_end, end - lp_end);
3578 return NULL;
3579 }
3581 if (lp_end != end) {
3582 result = ::mmap(lp_end, end - lp_end, prot,
3583 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3584 -1, 0);
3585 if (result == MAP_FAILED) {
3586 ::munmap(start, lp_end - start);
3587 return NULL;
3588 }
3589 }
3591 return start;
3592 }
3594 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3595 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3596 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3597 assert(is_power_of_2(alignment), "Must be");
3598 assert(is_power_of_2(os::large_page_size()), "Must be");
3599 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3601 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3602 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3603 } else {
3604 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3605 }
3606 }
3608 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3609 assert(UseLargePages, "only for large pages");
3611 char* addr;
3612 if (UseSHM) {
3613 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3614 } else {
3615 assert(UseHugeTLBFS, "must be");
3616 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3617 }
3619 if (addr != NULL) {
3620 if (UseNUMAInterleaving) {
3621 numa_make_global(addr, bytes);
3622 }
3624 // The memory is committed
3625 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3626 }
3628 return addr;
3629 }
3631 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3632 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3633 return shmdt(base) == 0;
3634 }
3636 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3637 return pd_release_memory(base, bytes);
3638 }
3640 bool os::release_memory_special(char* base, size_t bytes) {
3641 assert(UseLargePages, "only for large pages");
3643 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3645 bool res;
3646 if (UseSHM) {
3647 res = os::Linux::release_memory_special_shm(base, bytes);
3648 } else {
3649 assert(UseHugeTLBFS, "must be");
3650 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3651 }
3653 if (res) {
3654 tkr.record((address)base, bytes);
3655 } else {
3656 tkr.discard();
3657 }
3659 return res;
3660 }
3662 size_t os::large_page_size() {
3663 return _large_page_size;
3664 }
3666 // With SysV SHM the entire memory region must be allocated as shared
3667 // memory.
3668 // HugeTLBFS allows application to commit large page memory on demand.
3669 // However, when committing memory with HugeTLBFS fails, the region
3670 // that was supposed to be committed will lose the old reservation
3671 // and allow other threads to steal that memory region. Because of this
3672 // behavior we can't commit HugeTLBFS memory.
3673 bool os::can_commit_large_page_memory() {
3674 return UseTransparentHugePages;
3675 }
3677 bool os::can_execute_large_page_memory() {
3678 return UseTransparentHugePages || UseHugeTLBFS;
3679 }
3681 // Reserve memory at an arbitrary address, only if that area is
3682 // available (and not reserved for something else).
3684 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3685 const int max_tries = 10;
3686 char* base[max_tries];
3687 size_t size[max_tries];
3688 const size_t gap = 0x000000;
3690 // Assert only that the size is a multiple of the page size, since
3691 // that's all that mmap requires, and since that's all we really know
3692 // about at this low abstraction level. If we need higher alignment,
3693 // we can either pass an alignment to this method or verify alignment
3694 // in one of the methods further up the call chain. See bug 5044738.
3695 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3697 // Repeatedly allocate blocks until the block is allocated at the
3698 // right spot. Give up after max_tries. Note that reserve_memory() will
3699 // automatically update _highest_vm_reserved_address if the call is
3700 // successful. The variable tracks the highest memory address every reserved
3701 // by JVM. It is used to detect heap-stack collision if running with
3702 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3703 // space than needed, it could confuse the collision detecting code. To
3704 // solve the problem, save current _highest_vm_reserved_address and
3705 // calculate the correct value before return.
3706 address old_highest = _highest_vm_reserved_address;
3708 // Linux mmap allows caller to pass an address as hint; give it a try first,
3709 // if kernel honors the hint then we can return immediately.
3710 char * addr = anon_mmap(requested_addr, bytes, false);
3711 if (addr == requested_addr) {
3712 return requested_addr;
3713 }
3715 if (addr != NULL) {
3716 // mmap() is successful but it fails to reserve at the requested address
3717 anon_munmap(addr, bytes);
3718 }
3720 int i;
3721 for (i = 0; i < max_tries; ++i) {
3722 base[i] = reserve_memory(bytes);
3724 if (base[i] != NULL) {
3725 // Is this the block we wanted?
3726 if (base[i] == requested_addr) {
3727 size[i] = bytes;
3728 break;
3729 }
3731 // Does this overlap the block we wanted? Give back the overlapped
3732 // parts and try again.
3734 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3735 if (top_overlap >= 0 && top_overlap < bytes) {
3736 unmap_memory(base[i], top_overlap);
3737 base[i] += top_overlap;
3738 size[i] = bytes - top_overlap;
3739 } else {
3740 size_t bottom_overlap = base[i] + bytes - requested_addr;
3741 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3742 unmap_memory(requested_addr, bottom_overlap);
3743 size[i] = bytes - bottom_overlap;
3744 } else {
3745 size[i] = bytes;
3746 }
3747 }
3748 }
3749 }
3751 // Give back the unused reserved pieces.
3753 for (int j = 0; j < i; ++j) {
3754 if (base[j] != NULL) {
3755 unmap_memory(base[j], size[j]);
3756 }
3757 }
3759 if (i < max_tries) {
3760 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3761 return requested_addr;
3762 } else {
3763 _highest_vm_reserved_address = old_highest;
3764 return NULL;
3765 }
3766 }
3768 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3769 return ::read(fd, buf, nBytes);
3770 }
3772 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3773 // Solaris uses poll(), linux uses park().
3774 // Poll() is likely a better choice, assuming that Thread.interrupt()
3775 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3776 // SIGSEGV, see 4355769.
3778 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3779 assert(thread == Thread::current(), "thread consistency check");
3781 ParkEvent * const slp = thread->_SleepEvent ;
3782 slp->reset() ;
3783 OrderAccess::fence() ;
3785 if (interruptible) {
3786 jlong prevtime = javaTimeNanos();
3788 for (;;) {
3789 if (os::is_interrupted(thread, true)) {
3790 return OS_INTRPT;
3791 }
3793 jlong newtime = javaTimeNanos();
3795 if (newtime - prevtime < 0) {
3796 // time moving backwards, should only happen if no monotonic clock
3797 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3798 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3799 } else {
3800 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3801 }
3803 if(millis <= 0) {
3804 return OS_OK;
3805 }
3807 prevtime = newtime;
3809 {
3810 assert(thread->is_Java_thread(), "sanity check");
3811 JavaThread *jt = (JavaThread *) thread;
3812 ThreadBlockInVM tbivm(jt);
3813 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3815 jt->set_suspend_equivalent();
3816 // cleared by handle_special_suspend_equivalent_condition() or
3817 // java_suspend_self() via check_and_wait_while_suspended()
3819 slp->park(millis);
3821 // were we externally suspended while we were waiting?
3822 jt->check_and_wait_while_suspended();
3823 }
3824 }
3825 } else {
3826 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3827 jlong prevtime = javaTimeNanos();
3829 for (;;) {
3830 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3831 // the 1st iteration ...
3832 jlong newtime = javaTimeNanos();
3834 if (newtime - prevtime < 0) {
3835 // time moving backwards, should only happen if no monotonic clock
3836 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3837 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3838 } else {
3839 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3840 }
3842 if(millis <= 0) break ;
3844 prevtime = newtime;
3845 slp->park(millis);
3846 }
3847 return OS_OK ;
3848 }
3849 }
3851 //
3852 // Short sleep, direct OS call.
3853 //
3854 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3855 // sched_yield(2) will actually give up the CPU:
3856 //
3857 // * Alone on this pariticular CPU, keeps running.
3858 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3859 // (pre 2.6.39).
3860 //
3861 // So calling this with 0 is an alternative.
3862 //
3863 void os::naked_short_sleep(jlong ms) {
3864 struct timespec req;
3866 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3867 req.tv_sec = 0;
3868 if (ms > 0) {
3869 req.tv_nsec = (ms % 1000) * 1000000;
3870 }
3871 else {
3872 req.tv_nsec = 1;
3873 }
3875 nanosleep(&req, NULL);
3877 return;
3878 }
3880 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3881 void os::infinite_sleep() {
3882 while (true) { // sleep forever ...
3883 ::sleep(100); // ... 100 seconds at a time
3884 }
3885 }
3887 // Used to convert frequent JVM_Yield() to nops
3888 bool os::dont_yield() {
3889 return DontYieldALot;
3890 }
3892 void os::yield() {
3893 sched_yield();
3894 }
3896 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3898 void os::yield_all(int attempts) {
3899 // Yields to all threads, including threads with lower priorities
3900 // Threads on Linux are all with same priority. The Solaris style
3901 // os::yield_all() with nanosleep(1ms) is not necessary.
3902 sched_yield();
3903 }
3905 // Called from the tight loops to possibly influence time-sharing heuristics
3906 void os::loop_breaker(int attempts) {
3907 os::yield_all(attempts);
3908 }
3910 ////////////////////////////////////////////////////////////////////////////////
3911 // thread priority support
3913 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3914 // only supports dynamic priority, static priority must be zero. For real-time
3915 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3916 // However, for large multi-threaded applications, SCHED_RR is not only slower
3917 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3918 // of 5 runs - Sep 2005).
3919 //
3920 // The following code actually changes the niceness of kernel-thread/LWP. It
3921 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3922 // not the entire user process, and user level threads are 1:1 mapped to kernel
3923 // threads. It has always been the case, but could change in the future. For
3924 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3925 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3927 int os::java_to_os_priority[CriticalPriority + 1] = {
3928 19, // 0 Entry should never be used
3930 4, // 1 MinPriority
3931 3, // 2
3932 2, // 3
3934 1, // 4
3935 0, // 5 NormPriority
3936 -1, // 6
3938 -2, // 7
3939 -3, // 8
3940 -4, // 9 NearMaxPriority
3942 -5, // 10 MaxPriority
3944 -5 // 11 CriticalPriority
3945 };
3947 static int prio_init() {
3948 if (ThreadPriorityPolicy == 1) {
3949 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3950 // if effective uid is not root. Perhaps, a more elegant way of doing
3951 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3952 if (geteuid() != 0) {
3953 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3954 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3955 }
3956 ThreadPriorityPolicy = 0;
3957 }
3958 }
3959 if (UseCriticalJavaThreadPriority) {
3960 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3961 }
3962 return 0;
3963 }
3965 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3966 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3968 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3969 return (ret == 0) ? OS_OK : OS_ERR;
3970 }
3972 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3973 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3974 *priority_ptr = java_to_os_priority[NormPriority];
3975 return OS_OK;
3976 }
3978 errno = 0;
3979 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3980 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3981 }
3983 // Hint to the underlying OS that a task switch would not be good.
