Mon, 24 Mar 2014 13:36:32 -0700
8033464: Linux code cleanup
8033931: Several nightly tests failing with assert(imin < imax) failed: Unexpected page size
Summary: cleaned up warnings in linux specific os code.
Reviewed-by: dcubed, hseigel, coleenp, dsamersoff
Contributed-by: gerald.thornbrugh@oracle.com
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 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_share_linux.hpp"
40 #include "prims/jniFastGetField.hpp"
41 #include "prims/jvm.h"
42 #include "prims/jvm_misc.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/extendedPC.hpp"
45 #include "runtime/globals.hpp"
46 #include "runtime/interfaceSupport.hpp"
47 #include "runtime/init.hpp"
48 #include "runtime/java.hpp"
49 #include "runtime/javaCalls.hpp"
50 #include "runtime/mutexLocker.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/osThread.hpp"
53 #include "runtime/perfMemory.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/statSampler.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "runtime/thread.inline.hpp"
58 #include "runtime/threadCritical.hpp"
59 #include "runtime/timer.hpp"
60 #include "services/attachListener.hpp"
61 #include "services/memTracker.hpp"
62 #include "services/runtimeService.hpp"
63 #include "utilities/decoder.hpp"
64 #include "utilities/defaultStream.hpp"
65 #include "utilities/events.hpp"
66 #include "utilities/elfFile.hpp"
67 #include "utilities/growableArray.hpp"
68 #include "utilities/vmError.hpp"
70 // put OS-includes here
71 # include <sys/types.h>
72 # include <sys/mman.h>
73 # include <sys/stat.h>
74 # include <sys/select.h>
75 # include <pthread.h>
76 # include <signal.h>
77 # include <errno.h>
78 # include <dlfcn.h>
79 # include <stdio.h>
80 # include <unistd.h>
81 # include <sys/resource.h>
82 # include <pthread.h>
83 # include <sys/stat.h>
84 # include <sys/time.h>
85 # include <sys/times.h>
86 # include <sys/utsname.h>
87 # include <sys/socket.h>
88 # include <sys/wait.h>
89 # include <pwd.h>
90 # include <poll.h>
91 # include <semaphore.h>
92 # include <fcntl.h>
93 # include <string.h>
94 # include <syscall.h>
95 # include <sys/sysinfo.h>
96 # include <gnu/libc-version.h>
97 # include <sys/ipc.h>
98 # include <sys/shm.h>
99 # include <link.h>
100 # include <stdint.h>
101 # include <inttypes.h>
102 # include <sys/ioctl.h>
104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
105 // getrusage() is prepared to handle the associated failure.
106 #ifndef RUSAGE_THREAD
107 #define RUSAGE_THREAD (1) /* only the calling thread */
108 #endif
110 #define MAX_PATH (2 * K)
112 #define MAX_SECS 100000000
114 // for timer info max values which include all bits
115 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
117 #define LARGEPAGES_BIT (1 << 6)
118 ////////////////////////////////////////////////////////////////////////////////
119 // global variables
120 julong os::Linux::_physical_memory = 0;
122 address os::Linux::_initial_thread_stack_bottom = NULL;
123 uintptr_t os::Linux::_initial_thread_stack_size = 0;
125 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
126 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
127 Mutex* os::Linux::_createThread_lock = NULL;
128 pthread_t os::Linux::_main_thread;
129 int os::Linux::_page_size = -1;
130 const int os::Linux::_vm_default_page_size = (8 * K);
131 bool os::Linux::_is_floating_stack = false;
132 bool os::Linux::_is_NPTL = false;
133 bool os::Linux::_supports_fast_thread_cpu_time = false;
134 const char * os::Linux::_glibc_version = NULL;
135 const char * os::Linux::_libpthread_version = NULL;
136 pthread_condattr_t os::Linux::_condattr[1];
138 static jlong initial_time_count=0;
140 static int clock_tics_per_sec = 100;
142 // For diagnostics to print a message once. see run_periodic_checks
143 static sigset_t check_signal_done;
144 static bool check_signals = true;;
146 static pid_t _initial_pid = 0;
148 /* Signal number used to suspend/resume a thread */
150 /* do not use any signal number less than SIGSEGV, see 4355769 */
151 static int SR_signum = SIGUSR2;
152 sigset_t SR_sigset;
154 /* Used to protect dlsym() calls */
155 static pthread_mutex_t dl_mutex;
157 // Declarations
158 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
160 #ifdef JAVASE_EMBEDDED
161 class MemNotifyThread: public Thread {
162 friend class VMStructs;
163 public:
164 virtual void run();
166 private:
167 static MemNotifyThread* _memnotify_thread;
168 int _fd;
170 public:
172 // Constructor
173 MemNotifyThread(int fd);
175 // Tester
176 bool is_memnotify_thread() const { return true; }
178 // Printing
179 char* name() const { return (char*)"Linux MemNotify Thread"; }
181 // Returns the single instance of the MemNotifyThread
182 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
184 // Create and start the single instance of MemNotifyThread
185 static void start();
186 };
187 #endif // JAVASE_EMBEDDED
189 // utility functions
191 static int SR_initialize();
193 julong os::available_memory() {
194 return Linux::available_memory();
195 }
197 julong os::Linux::available_memory() {
198 // values in struct sysinfo are "unsigned long"
199 struct sysinfo si;
200 sysinfo(&si);
202 return (julong)si.freeram * si.mem_unit;
203 }
205 julong os::physical_memory() {
206 return Linux::physical_memory();
207 }
209 ////////////////////////////////////////////////////////////////////////////////
210 // environment support
212 bool os::getenv(const char* name, char* buf, int len) {
213 const char* val = ::getenv(name);
214 if (val != NULL && strlen(val) < (size_t)len) {
215 strcpy(buf, val);
216 return true;
217 }
218 if (len > 0) buf[0] = 0; // return a null string
219 return false;
220 }
223 // Return true if user is running as root.