3984 // Void return because it's a hint and can fail.
3985 void os::hint_no_preempt() {}
3987 ////////////////////////////////////////////////////////////////////////////////
3988 // suspend/resume support
3990 // the low-level signal-based suspend/resume support is a remnant from the
3991 // old VM-suspension that used to be for java-suspension, safepoints etc,
3992 // within hotspot. Now there is a single use-case for this:
3993 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3994 // that runs in the watcher thread.
3995 // The remaining code is greatly simplified from the more general suspension
3996 // code that used to be used.
3997 //
3998 // The protocol is quite simple:
3999 // - suspend:
4000 // - sends a signal to the target thread
4001 // - polls the suspend state of the osthread using a yield loop
4002 // - target thread signal handler (SR_handler) sets suspend state
4003 // and blocks in sigsuspend until continued
4004 // - resume:
4005 // - sets target osthread state to continue
4006 // - sends signal to end the sigsuspend loop in the SR_handler
4007 //
4008 // Note that the SR_lock plays no role in this suspend/resume protocol.
4009 //
4011 static void resume_clear_context(OSThread *osthread) {
4012 osthread->set_ucontext(NULL);
4013 osthread->set_siginfo(NULL);
4014 }
4016 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4017 osthread->set_ucontext(context);
4018 osthread->set_siginfo(siginfo);
4019 }
4021 //
4022 // Handler function invoked when a thread's execution is suspended or
4023 // resumed. We have to be careful that only async-safe functions are
4024 // called here (Note: most pthread functions are not async safe and
4025 // should be avoided.)
4026 //
4027 // Note: sigwait() is a more natural fit than sigsuspend() from an
4028 // interface point of view, but sigwait() prevents the signal hander
4029 // from being run. libpthread would get very confused by not having
4030 // its signal handlers run and prevents sigwait()'s use with the
4031 // mutex granting granting signal.
4032 //
4033 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4034 //
4035 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4036 // Save and restore errno to avoid confusing native code with EINTR
4037 // after sigsuspend.
4038 int old_errno = errno;
4040 Thread* thread = Thread::current();
4041 OSThread* osthread = thread->osthread();
4042 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4044 os::SuspendResume::State current = osthread->sr.state();
4045 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4046 suspend_save_context(osthread, siginfo, context);
4048 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4049 os::SuspendResume::State state = osthread->sr.suspended();
4050 if (state == os::SuspendResume::SR_SUSPENDED) {
4051 sigset_t suspend_set; // signals for sigsuspend()
4053 // get current set of blocked signals and unblock resume signal
4054 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4055 sigdelset(&suspend_set, SR_signum);
4057 sr_semaphore.signal();
4058 // wait here until we are resumed
4059 while (1) {
4060 sigsuspend(&suspend_set);
4062 os::SuspendResume::State result = osthread->sr.running();
4063 if (result == os::SuspendResume::SR_RUNNING) {
4064 sr_semaphore.signal();
4065 break;
4066 }
4067 }
4069 } else if (state == os::SuspendResume::SR_RUNNING) {
4070 // request was cancelled, continue
4071 } else {
4072 ShouldNotReachHere();
4073 }
4075 resume_clear_context(osthread);
4076 } else if (current == os::SuspendResume::SR_RUNNING) {
4077 // request was cancelled, continue
4078 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4079 // ignore
4080 } else {
4081 // ignore
4082 }
4084 errno = old_errno;
4085 }
4088 static int SR_initialize() {
4089 struct sigaction act;
4090 char *s;
4091 /* Get signal number to use for suspend/resume */
4092 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4093 int sig = ::strtol(s, 0, 10);
4094 if (sig > 0 || sig < _NSIG) {
4095 SR_signum = sig;
4096 }
4097 }
4099 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4100 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4102 sigemptyset(&SR_sigset);
4103 sigaddset(&SR_sigset, SR_signum);
4105 /* Set up signal handler for suspend/resume */
4106 act.sa_flags = SA_RESTART|SA_SIGINFO;
4107 act.sa_handler = (void (*)(int)) SR_handler;
4109 // SR_signum is blocked by default.
4110 // 4528190 - We also need to block pthread restart signal (32 on all
4111 // supported Linux platforms). Note that LinuxThreads need to block
4112 // this signal for all threads to work properly. So we don't have
4113 // to use hard-coded signal number when setting up the mask.
4114 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4116 if (sigaction(SR_signum, &act, 0) == -1) {
4117 return -1;
4118 }
4120 // Save signal flag
4121 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4122 return 0;
4123 }
4125 static int sr_notify(OSThread* osthread) {
4126 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4127 assert_status(status == 0, status, "pthread_kill");
4128 return status;
4129 }
4131 // "Randomly" selected value for how long we want to spin
4132 // before bailing out on suspending a thread, also how often
4133 // we send a signal to a thread we want to resume
4134 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4135 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4137 // returns true on success and false on error - really an error is fatal
4138 // but this seems the normal response to library errors
4139 static bool do_suspend(OSThread* osthread) {
4140 assert(osthread->sr.is_running(), "thread should be running");
4141 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4143 // mark as suspended and send signal
4144 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4145 // failed to switch, state wasn't running?
4146 ShouldNotReachHere();
4147 return false;
4148 }
4150 if (sr_notify(osthread) != 0) {
4151 ShouldNotReachHere();
4152 }
4154 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4155 while (true) {
4156 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4157 break;
4158 } else {
4159 // timeout
4160 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4161 if (cancelled == os::SuspendResume::SR_RUNNING) {
4162 return false;
4163 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4164 // make sure that we consume the signal on the semaphore as well
4165 sr_semaphore.wait();
4166 break;
4167 } else {
4168 ShouldNotReachHere();
4169 return false;
4170 }
4171 }
4172 }
4174 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4175 return true;
4176 }
4178 static void do_resume(OSThread* osthread) {
4179 assert(osthread->sr.is_suspended(), "thread should be suspended");
4180 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4182 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4183 // failed to switch to WAKEUP_REQUEST
4184 ShouldNotReachHere();
4185 return;
4186 }
4188 while (true) {
4189 if (sr_notify(osthread) == 0) {
4190 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4191 if (osthread->sr.is_running()) {
4192 return;
4193 }
4194 }
4195 } else {
4196 ShouldNotReachHere();
4197 }
4198 }
4200 guarantee(osthread->sr.is_running(), "Must be running!");
4201 }
4203 ////////////////////////////////////////////////////////////////////////////////
4204 // interrupt support
4206 void os::interrupt(Thread* thread) {
4207 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4208 "possibility of dangling Thread pointer");
4210 OSThread* osthread = thread->osthread();
4212 if (!osthread->interrupted()) {
4213 osthread->set_interrupted(true);
4214 // More than one thread can get here with the same value of osthread,
4215 // resulting in multiple notifications. We do, however, want the store
4216 // to interrupted() to be visible to other threads before we execute unpark().
4217 OrderAccess::fence();
4218 ParkEvent * const slp = thread->_SleepEvent ;
4219 if (slp != NULL) slp->unpark() ;
4220 }
4222 // For JSR166. Unpark even if interrupt status already was set
4223 if (thread->is_Java_thread())
4224 ((JavaThread*)thread)->parker()->unpark();
4226 ParkEvent * ev = thread->_ParkEvent ;
4227 if (ev != NULL) ev->unpark() ;
4229 }
4231 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4232 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4233 "possibility of dangling Thread pointer");
4235 OSThread* osthread = thread->osthread();
4237 bool interrupted = osthread->interrupted();
4239 if (interrupted && clear_interrupted) {
4240 osthread->set_interrupted(false);
4241 // consider thread->_SleepEvent->reset() ... optional optimization
4242 }
4244 return interrupted;
4245 }
4247 ///////////////////////////////////////////////////////////////////////////////////
4248 // signal handling (except suspend/resume)
4250 // This routine may be used by user applications as a "hook" to catch signals.
4251 // The user-defined signal handler must pass unrecognized signals to this
4252 // routine, and if it returns true (non-zero), then the signal handler must
4253 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4254 // routine will never retun false (zero), but instead will execute a VM panic
4255 // routine kill the process.
4256 //
4257 // If this routine returns false, it is OK to call it again. This allows
4258 // the user-defined signal handler to perform checks either before or after
4259 // the VM performs its own checks. Naturally, the user code would be making
4260 // a serious error if it tried to handle an exception (such as a null check
4261 // or breakpoint) that the VM was generating for its own correct operation.
4262 //
4263 // This routine may recognize any of the following kinds of signals:
4264 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4265 // It should be consulted by handlers for any of those signals.
4266 //
4267 // The caller of this routine must pass in the three arguments supplied
4268 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4269 // field of the structure passed to sigaction(). This routine assumes that
4270 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4271 //
4272 // Note that the VM will print warnings if it detects conflicting signal
4273 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4274 //
4275 extern "C" JNIEXPORT int
4276 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4277 void* ucontext, int abort_if_unrecognized);
4279 void signalHandler(int sig, siginfo_t* info, void* uc) {
4280 assert(info != NULL && uc != NULL, "it must be old kernel");
4281 int orig_errno = errno; // Preserve errno value over signal handler.
4282 JVM_handle_linux_signal(sig, info, uc, true);
4283 errno = orig_errno;
4284 }
4287 // This boolean allows users to forward their own non-matching signals
4288 // to JVM_handle_linux_signal, harmlessly.
4289 bool os::Linux::signal_handlers_are_installed = false;
4291 // For signal-chaining
4292 struct sigaction os::Linux::sigact[MAXSIGNUM];
4293 unsigned int os::Linux::sigs = 0;
4294 bool os::Linux::libjsig_is_loaded = false;
4295 typedef struct sigaction *(*get_signal_t)(int);
4296 get_signal_t os::Linux::get_signal_action = NULL;
4298 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4299 struct sigaction *actp = NULL;
4301 if (libjsig_is_loaded) {
4302 // Retrieve the old signal handler from libjsig
4303 actp = (*get_signal_action)(sig);
4304 }
4305 if (actp == NULL) {
4306 // Retrieve the preinstalled signal handler from jvm
4307 actp = get_preinstalled_handler(sig);
4308 }
4310 return actp;
4311 }
4313 static bool call_chained_handler(struct sigaction *actp, int sig,
4314 siginfo_t *siginfo, void *context) {
4315 // Call the old signal handler
4316 if (actp->sa_handler == SIG_DFL) {
4317 // It's more reasonable to let jvm treat it as an unexpected exception
4318 // instead of taking the default action.
4319 return false;
4320 } else if (actp->sa_handler != SIG_IGN) {
4321 if ((actp->sa_flags & SA_NODEFER) == 0) {
4322 // automaticlly block the signal
4323 sigaddset(&(actp->sa_mask), sig);
4324 }
4326 sa_handler_t hand;
4327 sa_sigaction_t sa;
4328 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4329 // retrieve the chained handler
4330 if (siginfo_flag_set) {
4331 sa = actp->sa_sigaction;
4332 } else {
4333 hand = actp->sa_handler;
4334 }
4336 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4337 actp->sa_handler = SIG_DFL;
4338 }
4340 // try to honor the signal mask
4341 sigset_t oset;
4342 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4344 // call into the chained handler
4345 if (siginfo_flag_set) {
4346 (*sa)(sig, siginfo, context);
4347 } else {
4348 (*hand)(sig);
4349 }
4351 // restore the signal mask
4352 pthread_sigmask(SIG_SETMASK, &oset, 0);
4353 }
4354 // Tell jvm's signal handler the signal is taken care of.
4355 return true;
4356 }
4358 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4359 bool chained = false;
4360 // signal-chaining
4361 if (UseSignalChaining) {
4362 struct sigaction *actp = get_chained_signal_action(sig);
4363 if (actp != NULL) {
4364 chained = call_chained_handler(actp, sig, siginfo, context);
4365 }
4366 }
4367 return chained;
4368 }
4370 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4371 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4372 return &sigact[sig];
4373 }
4374 return NULL;
4375 }
4377 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4378 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4379 sigact[sig] = oldAct;
4380 sigs |= (unsigned int)1 << sig;
4381 }
4383 // for diagnostic
4384 int os::Linux::sigflags[MAXSIGNUM];
4386 int os::Linux::get_our_sigflags(int sig) {
4387 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4388 return sigflags[sig];
4389 }
4391 void os::Linux::set_our_sigflags(int sig, int flags) {
4392 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4393 sigflags[sig] = flags;
4394 }
4396 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4397 // Check for overwrite.
4398 struct sigaction oldAct;
4399 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4401 void* oldhand = oldAct.sa_sigaction
4402 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4403 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4404 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4405 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4406 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4407 if (AllowUserSignalHandlers || !set_installed) {
4408 // Do not overwrite; user takes responsibility to forward to us.
4409 return;
4410 } else if (UseSignalChaining) {
4411 // save the old handler in jvm
4412 save_preinstalled_handler(sig, oldAct);
4413 // libjsig also interposes the sigaction() call below and saves the
4414 // old sigaction on it own.
4415 } else {
4416 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4417 "%#lx for signal %d.", (long)oldhand, sig));
4418 }
4419 }
4421 struct sigaction sigAct;
4422 sigfillset(&(sigAct.sa_mask));
4423 sigAct.sa_handler = SIG_DFL;
4424 if (!set_installed) {
4425 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4426 } else {
4427 sigAct.sa_sigaction = signalHandler;
4428 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4429 }
4430 // Save flags, which are set by ours
4431 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4432 sigflags[sig] = sigAct.sa_flags;
4434 int ret = sigaction(sig, &sigAct, &oldAct);
4435 assert(ret == 0, "check");
4437 void* oldhand2 = oldAct.sa_sigaction
4438 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4439 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4440 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4441 }
4443 // install signal handlers for signals that HotSpot needs to
4444 // handle in order to support Java-level exception handling.
4446 void os::Linux::install_signal_handlers() {
4447 if (!signal_handlers_are_installed) {
4448 signal_handlers_are_installed = true;
4450 // signal-chaining
4451 typedef void (*signal_setting_t)();
4452 signal_setting_t begin_signal_setting = NULL;
4453 signal_setting_t end_signal_setting = NULL;
4454 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4455 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4456 if (begin_signal_setting != NULL) {
4457 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4458 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4459 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4460 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4461 libjsig_is_loaded = true;
4462 assert(UseSignalChaining, "should enable signal-chaining");
4463 }
4464 if (libjsig_is_loaded) {
4465 // Tell libjsig jvm is setting signal handlers
4466 (*begin_signal_setting)();
4467 }
4469 set_signal_handler(SIGSEGV, true);
4470 set_signal_handler(SIGPIPE, true);
4471 set_signal_handler(SIGBUS, true);
4472 set_signal_handler(SIGILL, true);
4473 set_signal_handler(SIGFPE, true);
4474 #if defined(PPC64)
4475 set_signal_handler(SIGTRAP, true);
4476 #endif
4477 set_signal_handler(SIGXFSZ, true);
4479 if (libjsig_is_loaded) {
4480 // Tell libjsig jvm finishes setting signal handlers
4481 (*end_signal_setting)();
4482 }
4484 // We don't activate signal checker if libjsig is in place, we trust ourselves
4485 // and if UserSignalHandler is installed all bets are off.
4486 // Log that signal checking is off only if -verbose:jni is specified.
4487 if (CheckJNICalls) {
4488 if (libjsig_is_loaded) {
4489 if (PrintJNIResolving) {
4490 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4491 }
4492 check_signals = false;
4493 }
4494 if (AllowUserSignalHandlers) {
4495 if (PrintJNIResolving) {
4496 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4497 }
4498 check_signals = false;
4499 }
4500 }
4501 }
4502 }
4504 // This is the fastest way to get thread cpu time on Linux.
4505 // Returns cpu time (user+sys) for any thread, not only for current.
4506 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4507 // It might work on 2.6.10+ with a special kernel/glibc patch.
4508 // For reference, please, see IEEE Std 1003.1-2004:
4509 // http://www.unix.org/single_unix_specification
4511 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4512 struct timespec tp;
4513 int rc = os::Linux::clock_gettime(clockid, &tp);
4514 assert(rc == 0, "clock_gettime is expected to return 0 code");
4516 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4517 }
4519 /////
4520 // glibc on Linux platform uses non-documented flag
4521 // to indicate, that some special sort of signal
4522 // trampoline is used.
4523 // We will never set this flag, and we should
4524 // ignore this flag in our diagnostic
4525 #ifdef SIGNIFICANT_SIGNAL_MASK
4526 #undef SIGNIFICANT_SIGNAL_MASK
4527 #endif
4528 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4530 static const char* get_signal_handler_name(address handler,
4531 char* buf, int buflen) {
4532 int offset;
4533 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4534 if (found) {
4535 // skip directory names
4536 const char *p1, *p2;
4537 p1 = buf;
4538 size_t len = strlen(os::file_separator());
4539 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4540 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4541 } else {
4542 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4543 }
4544 return buf;
4545 }
4547 static void print_signal_handler(outputStream* st, int sig,
4548 char* buf, size_t buflen) {
4549 struct sigaction sa;
4551 sigaction(sig, NULL, &sa);
4553 // See comment for SIGNIFICANT_SIGNAL_MASK define
4554 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4556 st->print("%s: ", os::exception_name(sig, buf, buflen));
4558 address handler = (sa.sa_flags & SA_SIGINFO)
4559 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4560 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4562 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4563 st->print("SIG_DFL");
4564 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4565 st->print("SIG_IGN");
4566 } else {
4567 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4568 }
4570 st->print(", sa_mask[0]=");
4571 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4573 address rh = VMError::get_resetted_sighandler(sig);
4574 // May be, handler was resetted by VMError?
4575 if(rh != NULL) {
4576 handler = rh;
4577 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4578 }
4580 st->print(", sa_flags=");
4581 os::Posix::print_sa_flags(st, sa.sa_flags);
4583 // Check: is it our handler?
4584 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4585 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4586 // It is our signal handler
4587 // check for flags, reset system-used one!
4588 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4589 st->print(
4590 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4591 os::Linux::get_our_sigflags(sig));
4592 }
4593 }
4594 st->cr();
4595 }
4598 #define DO_SIGNAL_CHECK(sig) \
4599 if (!sigismember(&check_signal_done, sig)) \
4600 os::Linux::check_signal_handler(sig)
4602 // This method is a periodic task to check for misbehaving JNI applications
4603 // under CheckJNI, we can add any periodic checks here
4605 void os::run_periodic_checks() {
4607 if (check_signals == false) return;
4609 // SEGV and BUS if overridden could potentially prevent
4610 // generation of hs*.log in the event of a crash, debugging
4611 // such a case can be very challenging, so we absolutely
4612 // check the following for a good measure:
4613 DO_SIGNAL_CHECK(SIGSEGV);
4614 DO_SIGNAL_CHECK(SIGILL);
4615 DO_SIGNAL_CHECK(SIGFPE);
4616 DO_SIGNAL_CHECK(SIGBUS);
4617 DO_SIGNAL_CHECK(SIGPIPE);
4618 DO_SIGNAL_CHECK(SIGXFSZ);
4619 #if defined(PPC64)
4620 DO_SIGNAL_CHECK(SIGTRAP);
4621 #endif
4623 // ReduceSignalUsage allows the user to override these handlers
4624 // see comments at the very top and jvm_solaris.h
4625 if (!ReduceSignalUsage) {
4626 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4627 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4628 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4629 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4630 }
4632 DO_SIGNAL_CHECK(SR_signum);
4633 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4634 }
4636 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4638 static os_sigaction_t os_sigaction = NULL;
4640 void os::Linux::check_signal_handler(int sig) {
4641 char buf[O_BUFLEN];
4642 address jvmHandler = NULL;
4645 struct sigaction act;
4646 if (os_sigaction == NULL) {
4647 // only trust the default sigaction, in case it has been interposed
4648 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4649 if (os_sigaction == NULL) return;
4650 }
4652 os_sigaction(sig, (struct sigaction*)NULL, &act);
4655 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4657 address thisHandler = (act.sa_flags & SA_SIGINFO)
4658 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4659 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4662 switch(sig) {
4663 case SIGSEGV:
4664 case SIGBUS:
4665 case SIGFPE:
4666 case SIGPIPE:
4667 case SIGILL:
4668 case SIGXFSZ:
4669 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4670 break;
4672 case SHUTDOWN1_SIGNAL:
4673 case SHUTDOWN2_SIGNAL:
4674 case SHUTDOWN3_SIGNAL:
4675 case BREAK_SIGNAL:
4676 jvmHandler = (address)user_handler();
4677 break;
4679 case INTERRUPT_SIGNAL:
4680 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4681 break;
4683 default:
4684 if (sig == SR_signum) {
4685 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4686 } else {
4687 return;
4688 }
4689 break;
4690 }
4692 if (thisHandler != jvmHandler) {
4693 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4694 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4695 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4696 // No need to check this sig any longer
4697 sigaddset(&check_signal_done, sig);
4698 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4699 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4700 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4701 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4702 // No need to check this sig any longer
4703 sigaddset(&check_signal_done, sig);
4704 }
4706 // Dump all the signal
4707 if (sigismember(&check_signal_done, sig)) {
4708 print_signal_handlers(tty, buf, O_BUFLEN);
4709 }
4710 }
4712 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4714 extern bool signal_name(int signo, char* buf, size_t len);
4716 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4717 if (0 < exception_code && exception_code <= SIGRTMAX) {
4718 // signal
4719 if (!signal_name(exception_code, buf, size)) {
4720 jio_snprintf(buf, size, "SIG%d", exception_code);
4721 }
4722 return buf;
4723 } else {
4724 return NULL;
4725 }
4726 }
4728 // this is called _before_ the most of global arguments have been parsed
4729 void os::init(void) {
4730 char dummy; /* used to get a guess on initial stack address */
4731 // first_hrtime = gethrtime();
4733 // With LinuxThreads the JavaMain thread pid (primordial thread)
4734 // is different than the pid of the java launcher thread.