225 bool os::have_special_privileges() {
226 static bool init = false;
227 static bool privileges = false;
228 if (!init) {
229 privileges = (getuid() != geteuid()) || (getgid() != getegid());
230 init = true;
231 }
232 return privileges;
233 }
236 #ifndef SYS_gettid
237 // i386: 224, ia64: 1105, amd64: 186, sparc 143
238 #ifdef __ia64__
239 #define SYS_gettid 1105
240 #elif __i386__
241 #define SYS_gettid 224
242 #elif __amd64__
243 #define SYS_gettid 186
244 #elif __sparc__
245 #define SYS_gettid 143
246 #else
247 #error define gettid for the arch
248 #endif
249 #endif
251 // Cpu architecture string
252 #if defined(ZERO)
253 static char cpu_arch[] = ZERO_LIBARCH;
254 #elif defined(IA64)
255 static char cpu_arch[] = "ia64";
256 #elif defined(IA32)
257 static char cpu_arch[] = "i386";
258 #elif defined(AMD64)
259 static char cpu_arch[] = "amd64";
260 #elif defined(ARM)
261 static char cpu_arch[] = "arm";
262 #elif defined(PPC)
263 static char cpu_arch[] = "ppc";
264 #elif defined(SPARC)
265 # ifdef _LP64
266 static char cpu_arch[] = "sparcv9";
267 # else
268 static char cpu_arch[] = "sparc";
269 # endif
270 #else
271 #error Add appropriate cpu_arch setting
272 #endif
275 // pid_t gettid()
276 //
277 // Returns the kernel thread id of the currently running thread. Kernel
278 // thread id is used to access /proc.
279 //
280 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
281 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
282 //
283 pid_t os::Linux::gettid() {
284 int rslt = syscall(SYS_gettid);
285 if (rslt == -1) {
286 // old kernel, no NPTL support
287 return getpid();
288 } else {
289 return (pid_t)rslt;
290 }
291 }
293 // Most versions of linux have a bug where the number of processors are
294 // determined by looking at the /proc file system. In a chroot environment,
295 // the system call returns 1. This causes the VM to act as if it is
296 // a single processor and elide locking (see is_MP() call).
297 static bool unsafe_chroot_detected = false;
298 static const char *unstable_chroot_error = "/proc file system not found.\n"
299 "Java may be unstable running multithreaded in a chroot "
300 "environment on Linux when /proc filesystem is not mounted.";
302 void os::Linux::initialize_system_info() {
303 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
304 if (processor_count() == 1) {
305 pid_t pid = os::Linux::gettid();
306 char fname[32];
307 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
308 FILE *fp = fopen(fname, "r");
309 if (fp == NULL) {
310 unsafe_chroot_detected = true;
311 } else {
312 fclose(fp);
313 }
314 }
315 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
316 assert(processor_count() > 0, "linux error");
317 }
319 void os::init_system_properties_values() {
320 // char arch[12];
321 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
323 // The next steps are taken in the product version:
324 //
325 // Obtain the JAVA_HOME value from the location of libjvm.so.
326 // This library should be located at:
327 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
328 //
329 // If "/jre/lib/" appears at the right place in the path, then we
330 // assume libjvm.so is installed in a JDK and we use this path.
331 //
332 // Otherwise exit with message: "Could not create the Java virtual machine."
333 //
334 // The following extra steps are taken in the debugging version:
335 //
336 // If "/jre/lib/" does NOT appear at the right place in the path
337 // instead of exit check for $JAVA_HOME environment variable.
338 //
339 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
340 // then we append a fake suffix "hotspot/libjvm.so" to this path so
341 // it looks like libjvm.so is installed there
342 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
343 //
344 // Otherwise exit.
345 //
346 // Important note: if the location of libjvm.so changes this
347 // code needs to be changed accordingly.
349 // The next few definitions allow the code to be verbatim:
350 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
351 #define getenv(n) ::getenv(n)
353 /*
354 * See ld(1):
355 * The linker uses the following search paths to locate required
356 * shared libraries:
357 * 1: ...
358 * ...
359 * 7: The default directories, normally /lib and /usr/lib.
360 */
361 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
362 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
363 #else
364 #define DEFAULT_LIBPATH "/lib:/usr/lib"
365 #endif
367 #define EXTENSIONS_DIR "/lib/ext"
368 #define ENDORSED_DIR "/lib/endorsed"
369 #define REG_DIR "/usr/java/packages"
371 {
372 /* sysclasspath, java_home, dll_dir */
373 {
374 char *home_path;
375 char *dll_path;
376 char *pslash;
377 char buf[MAXPATHLEN];
378 os::jvm_path(buf, sizeof(buf));
380 // Found the full path to libjvm.so.
381 // Now cut the path to <java_home>/jre if we can.
382 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
383 pslash = strrchr(buf, '/');
384 if (pslash != NULL)
385 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
386 dll_path = malloc(strlen(buf) + 1);
387 if (dll_path == NULL)
388 return;
389 strcpy(dll_path, buf);
390 Arguments::set_dll_dir(dll_path);
392 if (pslash != NULL) {
393 pslash = strrchr(buf, '/');
394 if (pslash != NULL) {
395 *pslash = '\0'; /* get rid of /<arch> */
396 pslash = strrchr(buf, '/');
397 if (pslash != NULL)
398 *pslash = '\0'; /* get rid of /lib */
399 }
400 }
402 home_path = malloc(strlen(buf) + 1);
403 if (home_path == NULL)
404 return;
405 strcpy(home_path, buf);
406 Arguments::set_java_home(home_path);
408 if (!set_boot_path('/', ':'))
409 return;
410 }
412 /*
413 * Where to look for native libraries
414 *
415 * Note: Due to a legacy implementation, most of the library path
416 * is set in the launcher. This was to accomodate linking restrictions
417 * on legacy Linux implementations (which are no longer supported).
418 * Eventually, all the library path setting will be done here.
419 *
420 * However, to prevent the proliferation of improperly built native
421 * libraries, the new path component /usr/java/packages is added here.
422 * Eventually, all the library path setting will be done here.
423 */
424 {
425 char *ld_library_path;
427 /*
428 * Construct the invariant part of ld_library_path. Note that the
429 * space for the colon and the trailing null are provided by the
430 * nulls included by the sizeof operator (so actually we allocate
431 * a byte more than necessary).
432 */
433 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
434 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
435 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
437 /*
438 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
439 * should always exist (until the legacy problem cited above is
440 * addressed).
441 */
442 char *v = getenv("LD_LIBRARY_PATH");
443 if (v != NULL) {
444 char *t = ld_library_path;
445 /* That's +1 for the colon and +1 for the trailing '\0' */
446 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
447 sprintf(ld_library_path, "%s:%s", v, t);
448 }
449 Arguments::set_library_path(ld_library_path);
450 }
452 /*
453 * Extensions directories.
454 *
455 * Note that the space for the colon and the trailing null are provided
456 * by the nulls included by the sizeof operator (so actually one byte more
457 * than necessary is allocated).
458 */
459 {
460 char *buf = malloc(strlen(Arguments::get_java_home()) +
461 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
462 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
463 Arguments::get_java_home());
464 Arguments::set_ext_dirs(buf);
465 }
467 /* Endorsed standards default directory. */
468 {
469 char * buf;
470 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
471 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
472 Arguments::set_endorsed_dirs(buf);
473 }
474 }
476 #undef malloc
477 #undef getenv
478 #undef EXTENSIONS_DIR
479 #undef ENDORSED_DIR
481 // Done
482 return;
483 }
485 ////////////////////////////////////////////////////////////////////////////////
486 // breakpoint support
488 void os::breakpoint() {
489 BREAKPOINT;
490 }
492 extern "C" void breakpoint() {
493 // use debugger to set breakpoint here
494 }
496 ////////////////////////////////////////////////////////////////////////////////
497 // signal support
499 debug_only(static bool signal_sets_initialized = false);
500 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
502 bool os::Linux::is_sig_ignored(int sig) {
503 struct sigaction oact;
504 sigaction(sig, (struct sigaction*)NULL, &oact);
505 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
506 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
507 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
508 return true;
509 else
510 return false;
511 }
513 void os::Linux::signal_sets_init() {
514 // Should also have an assertion stating we are still single-threaded.
515 assert(!signal_sets_initialized, "Already initialized");
516 // Fill in signals that are necessarily unblocked for all threads in
517 // the VM. Currently, we unblock the following signals:
518 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
519 // by -Xrs (=ReduceSignalUsage));
520 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
521 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
522 // the dispositions or masks wrt these signals.
523 // Programs embedding the VM that want to use the above signals for their
524 // own purposes must, at this time, use the "-Xrs" option to prevent
525 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
526 // (See bug 4345157, and other related bugs).
527 // In reality, though, unblocking these signals is really a nop, since
528 // these signals are not blocked by default.
529 sigemptyset(&unblocked_sigs);
530 sigemptyset(&allowdebug_blocked_sigs);
531 sigaddset(&unblocked_sigs, SIGILL);
532 sigaddset(&unblocked_sigs, SIGSEGV);
533 sigaddset(&unblocked_sigs, SIGBUS);
534 sigaddset(&unblocked_sigs, SIGFPE);
535 sigaddset(&unblocked_sigs, SR_signum);
537 if (!ReduceSignalUsage) {
538 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
539 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
540 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
541 }
542 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
543 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
544 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
545 }
546 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
547 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
548 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
549 }
550 }
551 // Fill in signals that are blocked by all but the VM thread.
552 sigemptyset(&vm_sigs);
553 if (!ReduceSignalUsage)
554 sigaddset(&vm_sigs, BREAK_SIGNAL);
555 debug_only(signal_sets_initialized = true);
557 }
559 // These are signals that are unblocked while a thread is running Java.
560 // (For some reason, they get blocked by default.)
561 sigset_t* os::Linux::unblocked_signals() {
562 assert(signal_sets_initialized, "Not initialized");
563 return &unblocked_sigs;
564 }
566 // These are the signals that are blocked while a (non-VM) thread is
567 // running Java. Only the VM thread handles these signals.
568 sigset_t* os::Linux::vm_signals() {
569 assert(signal_sets_initialized, "Not initialized");
570 return &vm_sigs;
571 }
573 // These are signals that are blocked during cond_wait to allow debugger in
574 sigset_t* os::Linux::allowdebug_blocked_signals() {
575 assert(signal_sets_initialized, "Not initialized");
576 return &allowdebug_blocked_sigs;
577 }
579 void os::Linux::hotspot_sigmask(Thread* thread) {
581 //Save caller's signal mask before setting VM signal mask
582 sigset_t caller_sigmask;
583 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
585 OSThread* osthread = thread->osthread();
586 osthread->set_caller_sigmask(caller_sigmask);
588 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
590 if (!ReduceSignalUsage) {
591 if (thread->is_VM_thread()) {
592 // Only the VM thread handles BREAK_SIGNAL ...
593 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
594 } else {
595 // ... all other threads block BREAK_SIGNAL
596 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
597 }
598 }
599 }
601 //////////////////////////////////////////////////////////////////////////////
602 // detecting pthread library
604 void os::Linux::libpthread_init() {
605 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
606 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
607 // generic name for earlier versions.
608 // Define macros here so we can build HotSpot on old systems.
609 # ifndef _CS_GNU_LIBC_VERSION
610 # define _CS_GNU_LIBC_VERSION 2
611 # endif
612 # ifndef _CS_GNU_LIBPTHREAD_VERSION
613 # define _CS_GNU_LIBPTHREAD_VERSION 3
614 # endif
616 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
617 if (n > 0) {
618 char *str = (char *)malloc(n, mtInternal);
619 confstr(_CS_GNU_LIBC_VERSION, str, n);
620 os::Linux::set_glibc_version(str);
621 } else {
622 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
623 static char _gnu_libc_version[32];
624 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
625 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
626 os::Linux::set_glibc_version(_gnu_libc_version);
627 }
629 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
630 if (n > 0) {
631 char *str = (char *)malloc(n, mtInternal);
632 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
633 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
634 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
635 // is the case. LinuxThreads has a hard limit on max number of threads.
636 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
637 // On the other hand, NPTL does not have such a limit, sysconf()
638 // will return -1 and errno is not changed. Check if it is really NPTL.
639 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
640 strstr(str, "NPTL") &&
641 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
642 free(str);
643 os::Linux::set_libpthread_version("linuxthreads");
644 } else {
645 os::Linux::set_libpthread_version(str);
646 }
647 } else {
648 // glibc before 2.3.2 only has LinuxThreads.
649 os::Linux::set_libpthread_version("linuxthreads");
650 }
652 if (strstr(libpthread_version(), "NPTL")) {
653 os::Linux::set_is_NPTL();
654 } else {
655 os::Linux::set_is_LinuxThreads();
656 }
658 // LinuxThreads have two flavors: floating-stack mode, which allows variable
659 // stack size; and fixed-stack mode. NPTL is always floating-stack.