4735 // So, on Linux, the launcher thread pid is passed to the VM
4736 // via the sun.java.launcher.pid property.
4737 // Use this property instead of getpid() if it was correctly passed.
4738 // See bug 6351349.
4739 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4741 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4743 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4745 init_random(1234567);
4747 ThreadCritical::initialize();
4749 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4750 if (Linux::page_size() == -1) {
4751 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4752 strerror(errno)));
4753 }
4754 init_page_sizes((size_t) Linux::page_size());
4756 Linux::initialize_system_info();
4758 // main_thread points to the aboriginal thread
4759 Linux::_main_thread = pthread_self();
4761 Linux::clock_init();
4762 initial_time_count = javaTimeNanos();
4764 // pthread_condattr initialization for monotonic clock
4765 int status;
4766 pthread_condattr_t* _condattr = os::Linux::condAttr();
4767 if ((status = pthread_condattr_init(_condattr)) != 0) {
4768 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4769 }
4770 // Only set the clock if CLOCK_MONOTONIC is available
4771 if (Linux::supports_monotonic_clock()) {
4772 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4773 if (status == EINVAL) {
4774 warning("Unable to use monotonic clock with relative timed-waits" \
4775 " - changes to the time-of-day clock may have adverse affects");
4776 } else {
4777 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4778 }
4779 }
4780 }
4781 // else it defaults to CLOCK_REALTIME
4783 pthread_mutex_init(&dl_mutex, NULL);
4785 // If the pagesize of the VM is greater than 8K determine the appropriate
4786 // number of initial guard pages. The user can change this with the
4787 // command line arguments, if needed.
4788 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4789 StackYellowPages = 1;
4790 StackRedPages = 1;
4791 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4792 }
4793 }
4795 // To install functions for atexit system call
4796 extern "C" {
4797 static void perfMemory_exit_helper() {
4798 perfMemory_exit();
4799 }
4800 }
4802 // this is called _after_ the global arguments have been parsed
4803 jint os::init_2(void)
4804 {
4805 Linux::fast_thread_clock_init();
4807 // Allocate a single page and mark it as readable for safepoint polling
4808 #ifdef OPT_SAFEPOINT
4809 /* 2013/10/25 Jin: get polling_page from with 32-bit memory space.
4810 In our Loongson-3A Redhat-64 OS, the JDK's virtual memory space starts from 0x120000000.
4811 So we can select any address below 32-bit.
4812 */
4813 void * p = (void *)0x400000;
4814 address polling_page = (address) ::mmap(p, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4815 #else
4816 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4817 #endif
4819 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4821 os::set_polling_page( polling_page );
4823 #ifndef PRODUCT
4824 if(Verbose && PrintMiscellaneous)
4825 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4826 #endif
4828 if (!UseMembar) {
4829 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4830 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4831 os::set_memory_serialize_page( mem_serialize_page );
4833 #ifndef PRODUCT
4834 if(Verbose && PrintMiscellaneous)
4835 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4836 #endif
4837 }
4839 // initialize suspend/resume support - must do this before signal_sets_init()
4840 if (SR_initialize() != 0) {
4841 perror("SR_initialize failed");
4842 return JNI_ERR;
4843 }
4845 Linux::signal_sets_init();
4846 Linux::install_signal_handlers();
4848 // Check minimum allowable stack size for thread creation and to initialize
4849 // the java system classes, including StackOverflowError - depends on page
4850 // size. Add a page for compiler2 recursion in main thread.
4851 // Add in 2*BytesPerWord times page size to account for VM stack during
4852 // class initialization depending on 32 or 64 bit VM.
4854 /* 2014/1/2 Liao: JDK8 requires larger -Xss option.
4855 * TongWeb cannot run with -Xss192K.
4856 * We are not sure whether this causes errors, so simply print a warning. */
4857 size_t min_stack_allowed_jdk6 = os::Linux::min_stack_allowed;
4858 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4859 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4860 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4862 size_t threadStackSizeInBytes = ThreadStackSize * K;
4863 if (threadStackSizeInBytes != 0 &&
4864 threadStackSizeInBytes < min_stack_allowed_jdk6) {
4865 tty->print_cr("\nThe stack size specified is too small, "
4866 "Specify at least %dk",
4867 min_stack_allowed_jdk6/ K);
4868 return JNI_ERR;
4869 }
4871 if (threadStackSizeInBytes != 0 &&
4872 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4873 warning("JDK8 requires the stack size be at least %dk, the current value is %dk. "
4874 "Resize -Xss for best performance.", os::Linux::min_stack_allowed/ K, ThreadStackSize);
4875 }
4877 // Make the stack size a multiple of the page size so that
4878 // the yellow/red zones can be guarded.
4879 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4880 vm_page_size()));
4882 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4884 #if defined(IA32)
4885 workaround_expand_exec_shield_cs_limit();
4886 #endif
4888 Linux::libpthread_init();
4889 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4890 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4891 Linux::glibc_version(), Linux::libpthread_version(),
4892 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4893 }
4895 if (UseNUMA) {
4896 if (!Linux::libnuma_init()) {
4897 UseNUMA = false;
4898 } else {
4899 if ((Linux::numa_max_node() < 1)) {
4900 // There's only one node(they start from 0), disable NUMA.
4901 UseNUMA = false;
4902 }
4903 }
4904 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4905 // we can make the adaptive lgrp chunk resizing work. If the user specified
4906 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4907 // disable adaptive resizing.
4908 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4909 if (FLAG_IS_DEFAULT(UseNUMA)) {
4910 UseNUMA = false;
4911 } else {
4912 if (FLAG_IS_DEFAULT(UseLargePages) &&
4913 FLAG_IS_DEFAULT(UseSHM) &&
4914 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4915 UseLargePages = false;
4916 } else {
4917 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4918 UseAdaptiveSizePolicy = false;
4919 UseAdaptiveNUMAChunkSizing = false;
4920 }
4921 }
4922 }
4923 if (!UseNUMA && ForceNUMA) {
4924 UseNUMA = true;
4925 }
4926 }
4928 /*Liao:2013/11/18 UseOldNUMA: Let OldGen support NUMA*/
4929 if(UseNUMA == true && UseOldNUMA == true) {
4930 UseOldNUMA = true;
4931 } else {
4932 UseOldNUMA = false;
4933 }
4935 if(UseOldNUMA == false) {
4936 UseNUMAGC = false;
4937 UseNUMAThreadRoots = false;
4938 UseNUMASteal = false;
4939 BindGCTaskThreadsToCPUs = false;
4940 }
4942 if (MaxFDLimit) {
4943 // set the number of file descriptors to max. print out error
4944 // if getrlimit/setrlimit fails but continue regardless.
4945 struct rlimit nbr_files;
4946 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4947 if (status != 0) {
4948 if (PrintMiscellaneous && (Verbose || WizardMode))
4949 perror("os::init_2 getrlimit failed");
4950 } else {
4951 nbr_files.rlim_cur = nbr_files.rlim_max;
4952 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4953 if (status != 0) {
4954 if (PrintMiscellaneous && (Verbose || WizardMode))
4955 perror("os::init_2 setrlimit failed");
4956 }
4957 }
4958 }
4960 // Initialize lock used to serialize thread creation (see os::create_thread)
4961 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4963 // at-exit methods are called in the reverse order of their registration.
4964 // atexit functions are called on return from main or as a result of a
4965 // call to exit(3C). There can be only 32 of these functions registered
4966 // and atexit() does not set errno.
4968 if (PerfAllowAtExitRegistration) {
4969 // only register atexit functions if PerfAllowAtExitRegistration is set.
4970 // atexit functions can be delayed until process exit time, which
4971 // can be problematic for embedded VM situations. Embedded VMs should
4972 // call DestroyJavaVM() to assure that VM resources are released.
4974 // note: perfMemory_exit_helper atexit function may be removed in
4975 // the future if the appropriate cleanup code can be added to the
4976 // VM_Exit VMOperation's doit method.