660 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
661 os::Linux::set_is_floating_stack();
662 }
663 }
665 /////////////////////////////////////////////////////////////////////////////
666 // thread stack
668 // Force Linux kernel to expand current thread stack. If "bottom" is close
669 // to the stack guard, caller should block all signals.
670 //
671 // MAP_GROWSDOWN:
672 // A special mmap() flag that is used to implement thread stacks. It tells
673 // kernel that the memory region should extend downwards when needed. This
674 // allows early versions of LinuxThreads to only mmap the first few pages
675 // when creating a new thread. Linux kernel will automatically expand thread
676 // stack as needed (on page faults).
677 //
678 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
679 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
680 // region, it's hard to tell if the fault is due to a legitimate stack
681 // access or because of reading/writing non-exist memory (e.g. buffer
682 // overrun). As a rule, if the fault happens below current stack pointer,
683 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
684 // application (see Linux kernel fault.c).
685 //
686 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
687 // stack overflow detection.
688 //
689 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
690 // not use this flag. However, the stack of initial thread is not created
691 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
692 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
693 // and then attach the thread to JVM.
694 //
695 // To get around the problem and allow stack banging on Linux, we need to
696 // manually expand thread stack after receiving the SIGSEGV.
697 //
698 // There are two ways to expand thread stack to address "bottom", we used
699 // both of them in JVM before 1.5:
700 // 1. adjust stack pointer first so that it is below "bottom", and then
701 // touch "bottom"
702 // 2. mmap() the page in question
703 //
704 // Now alternate signal stack is gone, it's harder to use 2. For instance,
705 // if current sp is already near the lower end of page 101, and we need to
706 // call mmap() to map page 100, it is possible that part of the mmap() frame
707 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
708 // That will destroy the mmap() frame and cause VM to crash.
709 //
710 // The following code works by adjusting sp first, then accessing the "bottom"
711 // page to force a page fault. Linux kernel will then automatically expand the
712 // stack mapping.
713 //
714 // _expand_stack_to() assumes its frame size is less than page size, which
715 // should always be true if the function is not inlined.
717 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
718 #define NOINLINE
719 #else
720 #define NOINLINE __attribute__ ((noinline))
721 #endif
723 static void _expand_stack_to(address bottom) NOINLINE;
725 static void _expand_stack_to(address bottom) {
726 address sp;
727 size_t size;
728 volatile char *p;
730 // Adjust bottom to point to the largest address within the same page, it
731 // gives us a one-page buffer if alloca() allocates slightly more memory.
732 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
733 bottom += os::Linux::page_size() - 1;
735 // sp might be slightly above current stack pointer; if that's the case, we
736 // will alloca() a little more space than necessary, which is OK. Don't use
737 // os::current_stack_pointer(), as its result can be slightly below current
738 // stack pointer, causing us to not alloca enough to reach "bottom".
739 sp = (address)&sp;
741 if (sp > bottom) {
742 size = sp - bottom;
743 p = (volatile char *)alloca(size);
744 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
745 p[0] = '\0';
746 }
747 }
749 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
750 assert(t!=NULL, "just checking");
751 assert(t->osthread()->expanding_stack(), "expand should be set");
752 assert(t->stack_base() != NULL, "stack_base was not initialized");
754 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
755 sigset_t mask_all, old_sigset;
756 sigfillset(&mask_all);
757 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
758 _expand_stack_to(addr);
759 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
760 return true;
761 }
762 return false;
763 }
765 //////////////////////////////////////////////////////////////////////////////
766 // create new thread
768 static address highest_vm_reserved_address();
770 // check if it's safe to start a new thread
771 static bool _thread_safety_check(Thread* thread) {
772 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
773 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
774 // Heap is mmap'ed at lower end of memory space. Thread stacks are
775 // allocated (MAP_FIXED) from high address space. Every thread stack
776 // occupies a fixed size slot (usually 2Mbytes, but user can change
777 // it to other values if they rebuild LinuxThreads).
778 //
779 // Problem with MAP_FIXED is that mmap() can still succeed even part of
780 // the memory region has already been mmap'ed. That means if we have too
781 // many threads and/or very large heap, eventually thread stack will
782 // collide with heap.
783 //
784 // Here we try to prevent heap/stack collision by comparing current
785 // stack bottom with the highest address that has been mmap'ed by JVM
786 // plus a safety margin for memory maps created by native code.
787 //
788 // This feature can be disabled by setting ThreadSafetyMargin to 0
789 //
790 if (ThreadSafetyMargin > 0) {
791 address stack_bottom = os::current_stack_base() - os::current_stack_size();
793 // not safe if our stack extends below the safety margin
794 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
795 } else {
796 return true;
797 }
798 } else {
799 // Floating stack LinuxThreads or NPTL:
800 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
801 // there's not enough space left, pthread_create() will fail. If we come
802 // here, that means enough space has been reserved for stack.
803 return true;
804 }
805 }
807 // Thread start routine for all newly created threads
808 static void *java_start(Thread *thread) {
809 // Try to randomize the cache line index of hot stack frames.
810 // This helps when threads of the same stack traces evict each other's
811 // cache lines. The threads can be either from the same JVM instance, or
812 // from different JVM instances. The benefit is especially true for
813 // processors with hyperthreading technology.