4977 if (atexit(perfMemory_exit_helper) != 0) {
4978 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4979 }
4980 }
4982 // initialize thread priority policy
4983 prio_init();
4985 return JNI_OK;
4986 }
4988 // this is called at the end of vm_initialization
4989 void os::init_3(void) {
4990 #ifdef JAVASE_EMBEDDED
4991 // Start the MemNotifyThread
4992 if (LowMemoryProtection) {
4993 MemNotifyThread::start();
4994 }
4995 return;
4996 #endif
4997 }
4999 // Mark the polling page as unreadable
5000 void os::make_polling_page_unreadable(void) {
5001 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5002 fatal("Could not disable polling page");
5003 };
5005 // Mark the polling page as readable
5006 void os::make_polling_page_readable(void) {
5007 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5008 fatal("Could not enable polling page");
5009 }
5010 };
5012 int os::active_processor_count() {
5013 // Linux doesn't yet have a (official) notion of processor sets,
5014 // so just return the number of online processors.
5015 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5016 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
5017 return online_cpus;
5018 }
5020 void os::set_native_thread_name(const char *name) {
5021 // Not yet implemented.
5022 return;
5023 }
5025 bool os::distribute_processes(uint length, uint* distribution) {
5026 // Not yet implemented.
5027 return false;
5028 }
5030 bool os::bind_to_processor(uint processor_id) {
5031 /* 2014/7/7 implemented by Liao */
5032 if(BindGCTaskThreadsToCPUs) {
5033 cpu_set_t mask;
5034 cpu_set_t get;
5035 CPU_ZERO(&mask);
5036 CPU_SET(processor_id, &mask);
5038 if(sched_setaffinity(0, sizeof(mask), &mask) == -1) {
5039 tty->print_cr("Can't bind to processor_id = %d!", processor_id);
5040 return false;
5041 }
5042 else {
5043 return true;
5044 }
5045 }
5046 else {
5047 return false;
5048 }
5049 }
5051 ///
5053 void os::SuspendedThreadTask::internal_do_task() {
5054 if (do_suspend(_thread->osthread())) {
5055 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5056 do_task(context);
5057 do_resume(_thread->osthread());
5058 }
5059 }
5061 class PcFetcher : public os::SuspendedThreadTask {
5062 public:
5063 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5064 ExtendedPC result();
5065 protected:
5066 void do_task(const os::SuspendedThreadTaskContext& context);
5067 private:
5068 ExtendedPC _epc;
5069 };
5071 ExtendedPC PcFetcher::result() {
5072 guarantee(is_done(), "task is not done yet.");
5073 return _epc;
5074 }
5076 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5077 Thread* thread = context.thread();
5078 OSThread* osthread = thread->osthread();
5079 if (osthread->ucontext() != NULL) {
5080 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5081 } else {
5082 // NULL context is unexpected, double-check this is the VMThread
5083 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5084 }
5085 }
5087 // Suspends the target using the signal mechanism and then grabs the PC before
5088 // resuming the target. Used by the flat-profiler only
5089 ExtendedPC os::get_thread_pc(Thread* thread) {
5090 // Make sure that it is called by the watcher for the VMThread
5091 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5092 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5094 PcFetcher fetcher(thread);
5095 fetcher.run();
5096 return fetcher.result();
5097 }
5099 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5100 {
5101 if (is_NPTL()) {
5102 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5103 } else {
5104 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5105 // word back to default 64bit precision if condvar is signaled. Java
5106 // wants 53bit precision. Save and restore current value.
5107 int fpu = get_fpu_control_word();
5108 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5109 set_fpu_control_word(fpu);
5110 return status;
5111 }
5112 }
5114 ////////////////////////////////////////////////////////////////////////////////
5115 // debug support
5117 bool os::find(address addr, outputStream* st) {
5118 Dl_info dlinfo;
5119 memset(&dlinfo, 0, sizeof(dlinfo));
5120 if (dladdr(addr, &dlinfo) != 0) {
5121 st->print(PTR_FORMAT ": ", addr);
5122 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5123 st->print("%s+%#x", dlinfo.dli_sname,
5124 addr - (intptr_t)dlinfo.dli_saddr);
5125 } else if (dlinfo.dli_fbase != NULL) {
5126 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5127 } else {
5128 st->print("<absolute address>");
5129 }
5130 if (dlinfo.dli_fname != NULL) {
5131 st->print(" in %s", dlinfo.dli_fname);
5132 }
5133 if (dlinfo.dli_fbase != NULL) {
5134 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5135 }
5136 st->cr();
5138 if (Verbose) {
5139 // decode some bytes around the PC
5140 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5141 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5142 address lowest = (address) dlinfo.dli_sname;
5143 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5144 if (begin < lowest) begin = lowest;
5145 Dl_info dlinfo2;
5146 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5147 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5148 end = (address) dlinfo2.dli_saddr;
5149 Disassembler::decode(begin, end, st);
5150 }
5151 return true;
5152 }
5153 return false;
5154 }
5156 ////////////////////////////////////////////////////////////////////////////////
5157 // misc
5159 // This does not do anything on Linux. This is basically a hook for being
5160 // able to use structured exception handling (thread-local exception filters)
5161 // on, e.g., Win32.
5162 void
5163 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5164 JavaCallArguments* args, Thread* thread) {
5165 f(value, method, args, thread);
5166 }
5168 void os::print_statistics() {
5169 }
5171 int os::message_box(const char* title, const char* message) {
5172 int i;
5173 fdStream err(defaultStream::error_fd());
5174 for (i = 0; i < 78; i++) err.print_raw("=");
5175 err.cr();
5176 err.print_raw_cr(title);
5177 for (i = 0; i < 78; i++) err.print_raw("-");
5178 err.cr();
5179 err.print_raw_cr(message);
5180 for (i = 0; i < 78; i++) err.print_raw("=");
5181 err.cr();
5183 char buf[16];
5184 // Prevent process from exiting upon "read error" without consuming all CPU
5185 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5187 return buf[0] == 'y' || buf[0] == 'Y';
5188 }
5190 int os::stat(const char *path, struct stat *sbuf) {
5191 char pathbuf[MAX_PATH];
5192 if (strlen(path) > MAX_PATH - 1) {
5193 errno = ENAMETOOLONG;
5194 return -1;
5195 }
5196 os::native_path(strcpy(pathbuf, path));
5197 return ::stat(pathbuf, sbuf);
5198 }
5200 bool os::check_heap(bool force) {
5201 return true;
5202 }
5204 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5205 return ::vsnprintf(buf, count, format, args);
5206 }
5208 // Is a (classpath) directory empty?
5209 bool os::dir_is_empty(const char* path) {
5210 DIR *dir = NULL;
5211 struct dirent *ptr;
5213 dir = opendir(path);
5214 if (dir == NULL) return true;
5216 /* Scan the directory */
5217 bool result = true;
5218 char buf[sizeof(struct dirent) + MAX_PATH];
5219 while (result && (ptr = ::readdir(dir)) != NULL) {
5220 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5221 result = false;
5222 }
5223 }
5224 closedir(dir);
5225 return result;
5226 }
5228 // This code originates from JDK's sysOpen and open64_w
5229 // from src/solaris/hpi/src/system_md.c
5231 #ifndef O_DELETE
5232 #define O_DELETE 0x10000
5233 #endif
5235 // Open a file. Unlink the file immediately after open returns
5236 // if the specified oflag has the O_DELETE flag set.
5237 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5239 int os::open(const char *path, int oflag, int mode) {
5241 if (strlen(path) > MAX_PATH - 1) {
5242 errno = ENAMETOOLONG;
5243 return -1;
5244 }
5245 int fd;
5246 int o_delete = (oflag & O_DELETE);
5247 oflag = oflag & ~O_DELETE;
5249 fd = ::open64(path, oflag, mode);
5250 if (fd == -1) return -1;
5252 //If the open succeeded, the file might still be a directory
5253 {
5254 struct stat64 buf64;
5255 int ret = ::fstat64(fd, &buf64);
5256 int st_mode = buf64.st_mode;
5258 if (ret != -1) {
5259 if ((st_mode & S_IFMT) == S_IFDIR) {
5260 errno = EISDIR;
5261 ::close(fd);
5262 return -1;
5263 }
5264 } else {
5265 ::close(fd);
5266 return -1;
5267 }
5268 }
5270 /*
5271 * All file descriptors that are opened in the JVM and not
5272 * specifically destined for a subprocess should have the
5273 * close-on-exec flag set. If we don't set it, then careless 3rd
5274 * party native code might fork and exec without closing all
5275 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5276 * UNIXProcess.c), and this in turn might:
5277 *
5278 * - cause end-of-file to fail to be detected on some file
5279 * descriptors, resulting in mysterious hangs, or
5280 *
5281 * - might cause an fopen in the subprocess to fail on a system
5282 * suffering from bug 1085341.
5283 *
5284 * (Yes, the default setting of the close-on-exec flag is a Unix
5285 * design flaw)
5286 *
5287 * See:
5288 * 1085341: 32-bit stdio routines should support file descriptors >255
5289 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5290 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5291 */
5292 #ifdef FD_CLOEXEC
5293 {
5294 int flags = ::fcntl(fd, F_GETFD);
5295 if (flags != -1)
5296 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5297 }
5298 #endif
5300 if (o_delete != 0) {
5301 ::unlink(path);
5302 }
5303 return fd;
5304 }
5307 // create binary file, rewriting existing file if required
5308 int os::create_binary_file(const char* path, bool rewrite_existing) {
5309 int oflags = O_WRONLY | O_CREAT;
5310 if (!rewrite_existing) {
5311 oflags |= O_EXCL;
5312 }
5313 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5314 }
5316 // return current position of file pointer
5317 jlong os::current_file_offset(int fd) {
5318 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5319 }
5321 // move file pointer to the specified offset
5322 jlong os::seek_to_file_offset(int fd, jlong offset) {
5323 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5324 }
5326 // This code originates from JDK's sysAvailable
5327 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5329 int os::available(int fd, jlong *bytes) {
5330 jlong cur, end;
5331 int mode;
5332 struct stat64 buf64;
5334 if (::fstat64(fd, &buf64) >= 0) {
5335 mode = buf64.st_mode;
5336 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5337 /*
5338 * XXX: is the following call interruptible? If so, this might
5339 * need to go through the INTERRUPT_IO() wrapper as for other
5340 * blocking, interruptible calls in this file.
5341 */
5342 int n;
5343 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5344 *bytes = n;
5345 return 1;
5346 }
5347 }
5348 }
5349 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5350 return 0;
5351 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5352 return 0;
5353 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5354 return 0;
5355 }
5356 *bytes = end - cur;
5357 return 1;
5358 }
5360 int os::socket_available(int fd, jint *pbytes) {
5361 // Linux doc says EINTR not returned, unlike Solaris
5362 int ret = ::ioctl(fd, FIONREAD, pbytes);
5364 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5365 // is expected to return 0 on failure and 1 on success to the jdk.
5366 return (ret < 0) ? 0 : 1;
5367 }
5369 // Map a block of memory.
5370 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5371 char *addr, size_t bytes, bool read_only,
5372 bool allow_exec) {
5373 int prot;
5374 int flags = MAP_PRIVATE;
5376 if (read_only) {
5377 prot = PROT_READ;
5378 } else {
5379 prot = PROT_READ | PROT_WRITE;
5380 }
5382 if (allow_exec) {
5383 prot |= PROT_EXEC;
5384 }
5386 if (addr != NULL) {
5387 flags |= MAP_FIXED;
5388 }
5390 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5391 fd, file_offset);
5392 if (mapped_address == MAP_FAILED) {
5393 return NULL;
5394 }
5395 return mapped_address;
5396 }
5399 // Remap a block of memory.
5400 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5401 char *addr, size_t bytes, bool read_only,
5402 bool allow_exec) {
5403 // same as map_memory() on this OS
5404 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5405 allow_exec);
5406 }
5409 // Unmap a block of memory.
5410 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5411 return munmap(addr, bytes) == 0;
5412 }
5414 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5416 static clockid_t thread_cpu_clockid(Thread* thread) {
5417 pthread_t tid = thread->osthread()->pthread_id();
5418 clockid_t clockid;
5420 // Get thread clockid
5421 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5422 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5423 return clockid;
5424 }
5426 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5427 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5428 // of a thread.
5429 //
5430 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5431 // the fast estimate available on the platform.
5433 jlong os::current_thread_cpu_time() {
5434 if (os::Linux::supports_fast_thread_cpu_time()) {
5435 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5436 } else {
5437 // return user + sys since the cost is the same
5438 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5439 }
5440 }
5442 jlong os::thread_cpu_time(Thread* thread) {
5443 // consistent with what current_thread_cpu_time() returns
5444 if (os::Linux::supports_fast_thread_cpu_time()) {
5445 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5446 } else {
5447 return slow_thread_cpu_time(thread, true /* user + sys */);
5448 }
5449 }
5451 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5452 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5453 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5454 } else {
5455 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5456 }
5457 }
5459 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5460 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5461 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5462 } else {
5463 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5464 }
5465 }
5467 //
5468 // -1 on error.
5469 //
5471 PRAGMA_DIAG_PUSH
5472 PRAGMA_FORMAT_NONLITERAL_IGNORED
5473 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5474 static bool proc_task_unchecked = true;
5475 static const char *proc_stat_path = "/proc/%d/stat";
5476 pid_t tid = thread->osthread()->thread_id();
5477 char *s;
5478 char stat[2048];
5479 int statlen;
5480 char proc_name[64];
5481 int count;
5482 long sys_time, user_time;
5483 char cdummy;
5484 int idummy;
5485 long ldummy;
5486 FILE *fp;
5488 // The /proc/<tid>/stat aggregates per-process usage on
5489 // new Linux kernels 2.6+ where NPTL is supported.
5490 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5491 // See bug 6328462.
5492 // There possibly can be cases where there is no directory
5493 // /proc/self/task, so we check its availability.
5494 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5495 // This is executed only once
5496 proc_task_unchecked = false;
5497 fp = fopen("/proc/self/task", "r");
5498 if (fp != NULL) {
5499 proc_stat_path = "/proc/self/task/%d/stat";
5500 fclose(fp);
5501 }
5502 }
5504 sprintf(proc_name, proc_stat_path, tid);
5505 fp = fopen(proc_name, "r");
5506 if ( fp == NULL ) return -1;
5507 statlen = fread(stat, 1, 2047, fp);
5508 stat[statlen] = '\0';
5509 fclose(fp);
5511 // Skip pid and the command string. Note that we could be dealing with
5512 // weird command names, e.g. user could decide to rename java launcher
5513 // to "java 1.4.2 :)", then the stat file would look like
5514 // 1234 (java 1.4.2 :)) R ... ...
5515 // We don't really need to know the command string, just find the last
5516 // occurrence of ")" and then start parsing from there. See bug 4726580.
5517 s = strrchr(stat, ')');
5518 if (s == NULL ) return -1;
5520 // Skip blank chars
5521 do s++; while (isspace(*s));
5523 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5524 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5525 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5526 &user_time, &sys_time);
5527 if ( count != 13 ) return -1;
5528 if (user_sys_cpu_time) {
5529 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5530 } else {
5531 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5532 }
5533 }
5534 PRAGMA_DIAG_POP
5536 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5537 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5538 info_ptr->may_skip_backward = false; // elapsed time not wall time
5539 info_ptr->may_skip_forward = false; // elapsed time not wall time
5540 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5541 }
5543 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5544 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5545 info_ptr->may_skip_backward = false; // elapsed time not wall time
5546 info_ptr->may_skip_forward = false; // elapsed time not wall time
5547 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5548 }
5550 bool os::is_thread_cpu_time_supported() {
5551 return true;
5552 }
5554 // System loadavg support. Returns -1 if load average cannot be obtained.
5555 // Linux doesn't yet have a (official) notion of processor sets,
5556 // so just return the system wide load average.
5557 int os::loadavg(double loadavg[], int nelem) {
5558 return ::getloadavg(loadavg, nelem);
5559 }
5561 void os::pause() {
5562 char filename[MAX_PATH];
5563 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5564 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5565 } else {
5566 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5567 }
5569 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5570 if (fd != -1) {
5571 struct stat buf;
5572 ::close(fd);
5573 while (::stat(filename, &buf) == 0) {
5574 (void)::poll(NULL, 0, 100);
5575 }
5576 } else {
5577 jio_fprintf(stderr,
5578 "Could not open pause file '%s', continuing immediately.\n", filename);
5579 }
5580 }
5583 // Refer to the comments in os_solaris.cpp park-unpark.
5584 //
5585 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5586 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5587 // For specifics regarding the bug see GLIBC BUGID 261237 :
5588 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5589 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5590 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5591 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5592 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5593 // and monitorenter when we're using 1-0 locking. All those operations may result in
5594 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5595 // of libpthread avoids the problem, but isn't practical.
5596 //
5597 // Possible remedies:
5598 //
5599 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5600 // This is palliative and probabilistic, however. If the thread is preempted
5601 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5602 // than the minimum period may have passed, and the abstime may be stale (in the
5603 // past) resultin in a hang. Using this technique reduces the odds of a hang
5604 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5605 //
5606 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5607 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5608 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5609 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5610 // thread.
5611 //
5612 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5613 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5614 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5615 // This also works well. In fact it avoids kernel-level scalability impediments
5616 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5617 // timers in a graceful fashion.
5618 //
5619 // 4. When the abstime value is in the past it appears that control returns
5620 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5621 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5622 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5623 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5624 // It may be possible to avoid reinitialization by checking the return
5625 // value from pthread_cond_timedwait(). In addition to reinitializing the
5626 // condvar we must establish the invariant that cond_signal() is only called
5627 // within critical sections protected by the adjunct mutex. This prevents
5628 // cond_signal() from "seeing" a condvar that's in the midst of being
5629 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5630 // desirable signal-after-unlock optimization that avoids futile context switching.
5631 //
5632 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5633 // structure when a condvar is used or initialized. cond_destroy() would
5634 // release the helper structure. Our reinitialize-after-timedwait fix
5635 // put excessive stress on malloc/free and locks protecting the c-heap.
5636 //
5637 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5638 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5639 // and only enabling the work-around for vulnerable environments.
5641 // utility to compute the abstime argument to timedwait:
5642 // millis is the relative timeout time
5643 // abstime will be the absolute timeout time
5644 // TODO: replace compute_abstime() with unpackTime()
5646 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5647 if (millis < 0) millis = 0;
5649 jlong seconds = millis / 1000;
5650 millis %= 1000;
5651 if (seconds > 50000000) { // see man cond_timedwait(3T)
5652 seconds = 50000000;
5653 }
5655 if (os::Linux::supports_monotonic_clock()) {
5656 struct timespec now;
5657 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5658 assert_status(status == 0, status, "clock_gettime");
5659 abstime->tv_sec = now.tv_sec + seconds;
5660 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5661 if (nanos >= NANOSECS_PER_SEC) {
5662 abstime->tv_sec += 1;
5663 nanos -= NANOSECS_PER_SEC;
5664 }
5665 abstime->tv_nsec = nanos;
5666 } else {
5667 struct timeval now;
5668 int status = gettimeofday(&now, NULL);
5669 assert(status == 0, "gettimeofday");
5670 abstime->tv_sec = now.tv_sec + seconds;
5671 long usec = now.tv_usec + millis * 1000;
5672 if (usec >= 1000000) {
5673 abstime->tv_sec += 1;
5674 usec -= 1000000;
5675 }
5676 abstime->tv_nsec = usec * 1000;
5677 }
5678 return abstime;
5679 }
5682 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5683 // Conceptually TryPark() should be equivalent to park(0).
5685 int os::PlatformEvent::TryPark() {
5686 for (;;) {
5687 const int v = _Event ;
5688 guarantee ((v == 0) || (v == 1), "invariant") ;
5689 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5690 }
5691 }
5693 void os::PlatformEvent::park() { // AKA "down()"
5694 // Invariant: Only the thread associated with the Event/PlatformEvent
5695 // may call park().
5696 // TODO: assert that _Assoc != NULL or _Assoc == Self
5697 int v ;
5698 for (;;) {
5699 v = _Event ;
5700 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5701 }
5702 guarantee (v >= 0, "invariant") ;
5703 if (v == 0) {
5704 // Do this the hard way by blocking ...
5705 int status = pthread_mutex_lock(_mutex);
5706 assert_status(status == 0, status, "mutex_lock");
5707 guarantee (_nParked == 0, "invariant") ;
5708 ++ _nParked ;
5709 while (_Event < 0) {
5710 status = pthread_cond_wait(_cond, _mutex);
5711 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5712 // Treat this the same as if the wait was interrupted
5713 if (status == ETIME) { status = EINTR; }
5714 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5715 }
5716 -- _nParked ;
5718 _Event = 0 ;
5719 status = pthread_mutex_unlock(_mutex);
5720 assert_status(status == 0, status, "mutex_unlock");
5721 // Paranoia to ensure our locked and lock-free paths interact
5722 // correctly with each other.