814 static int counter = 0;
815 int pid = os::current_process_id();
816 alloca(((pid ^ counter++) & 7) * 128);
818 ThreadLocalStorage::set_thread(thread);
820 OSThread* osthread = thread->osthread();
821 Monitor* sync = osthread->startThread_lock();
823 // non floating stack LinuxThreads needs extra check, see above
824 if (!_thread_safety_check(thread)) {
825 // notify parent thread
826 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
827 osthread->set_state(ZOMBIE);
828 sync->notify_all();
829 return NULL;
830 }
832 // thread_id is kernel thread id (similar to Solaris LWP id)
833 osthread->set_thread_id(os::Linux::gettid());
835 if (UseNUMA) {
836 int lgrp_id = os::numa_get_group_id();
837 if (lgrp_id != -1) {
838 thread->set_lgrp_id(lgrp_id);
839 }
840 }
841 // initialize signal mask for this thread
842 os::Linux::hotspot_sigmask(thread);
844 // initialize floating point control register
845 os::Linux::init_thread_fpu_state();
847 // handshaking with parent thread
848 {
849 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
851 // notify parent thread
852 osthread->set_state(INITIALIZED);
853 sync->notify_all();
855 // wait until os::start_thread()
856 while (osthread->get_state() == INITIALIZED) {
857 sync->wait(Mutex::_no_safepoint_check_flag);
858 }
859 }
861 // call one more level start routine
862 thread->run();
864 return 0;
865 }
867 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
868 assert(thread->osthread() == NULL, "caller responsible");
870 // Allocate the OSThread object
871 OSThread* osthread = new OSThread(NULL, NULL);
872 if (osthread == NULL) {
873 return false;
874 }
876 // set the correct thread state
877 osthread->set_thread_type(thr_type);
879 // Initial state is ALLOCATED but not INITIALIZED
880 osthread->set_state(ALLOCATED);
882 thread->set_osthread(osthread);
884 // init thread attributes
885 pthread_attr_t attr;
886 pthread_attr_init(&attr);
887 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
889 // stack size
890 if (os::Linux::supports_variable_stack_size()) {
891 // calculate stack size if it's not specified by caller
892 if (stack_size == 0) {
893 stack_size = os::Linux::default_stack_size(thr_type);
895 switch (thr_type) {
896 case os::java_thread:
897 // Java threads use ThreadStackSize which default value can be
898 // changed with the flag -Xss
899 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
900 stack_size = JavaThread::stack_size_at_create();
901 break;
902 case os::compiler_thread:
903 if (CompilerThreadStackSize > 0) {
904 stack_size = (size_t)(CompilerThreadStackSize * K);
905 break;
906 } // else fall through:
907 // use VMThreadStackSize if CompilerThreadStackSize is not defined
908 case os::vm_thread:
909 case os::pgc_thread:
910 case os::cgc_thread:
911 case os::watcher_thread:
912 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
913 break;
914 }
915 }
917 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
918 pthread_attr_setstacksize(&attr, stack_size);
919 } else {
920 // let pthread_create() pick the default value.
921 }
923 // glibc guard page
924 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
926 ThreadState state;
928 {
929 // Serialize thread creation if we are running with fixed stack LinuxThreads
930 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
931 if (lock) {
932 os::Linux::createThread_lock()->lock_without_safepoint_check();
933 }
935 pthread_t tid;
936 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
938 pthread_attr_destroy(&attr);
940 if (ret != 0) {
941 if (PrintMiscellaneous && (Verbose || WizardMode)) {
942 perror("pthread_create()");
943 }
944 // Need to clean up stuff we've allocated so far
945 thread->set_osthread(NULL);
946 delete osthread;
947 if (lock) os::Linux::createThread_lock()->unlock();
948 return false;
949 }
951 // Store pthread info into the OSThread
952 osthread->set_pthread_id(tid);
954 // Wait until child thread is either initialized or aborted
955 {
956 Monitor* sync_with_child = osthread->startThread_lock();
957 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
958 while ((state = osthread->get_state()) == ALLOCATED) {
959 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
960 }
961 }
963 if (lock) {
964 os::Linux::createThread_lock()->unlock();
965 }
966 }
968 // Aborted due to thread limit being reached
969 if (state == ZOMBIE) {
970 thread->set_osthread(NULL);
971 delete osthread;
972 return false;
973 }
975 // The thread is returned suspended (in state INITIALIZED),
976 // and is started higher up in the call chain
977 assert(state == INITIALIZED, "race condition");
978 return true;
979 }
981 /////////////////////////////////////////////////////////////////////////////
982 // attach existing thread
984 // bootstrap the main thread
985 bool os::create_main_thread(JavaThread* thread) {
986 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
987 return create_attached_thread(thread);
988 }
990 bool os::create_attached_thread(JavaThread* thread) {
991 #ifdef ASSERT
992 thread->verify_not_published();
993 #endif
995 // Allocate the OSThread object
996 OSThread* osthread = new OSThread(NULL, NULL);
998 if (osthread == NULL) {
999 return false;
1000 }
1002 // Store pthread info into the OSThread
1003 osthread->set_thread_id(os::Linux::gettid());
1004 osthread->set_pthread_id(::pthread_self());
1006 // initialize floating point control register
1007 os::Linux::init_thread_fpu_state();
1009 // Initial thread state is RUNNABLE
1010 osthread->set_state(RUNNABLE);
1012 thread->set_osthread(osthread);
1014 if (UseNUMA) {
1015 int lgrp_id = os::numa_get_group_id();
1016 if (lgrp_id != -1) {
1017 thread->set_lgrp_id(lgrp_id);
1018 }
1019 }
1021 if (os::Linux::is_initial_thread()) {
1022 // If current thread is initial thread, its stack is mapped on demand,
1023 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1024 // the entire stack region to avoid SEGV in stack banging.
1025 // It is also useful to get around the heap-stack-gap problem on SuSE
1026 // kernel (see 4821821 for details). We first expand stack to the top
1027 // of yellow zone, then enable stack yellow zone (order is significant,
1028 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1029 // is no gap between the last two virtual memory regions.
1031 JavaThread *jt = (JavaThread *)thread;
1032 address addr = jt->stack_yellow_zone_base();
1033 assert(addr != NULL, "initialization problem?");
1034 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1036 osthread->set_expanding_stack();
1037 os::Linux::manually_expand_stack(jt, addr);
1038 osthread->clear_expanding_stack();
1039 }
1041 // initialize signal mask for this thread
1042 // and save the caller's signal mask
1043 os::Linux::hotspot_sigmask(thread);
1045 return true;
1046 }
1048 void os::pd_start_thread(Thread* thread) {
1049 OSThread * osthread = thread->osthread();
1050 assert(osthread->get_state() != INITIALIZED, "just checking");
1051 Monitor* sync_with_child = osthread->startThread_lock();
1052 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1053 sync_with_child->notify();
1054 }
1056 // Free Linux resources related to the OSThread
1057 void os::free_thread(OSThread* osthread) {
1058 assert(osthread != NULL, "osthread not set");
1060 if (Thread::current()->osthread() == osthread) {
1061 // Restore caller's signal mask
1062 sigset_t sigmask = osthread->caller_sigmask();
1063 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1064 }
1066 delete osthread;
1067 }
1069 //////////////////////////////////////////////////////////////////////////////
1070 // thread local storage
1072 int os::allocate_thread_local_storage() {
1073 pthread_key_t key;
1074 int rslt = pthread_key_create(&key, NULL);
1075 assert(rslt == 0, "cannot allocate thread local storage");
1076 return (int)key;
1077 }
1079 // Note: This is currently not used by VM, as we don't destroy TLS key
1080 // on VM exit.
1081 void os::free_thread_local_storage(int index) {
1082 int rslt = pthread_key_delete((pthread_key_t)index);
1083 assert(rslt == 0, "invalid index");
1084 }
1086 void os::thread_local_storage_at_put(int index, void* value) {
1087 int rslt = pthread_setspecific((pthread_key_t)index, value);
1088 assert(rslt == 0, "pthread_setspecific failed");
1089 }
1091 extern "C" Thread* get_thread() {
1092 return ThreadLocalStorage::thread();
1093 }
1095 //////////////////////////////////////////////////////////////////////////////
1096 // initial thread
1098 // Check if current thread is the initial thread, similar to Solaris thr_main.
1099 bool os::Linux::is_initial_thread(void) {
1100 char dummy;
1101 // If called before init complete, thread stack bottom will be null.
1102 // Can be called if fatal error occurs before initialization.
1103 if (initial_thread_stack_bottom() == NULL) return false;
1104 assert(initial_thread_stack_bottom() != NULL &&
1105 initial_thread_stack_size() != 0,
1106 "os::init did not locate initial thread's stack region");
1107 if ((address)&dummy >= initial_thread_stack_bottom() &&
1108 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1109 return true;
1110 else return false;
1111 }
1113 // Find the virtual memory area that contains addr
1114 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1115 FILE *fp = fopen("/proc/self/maps", "r");
1116 if (fp) {
1117 address low, high;
1118 while (!feof(fp)) {
1119 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1120 if (low <= addr && addr < high) {
1121 if (vma_low) *vma_low = low;
1122 if (vma_high) *vma_high = high;
1123 fclose (fp);
1124 return true;
1125 }
1126 }
1127 for (;;) {
1128 int ch = fgetc(fp);
1129 if (ch == EOF || ch == (int)'\n') break;
1130 }
1131 }
1132 fclose(fp);
1133 }
1134 return false;
1135 }
1137 // Locate initial thread stack. This special handling of initial thread stack
1138 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1139 // bogus value for initial thread.
1140 void os::Linux::capture_initial_stack(size_t max_size) {
1141 // stack size is the easy part, get it from RLIMIT_STACK
1142 size_t stack_size;
1143 struct rlimit rlim;
1144 getrlimit(RLIMIT_STACK, &rlim);
1145 stack_size = rlim.rlim_cur;
1147 // 6308388: a bug in ld.so will relocate its own .data section to the
1148 // lower end of primordial stack; reduce ulimit -s value a little bit
1149 // so we won't install guard page on ld.so's data section.
1150 stack_size -= 2 * page_size();
1152 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1153 // 7.1, in both cases we will get 2G in return value.
1154 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1155 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1156 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1157 // in case other parts in glibc still assumes 2M max stack size.
1158 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1159 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1160 if (stack_size > 2 * K * K IA64_ONLY(*2))
1161 stack_size = 2 * K * K IA64_ONLY(*2);
1162 // Try to figure out where the stack base (top) is. This is harder.
1163 //
1164 // When an application is started, glibc saves the initial stack pointer in
1165 // a global variable "__libc_stack_end", which is then used by system
1166 // libraries. __libc_stack_end should be pretty close to stack top. The
1167 // variable is available since the very early days. However, because it is
1168 // a private interface, it could disappear in the future.
1169 //
1170 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1171 // to __libc_stack_end, it is very close to stack top, but isn't the real
1172 // stack top. Note that /proc may not exist if VM is running as a chroot
1173 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1174 // /proc/<pid>/stat could change in the future (though unlikely).
1175 //
1176 // We try __libc_stack_end first. If that doesn't work, look for
1177 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1178 // as a hint, which should work well in most cases.
1180 uintptr_t stack_start;
1182 // try __libc_stack_end first
1183 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1184 if (p && *p) {
1185 stack_start = *p;
1186 } else {
1187 // see if we can get the start_stack field from /proc/self/stat
1188 FILE *fp;
1189 int pid;
1190 char state;
1191 int ppid;
1192 int pgrp;
1193 int session;
1194 int nr;
1195 int tpgrp;
1196 unsigned long flags;
1197 unsigned long minflt;
1198 unsigned long cminflt;
1199 unsigned long majflt;
1200 unsigned long cmajflt;
1201 unsigned long utime;
1202 unsigned long stime;
1203 long cutime;
1204 long cstime;
1205 long prio;
1206 long nice;
1207 long junk;
1208 long it_real;
1209 uintptr_t start;
1210 uintptr_t vsize;
1211 intptr_t rss;
1212 uintptr_t rsslim;
1213 uintptr_t scodes;
1214 uintptr_t ecode;
1215 int i;
1217 // Figure what the primordial thread stack base is. Code is inspired
1218 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1219 // followed by command name surrounded by parentheses, state, etc.
1220 char stat[2048];
1221 int statlen;
1223 fp = fopen("/proc/self/stat", "r");
1224 if (fp) {
1225 statlen = fread(stat, 1, 2047, fp);
1226 stat[statlen] = '\0';
1227 fclose(fp);
1229 // Skip pid and the command string. Note that we could be dealing with
1230 // weird command names, e.g. user could decide to rename java launcher
1231 // to "java 1.4.2 :)", then the stat file would look like
1232 // 1234 (java 1.4.2 :)) R ... ...
1233 // We don't really need to know the command string, just find the last
1234 // occurrence of ")" and then start parsing from there. See bug 4726580.