5723 OrderAccess::fence();
5724 }
5725 guarantee (_Event >= 0, "invariant") ;
5726 }
5728 int os::PlatformEvent::park(jlong millis) {
5729 guarantee (_nParked == 0, "invariant") ;
5731 int v ;
5732 for (;;) {
5733 v = _Event ;
5734 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5735 }
5736 guarantee (v >= 0, "invariant") ;
5737 if (v != 0) return OS_OK ;
5739 // We do this the hard way, by blocking the thread.
5740 // Consider enforcing a minimum timeout value.
5741 struct timespec abst;
5742 compute_abstime(&abst, millis);
5744 int ret = OS_TIMEOUT;
5745 int status = pthread_mutex_lock(_mutex);
5746 assert_status(status == 0, status, "mutex_lock");
5747 guarantee (_nParked == 0, "invariant") ;
5748 ++_nParked ;
5750 // Object.wait(timo) will return because of
5751 // (a) notification
5752 // (b) timeout
5753 // (c) thread.interrupt
5754 //
5755 // Thread.interrupt and object.notify{All} both call Event::set.
5756 // That is, we treat thread.interrupt as a special case of notification.
5757 // The underlying Solaris implementation, cond_timedwait, admits
5758 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5759 // JVM from making those visible to Java code. As such, we must
5760 // filter out spurious wakeups. We assume all ETIME returns are valid.
5761 //
5762 // TODO: properly differentiate simultaneous notify+interrupt.
5763 // In that case, we should propagate the notify to another waiter.
5765 while (_Event < 0) {
5766 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5767 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5768 pthread_cond_destroy (_cond);
5769 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5770 }
5771 assert_status(status == 0 || status == EINTR ||
5772 status == ETIME || status == ETIMEDOUT,
5773 status, "cond_timedwait");
5774 if (!FilterSpuriousWakeups) break ; // previous semantics
5775 if (status == ETIME || status == ETIMEDOUT) break ;
5776 // We consume and ignore EINTR and spurious wakeups.
5777 }
5778 --_nParked ;
5779 if (_Event >= 0) {
5780 ret = OS_OK;
5781 }
5782 _Event = 0 ;
5783 status = pthread_mutex_unlock(_mutex);
5784 assert_status(status == 0, status, "mutex_unlock");
5785 assert (_nParked == 0, "invariant") ;
5786 // Paranoia to ensure our locked and lock-free paths interact
5787 // correctly with each other.
5788 OrderAccess::fence();
5789 return ret;
5790 }
5792 void os::PlatformEvent::unpark() {
5793 // Transitions for _Event:
5794 // 0 :=> 1
5795 // 1 :=> 1
5796 // -1 :=> either 0 or 1; must signal target thread
5797 // That is, we can safely transition _Event from -1 to either
5798 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5799 // unpark() calls.
5800 // See also: "Semaphores in Plan 9" by Mullender & Cox
5801 //
5802 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5803 // that it will take two back-to-back park() calls for the owning
5804 // thread to block. This has the benefit of forcing a spurious return
5805 // from the first park() call after an unpark() call which will help
5806 // shake out uses of park() and unpark() without condition variables.
5808 if (Atomic::xchg(1, &_Event) >= 0) return;
5810 // Wait for the thread associated with the event to vacate
5811 int status = pthread_mutex_lock(_mutex);
5812 assert_status(status == 0, status, "mutex_lock");
5813 int AnyWaiters = _nParked;
5814 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5815 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5816 AnyWaiters = 0;
5817 pthread_cond_signal(_cond);
5818 }
5819 status = pthread_mutex_unlock(_mutex);
5820 assert_status(status == 0, status, "mutex_unlock");
5821 if (AnyWaiters != 0) {
5822 status = pthread_cond_signal(_cond);
5823 assert_status(status == 0, status, "cond_signal");
5824 }
5826 // Note that we signal() _after dropping the lock for "immortal" Events.
5827 // This is safe and avoids a common class of futile wakeups. In rare
5828 // circumstances this can cause a thread to return prematurely from
5829 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5830 // simply re-test the condition and re-park itself.
5831 }
5834 // JSR166
5835 // -------------------------------------------------------
5837 /*
5838 * The solaris and linux implementations of park/unpark are fairly
5839 * conservative for now, but can be improved. They currently use a
5840 * mutex/condvar pair, plus a a count.
5841 * Park decrements count if > 0, else does a condvar wait. Unpark
5842 * sets count to 1 and signals condvar. Only one thread ever waits
5843 * on the condvar. Contention seen when trying to park implies that someone
5844 * is unparking you, so don't wait. And spurious returns are fine, so there
5845 * is no need to track notifications.
5846 */
5848 /*
5849 * This code is common to linux and solaris and will be moved to a
5850 * common place in dolphin.
5851 *
5852 * The passed in time value is either a relative time in nanoseconds
5853 * or an absolute time in milliseconds. Either way it has to be unpacked
5854 * into suitable seconds and nanoseconds components and stored in the
5855 * given timespec structure.
5856 * Given time is a 64-bit value and the time_t used in the timespec is only
5857 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5858 * overflow if times way in the future are given. Further on Solaris versions
5859 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5860 * number of seconds, in abstime, is less than current_time + 100,000,000.
5861 * As it will be 28 years before "now + 100000000" will overflow we can
5862 * ignore overflow and just impose a hard-limit on seconds using the value
5863 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5864 * years from "now".
5865 */
5867 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5868 assert (time > 0, "convertTime");
5869 time_t max_secs = 0;
5871 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5872 struct timeval now;
5873 int status = gettimeofday(&now, NULL);
5874 assert(status == 0, "gettimeofday");
5876 max_secs = now.tv_sec + MAX_SECS;
5878 if (isAbsolute) {
5879 jlong secs = time / 1000;
5880 if (secs > max_secs) {
5881 absTime->tv_sec = max_secs;
5882 } else {
5883 absTime->tv_sec = secs;
5884 }
5885 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5886 } else {
5887 jlong secs = time / NANOSECS_PER_SEC;
5888 if (secs >= MAX_SECS) {
5889 absTime->tv_sec = max_secs;
5890 absTime->tv_nsec = 0;
5891 } else {
5892 absTime->tv_sec = now.tv_sec + secs;
5893 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5894 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5895 absTime->tv_nsec -= NANOSECS_PER_SEC;
5896 ++absTime->tv_sec; // note: this must be <= max_secs
5897 }
5898 }
5899 }
5900 } else {
5901 // must be relative using monotonic clock
5902 struct timespec now;
5903 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5904 assert_status(status == 0, status, "clock_gettime");
5905 max_secs = now.tv_sec + MAX_SECS;
5906 jlong secs = time / NANOSECS_PER_SEC;
5907 if (secs >= MAX_SECS) {
5908 absTime->tv_sec = max_secs;
5909 absTime->tv_nsec = 0;
5910 } else {
5911 absTime->tv_sec = now.tv_sec + secs;
5912 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5913 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5914 absTime->tv_nsec -= NANOSECS_PER_SEC;
5915 ++absTime->tv_sec; // note: this must be <= max_secs
5916 }
5917 }
5918 }
5919 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5920 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5921 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5922 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5923 }
5925 void Parker::park(bool isAbsolute, jlong time) {
5926 // Ideally we'd do something useful while spinning, such
5927 // as calling unpackTime().
5929 // Optional fast-path check:
5930 // Return immediately if a permit is available.
5931 // We depend on Atomic::xchg() having full barrier semantics
5932 // since we are doing a lock-free update to _counter.
5933 if (Atomic::xchg(0, &_counter) > 0) return;
5935 Thread* thread = Thread::current();
5936 assert(thread->is_Java_thread(), "Must be JavaThread");
5937 JavaThread *jt = (JavaThread *)thread;
5939 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5940 // Check interrupt before trying to wait
5941 if (Thread::is_interrupted(thread, false)) {
5942 return;
5943 }
5945 // Next, demultiplex/decode time arguments
5946 timespec absTime;
5947 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5948 return;
5949 }
5950 if (time > 0) {
5951 unpackTime(&absTime, isAbsolute, time);
5952 }
5955 // Enter safepoint region
5956 // Beware of deadlocks such as 6317397.
5957 // The per-thread Parker:: mutex is a classic leaf-lock.
5958 // In particular a thread must never block on the Threads_lock while
5959 // holding the Parker:: mutex. If safepoints are pending both the
5960 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5961 ThreadBlockInVM tbivm(jt);
5963 // Don't wait if cannot get lock since interference arises from
5964 // unblocking. Also. check interrupt before trying wait
5965 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5966 return;
5967 }
5969 int status ;
5970 if (_counter > 0) { // no wait needed
5971 _counter = 0;
5972 status = pthread_mutex_unlock(_mutex);
5973 assert (status == 0, "invariant") ;
5974 // Paranoia to ensure our locked and lock-free paths interact
5975 // correctly with each other and Java-level accesses.
5976 OrderAccess::fence();
5977 return;
5978 }
5980 #ifdef ASSERT
5981 // Don't catch signals while blocked; let the running threads have the signals.
5982 // (This allows a debugger to break into the running thread.)
5983 sigset_t oldsigs;
5984 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5985 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5986 #endif
5988 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5989 jt->set_suspend_equivalent();
5990 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5992 assert(_cur_index == -1, "invariant");
5993 if (time == 0) {
5994 _cur_index = REL_INDEX; // arbitrary choice when not timed
5995 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5996 } else {
5997 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5998 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5999 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6000 pthread_cond_destroy (&_cond[_cur_index]) ;
6001 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6002 }
6003 }
6004 _cur_index = -1;
6005 assert_status(status == 0 || status == EINTR ||
6006 status == ETIME || status == ETIMEDOUT,
6007 status, "cond_timedwait");
6009 #ifdef ASSERT
6010 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6011 #endif
6013 _counter = 0 ;
6014 status = pthread_mutex_unlock(_mutex) ;
6015 assert_status(status == 0, status, "invariant") ;
6016 // Paranoia to ensure our locked and lock-free paths interact
6017 // correctly with each other and Java-level accesses.
6018 OrderAccess::fence();
6020 // If externally suspended while waiting, re-suspend
6021 if (jt->handle_special_suspend_equivalent_condition()) {
6022 jt->java_suspend_self();
6023 }
6024 }
6026 void Parker::unpark() {
6027 int s, status ;
6028 status = pthread_mutex_lock(_mutex);
6029 assert (status == 0, "invariant") ;
6030 s = _counter;
6031 _counter = 1;
6032 if (s < 1) {
6033 // thread might be parked
6034 if (_cur_index != -1) {
6035 // thread is definitely parked
6036 if (WorkAroundNPTLTimedWaitHang) {
6037 status = pthread_cond_signal (&_cond[_cur_index]);
6038 assert (status == 0, "invariant");
6039 status = pthread_mutex_unlock(_mutex);
6040 assert (status == 0, "invariant");
6041 } else {
6042 status = pthread_mutex_unlock(_mutex);
6043 assert (status == 0, "invariant");
6044 status = pthread_cond_signal (&_cond[_cur_index]);
6045 assert (status == 0, "invariant");
6046 }
6047 } else {
6048 pthread_mutex_unlock(_mutex);
6049 assert (status == 0, "invariant") ;
6050 }
6051 } else {
6052 pthread_mutex_unlock(_mutex);
6053 assert (status == 0, "invariant") ;
6054 }
6055 }
6058 extern char** environ;
6060 #ifndef __NR_fork
6061 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
6062 #endif
6064 #ifndef __NR_execve
6065 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
6066 #endif
6068 // Run the specified command in a separate process. Return its exit value,
6069 // or -1 on failure (e.g. can't fork a new process).
6070 // Unlike system(), this function can be called from signal handler. It
6071 // doesn't block SIGINT et al.
6072 int os::fork_and_exec(char* cmd) {
6073 const char * argv[4] = {"sh", "-c", cmd, NULL};
6075 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
6076 // pthread_atfork handlers and reset pthread library. All we need is a
6077 // separate process to execve. Make a direct syscall to fork process.
6078 // On IA64 there's no fork syscall, we have to use fork() and hope for
6079 // the best...
6080 pid_t pid = NOT_IA64(syscall(__NR_fork);)
6081 IA64_ONLY(fork();)
6083 if (pid < 0) {
6084 // fork failed
6085 return -1;
6087 } else if (pid == 0) {
6088 // child process
6090 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6091 // first to kill every thread on the thread list. Because this list is
6092 // not reset by fork() (see notes above), execve() will instead kill
6093 // every thread in the parent process. We know this is the only thread
6094 // in the new process, so make a system call directly.
6095 // IA64 should use normal execve() from glibc to match the glibc fork()
6096 // above.
6097 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6098 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6100 // execve failed
6101 _exit(-1);
6103 } else {
6104 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6105 // care about the actual exit code, for now.
6107 int status;
6109 // Wait for the child process to exit. This returns immediately if
6110 // the child has already exited. */
6111 while (waitpid(pid, &status, 0) < 0) {
6112 switch (errno) {
6113 case ECHILD: return 0;
6114 case EINTR: break;
6115 default: return -1;
6116 }
6117 }
6119 if (WIFEXITED(status)) {
6120 // The child exited normally; get its exit code.
6121 return WEXITSTATUS(status);
6122 } else if (WIFSIGNALED(status)) {
6123 // The child exited because of a signal
6124 // The best value to return is 0x80 + signal number,
6125 // because that is what all Unix shells do, and because
6126 // it allows callers to distinguish between process exit and
6127 // process death by signal.
6128 return 0x80 + WTERMSIG(status);
6129 } else {
6130 // Unknown exit code; pass it through
6131 return status;
6132 }
6133 }
6134 }
6136 // is_headless_jre()
6137 //
6138 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6139 // in order to report if we are running in a headless jre
6140 //
6141 // Since JDK8 xawt/libmawt.so was moved into the same directory
6142 // as libawt.so, and renamed libawt_xawt.so
6143 //
6144 bool os::is_headless_jre() {
6145 struct stat statbuf;
6146 char buf[MAXPATHLEN];
6147 char libmawtpath[MAXPATHLEN];
6148 const char *xawtstr = "/xawt/libmawt.so";
6149 const char *new_xawtstr = "/libawt_xawt.so";
6150 char *p;
6152 // Get path to libjvm.so
6153 os::jvm_path(buf, sizeof(buf));
6155 // Get rid of libjvm.so
6156 p = strrchr(buf, '/');
6157 if (p == NULL) return false;
6158 else *p = '\0';
6160 // Get rid of client or server
6161 p = strrchr(buf, '/');
6162 if (p == NULL) return false;
6163 else *p = '\0';
6165 // check xawt/libmawt.so
6166 strcpy(libmawtpath, buf);
6167 strcat(libmawtpath, xawtstr);
6168 if (::stat(libmawtpath, &statbuf) == 0) return false;
6170 // check libawt_xawt.so
6171 strcpy(libmawtpath, buf);
6172 strcat(libmawtpath, new_xawtstr);
6173 if (::stat(libmawtpath, &statbuf) == 0) return false;
6175 return true;
6176 }
6178 // Get the default path to the core file
6179 // Returns the length of the string
6180 int os::get_core_path(char* buffer, size_t bufferSize) {
6181 const char* p = get_current_directory(buffer, bufferSize);
6183 if (p == NULL) {
6184 assert(p != NULL, "failed to get current directory");
6185 return 0;
6186 }
6188 return strlen(buffer);
6189 }
6191 #ifdef JAVASE_EMBEDDED
6192 //
6193 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6194 //
6195 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6197 // ctor
6198 //
6199 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6200 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6201 _fd = fd;
6203 if (os::create_thread(this, os::os_thread)) {
6204 _memnotify_thread = this;
6205 os::set_priority(this, NearMaxPriority);
6206 os::start_thread(this);
6207 }
6208 }
6210 // Where all the work gets done
6211 //
6212 void MemNotifyThread::run() {
6213 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6215 // Set up the select arguments
6216 fd_set rfds;
6217 if (_fd != -1) {
6218 FD_ZERO(&rfds);
6219 FD_SET(_fd, &rfds);
6220 }
6222 // Now wait for the mem_notify device to wake up
6223 while (1) {
6224 // Wait for the mem_notify device to signal us..
6225 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6226 if (rc == -1) {
6227 perror("select!\n");
6228 break;
6229 } else if (rc) {
6230 //ssize_t free_before = os::available_memory();
6231 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6233 // The kernel is telling us there is not much memory left...
6234 // try to do something about that
6236 // If we are not already in a GC, try one.
6237 if (!Universe::heap()->is_gc_active()) {
6238 Universe::heap()->collect(GCCause::_allocation_failure);
6240 //ssize_t free_after = os::available_memory();
6241 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6242 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6243 }
6244 // We might want to do something like the following if we find the GC's are not helping...
6245 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6246 }
6247 }
6248 }
6250 //
6251 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6252 //
6253 void MemNotifyThread::start() {
6254 int fd;
6255 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6256 if (fd < 0) {
6257 return;
6258 }
6260 if (memnotify_thread() == NULL) {
6261 new MemNotifyThread(fd);
6262 }
6263 }
6265 #endif // JAVASE_EMBEDDED
6268 /////////////// Unit tests ///////////////
6270 #ifndef PRODUCT
6272 #define test_log(...) \
6273 do {\
6274 if (VerboseInternalVMTests) { \
6275 tty->print_cr(__VA_ARGS__); \
6276 tty->flush(); \
6277 }\
6278 } while (false)
6280 class TestReserveMemorySpecial : AllStatic {
6281 public:
6282 static void small_page_write(void* addr, size_t size) {
6283 size_t page_size = os::vm_page_size();
6285 char* end = (char*)addr + size;
6286 for (char* p = (char*)addr; p < end; p += page_size) {
6287 *p = 1;
6288 }
6289 }
6291 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6292 if (!UseHugeTLBFS) {
6293 return;
6294 }
6296 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6298 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6300 if (addr != NULL) {
6301 small_page_write(addr, size);
6303 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6304 }
6305 }
6307 static void test_reserve_memory_special_huge_tlbfs_only() {
6308 if (!UseHugeTLBFS) {
6309 return;
6310 }
6312 size_t lp = os::large_page_size();
6314 for (size_t size = lp; size <= lp * 10; size += lp) {
6315 test_reserve_memory_special_huge_tlbfs_only(size);
6316 }
6317 }
6319 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6320 if (!UseHugeTLBFS) {
6321 return;
6322 }
6324 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6325 size, alignment);
6327 assert(size >= os::large_page_size(), "Incorrect input to test");
6329 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6331 if (addr != NULL) {
6332 small_page_write(addr, size);
6334 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6335 }
6336 }
6338 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6339 size_t lp = os::large_page_size();
6340 size_t ag = os::vm_allocation_granularity();
6342 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6343 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6344 }
6345 }
6347 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6348 size_t lp = os::large_page_size();
6349 size_t ag = os::vm_allocation_granularity();
6351 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6352 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6353 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6354 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6355 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6356 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6357 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6358 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6359 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6360 }
6362 static void test_reserve_memory_special_huge_tlbfs() {
6363 if (!UseHugeTLBFS) {
6364 return;
6365 }
6367 test_reserve_memory_special_huge_tlbfs_only();
6368 test_reserve_memory_special_huge_tlbfs_mixed();
6369 }
6371 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6372 if (!UseSHM) {
6373 return;
6374 }
6376 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6378 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6380 if (addr != NULL) {
6381 assert(is_ptr_aligned(addr, alignment), "Check");
6382 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6384 small_page_write(addr, size);
6386 os::Linux::release_memory_special_shm(addr, size);
6387 }
6388 }
6390 static void test_reserve_memory_special_shm() {
6391 size_t lp = os::large_page_size();
6392 size_t ag = os::vm_allocation_granularity();
6394 for (size_t size = ag; size < lp * 3; size += ag) {
6395 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6396 test_reserve_memory_special_shm(size, alignment);
6397 }
6398 }
6399 }
6401 static void test() {
6402 test_reserve_memory_special_huge_tlbfs();
6403 test_reserve_memory_special_shm();
6404 }
6405 };
6407 void TestReserveMemorySpecial_test() {
6408 TestReserveMemorySpecial::test();
6409 }
6411 #endif