1235 char * s = strrchr(stat, ')');
1237 i = 0;
1238 if (s) {
1239 // Skip blank chars
1240 do s++; while (isspace(*s));
1242 #define _UFM UINTX_FORMAT
1243 #define _DFM INTX_FORMAT
1245 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1246 /* 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 */
1247 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,
1248 &state, /* 3 %c */
1249 &ppid, /* 4 %d */
1250 &pgrp, /* 5 %d */
1251 &session, /* 6 %d */
1252 &nr, /* 7 %d */
1253 &tpgrp, /* 8 %d */
1254 &flags, /* 9 %lu */
1255 &minflt, /* 10 %lu */
1256 &cminflt, /* 11 %lu */
1257 &majflt, /* 12 %lu */
1258 &cmajflt, /* 13 %lu */
1259 &utime, /* 14 %lu */
1260 &stime, /* 15 %lu */
1261 &cutime, /* 16 %ld */
1262 &cstime, /* 17 %ld */
1263 &prio, /* 18 %ld */
1264 &nice, /* 19 %ld */
1265 &junk, /* 20 %ld */
1266 &it_real, /* 21 %ld */
1267 &start, /* 22 UINTX_FORMAT */
1268 &vsize, /* 23 UINTX_FORMAT */
1269 &rss, /* 24 INTX_FORMAT */
1270 &rsslim, /* 25 UINTX_FORMAT */
1271 &scodes, /* 26 UINTX_FORMAT */
1272 &ecode, /* 27 UINTX_FORMAT */
1273 &stack_start); /* 28 UINTX_FORMAT */
1274 }
1276 #undef _UFM
1277 #undef _DFM
1279 if (i != 28 - 2) {
1280 assert(false, "Bad conversion from /proc/self/stat");
1281 // product mode - assume we are the initial thread, good luck in the
1282 // embedded case.
1283 warning("Can't detect initial thread stack location - bad conversion");
1284 stack_start = (uintptr_t) &rlim;
1285 }
1286 } else {
1287 // For some reason we can't open /proc/self/stat (for example, running on
1288 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1289 // most cases, so don't abort:
1290 warning("Can't detect initial thread stack location - no /proc/self/stat");
1291 stack_start = (uintptr_t) &rlim;
1292 }
1293 }
1295 // Now we have a pointer (stack_start) very close to the stack top, the
1296 // next thing to do is to figure out the exact location of stack top. We
1297 // can find out the virtual memory area that contains stack_start by
1298 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1299 // and its upper limit is the real stack top. (again, this would fail if
1300 // running inside chroot, because /proc may not exist.)
1302 uintptr_t stack_top;
1303 address low, high;
1304 if (find_vma((address)stack_start, &low, &high)) {
1305 // success, "high" is the true stack top. (ignore "low", because initial
1306 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1307 stack_top = (uintptr_t)high;
1308 } else {
1309 // failed, likely because /proc/self/maps does not exist
1310 warning("Can't detect initial thread stack location - find_vma failed");
1311 // best effort: stack_start is normally within a few pages below the real
1312 // stack top, use it as stack top, and reduce stack size so we won't put
1313 // guard page outside stack.
1314 stack_top = stack_start;
1315 stack_size -= 16 * page_size();
1316 }
1318 // stack_top could be partially down the page so align it
1319 stack_top = align_size_up(stack_top, page_size());
1321 if (max_size && stack_size > max_size) {
1322 _initial_thread_stack_size = max_size;
1323 } else {
1324 _initial_thread_stack_size = stack_size;
1325 }
1327 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1328 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1329 }
1331 ////////////////////////////////////////////////////////////////////////////////
1332 // time support
1334 // Time since start-up in seconds to a fine granularity.
1335 // Used by VMSelfDestructTimer and the MemProfiler.
1336 double os::elapsedTime() {
1338 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1339 }
1341 jlong os::elapsed_counter() {
1342 return javaTimeNanos() - initial_time_count;
1343 }
1345 jlong os::elapsed_frequency() {
1346 return NANOSECS_PER_SEC; // nanosecond resolution
1347 }
1349 bool os::supports_vtime() { return true; }
1350 bool os::enable_vtime() { return false; }
1351 bool os::vtime_enabled() { return false; }
1353 double os::elapsedVTime() {
1354 struct rusage usage;
1355 int retval = getrusage(RUSAGE_THREAD, &usage);
1356 if (retval == 0) {
1357 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);
1358 } else {
1359 // better than nothing, but not much
1360 return elapsedTime();
1361 }
1362 }
1364 jlong os::javaTimeMillis() {
1365 timeval time;
1366 int status = gettimeofday(&time, NULL);
1367 assert(status != -1, "linux error");
1368 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1369 }
1371 #ifndef CLOCK_MONOTONIC
1372 #define CLOCK_MONOTONIC (1)
1373 #endif
1375 void os::Linux::clock_init() {
1376 // we do dlopen's in this particular order due to bug in linux
1377 // dynamical loader (see 6348968) leading to crash on exit
1378 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1379 if (handle == NULL) {
1380 handle = dlopen("librt.so", RTLD_LAZY);
1381 }
1383 if (handle) {
1384 int (*clock_getres_func)(clockid_t, struct timespec*) =
1385 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1386 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1387 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1388 if (clock_getres_func && clock_gettime_func) {
1389 // See if monotonic clock is supported by the kernel. Note that some
1390 // early implementations simply return kernel jiffies (updated every
1391 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1392 // for nano time (though the monotonic property is still nice to have).
1393 // It's fixed in newer kernels, however clock_getres() still returns
1394 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1395 // resolution for now. Hopefully as people move to new kernels, this
1396 // won't be a problem.
1397 struct timespec res;
1398 struct timespec tp;
1399 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1400 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1401 // yes, monotonic clock is supported
1402 _clock_gettime = clock_gettime_func;
1403 return;
1404 } else {
1405 // close librt if there is no monotonic clock
1406 dlclose(handle);
1407 }
1408 }
1409 }
1410 warning("No monotonic clock was available - timed services may " \
1411 "be adversely affected if the time-of-day clock changes");
1412 }
1414 #ifndef SYS_clock_getres
1416 #if defined(IA32) || defined(AMD64)
1417 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1418 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1419 #else
1420 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1421 #define sys_clock_getres(x,y) -1
1422 #endif
1424 #else
1425 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1426 #endif
1428 void os::Linux::fast_thread_clock_init() {
1429 if (!UseLinuxPosixThreadCPUClocks) {
1430 return;
1431 }
1432 clockid_t clockid;
1433 struct timespec tp;
1434 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1435 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1437 // Switch to using fast clocks for thread cpu time if
1438 // the sys_clock_getres() returns 0 error code.
1439 // Note, that some kernels may support the current thread
1440 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1441 // returned by the pthread_getcpuclockid().
1442 // If the fast Posix clocks are supported then the sys_clock_getres()
1443 // must return at least tp.tv_sec == 0 which means a resolution
1444 // better than 1 sec. This is extra check for reliability.
1446 if(pthread_getcpuclockid_func &&
1447 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1448 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1450 _supports_fast_thread_cpu_time = true;
1451 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1452 }
1453 }
1455 jlong os::javaTimeNanos() {
1456 if (Linux::supports_monotonic_clock()) {
1457 struct timespec tp;
1458 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1459 assert(status == 0, "gettime error");
1460 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1461 return result;
1462 } else {
1463 timeval time;
1464 int status = gettimeofday(&time, NULL);
1465 assert(status != -1, "linux error");
1466 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1467 return 1000 * usecs;
1468 }
1469 }
1471 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1472 if (Linux::supports_monotonic_clock()) {
1473 info_ptr->max_value = ALL_64_BITS;
1475 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1476 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1477 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1478 } else {
1479 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1480 info_ptr->max_value = ALL_64_BITS;
1482 // gettimeofday is a real time clock so it skips
1483 info_ptr->may_skip_backward = true;
1484 info_ptr->may_skip_forward = true;
1485 }
1487 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1488 }
1490 // Return the real, user, and system times in seconds from an
1491 // arbitrary fixed point in the past.
1492 bool os::getTimesSecs(double* process_real_time,
1493 double* process_user_time,
1494 double* process_system_time) {
1495 struct tms ticks;
1496 clock_t real_ticks = times(&ticks);
1498 if (real_ticks == (clock_t) (-1)) {
1499 return false;
1500 } else {
1501 double ticks_per_second = (double) clock_tics_per_sec;
1502 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1503 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1504 *process_real_time = ((double) real_ticks) / ticks_per_second;
1506 return true;
1507 }
1508 }
1511 char * os::local_time_string(char *buf, size_t buflen) {
1512 struct tm t;
1513 time_t long_time;
1514 time(&long_time);
1515 localtime_r(&long_time, &t);
1516 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1517 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1518 t.tm_hour, t.tm_min, t.tm_sec);
1519 return buf;
1520 }
1522 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1523 return localtime_r(clock, res);
1524 }
1526 ////////////////////////////////////////////////////////////////////////////////
1527 // runtime exit support
1529 // Note: os::shutdown() might be called very early during initialization, or
1530 // called from signal handler. Before adding something to os::shutdown(), make
1531 // sure it is async-safe and can handle partially initialized VM.
1532 void os::shutdown() {
1534 // allow PerfMemory to attempt cleanup of any persistent resources
1535 perfMemory_exit();
1537 // needs to remove object in file system
1538 AttachListener::abort();
1540 // flush buffered output, finish log files
1541 ostream_abort();
1543 // Check for abort hook
1544 abort_hook_t abort_hook = Arguments::abort_hook();
1545 if (abort_hook != NULL) {
1546 abort_hook();
1547 }
1549 }
1551 // Note: os::abort() might be called very early during initialization, or
1552 // called from signal handler. Before adding something to os::abort(), make
1553 // sure it is async-safe and can handle partially initialized VM.
1554 void os::abort(bool dump_core) {
1555 os::shutdown();
1556 if (dump_core) {
1557 #ifndef PRODUCT
1558 fdStream out(defaultStream::output_fd());
1559 out.print_raw("Current thread is ");
1560 char buf[16];
1561 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1562 out.print_raw_cr(buf);
1563 out.print_raw_cr("Dumping core ...");
1564 #endif
1565 ::abort(); // dump core
1566 }
1568 ::exit(1);
1569 }
1571 // Die immediately, no exit hook, no abort hook, no cleanup.
1572 void os::die() {
1573 // _exit() on LinuxThreads only kills current thread
1574 ::abort();
1575 }
1577 // unused on linux for now.
1578 void os::set_error_file(const char *logfile) {}
1581 // This method is a copy of JDK's sysGetLastErrorString
1582 // from src/solaris/hpi/src/system_md.c
1584 size_t os::lasterror(char *buf, size_t len) {
1586 if (errno == 0) return 0;
1588 const char *s = ::strerror(errno);
1589 size_t n = ::strlen(s);
1590 if (n >= len) {
1591 n = len - 1;
1592 }
1593 ::strncpy(buf, s, n);
1594 buf[n] = '\0';
1595 return n;
1596 }
1598 intx os::current_thread_id() { return (intx)pthread_self(); }
1599 int os::current_process_id() {
1601 // Under the old linux thread library, linux gives each thread
1602 // its own process id. Because of this each thread will return
1603 // a different pid if this method were to return the result
1604 // of getpid(2). Linux provides no api that returns the pid
1605 // of the launcher thread for the vm. This implementation
1606 // returns a unique pid, the pid of the launcher thread
1607 // that starts the vm 'process'.
1609 // Under the NPTL, getpid() returns the same pid as the
1610 // launcher thread rather than a unique pid per thread.
1611 // Use gettid() if you want the old pre NPTL behaviour.
1613 // if you are looking for the result of a call to getpid() that
1614 // returns a unique pid for the calling thread, then look at the
1615 // OSThread::thread_id() method in osThread_linux.hpp file
1617 return (int)(_initial_pid ? _initial_pid : getpid());
1618 }
1620 // DLL functions
1622 const char* os::dll_file_extension() { return ".so"; }
1624 // This must be hard coded because it's the system's temporary
1625 // directory not the java application's temp directory, ala java.io.tmpdir.
1626 const char* os::get_temp_directory() { return "/tmp"; }
1628 static bool file_exists(const char* filename) {
1629 struct stat statbuf;
1630 if (filename == NULL || strlen(filename) == 0) {
1631 return false;
1632 }
1633 return os::stat(filename, &statbuf) == 0;
1634 }
1636 bool os::dll_build_name(char* buffer, size_t buflen,
1637 const char* pname, const char* fname) {
1638 bool retval = false;
1639 // Copied from libhpi
1640 const size_t pnamelen = pname ? strlen(pname) : 0;
1642 // Return error on buffer overflow.
1643 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1644 return retval;
1645 }
1647 if (pnamelen == 0) {
1648 snprintf(buffer, buflen, "lib%s.so", fname);
1649 retval = true;
1650 } else if (strchr(pname, *os::path_separator()) != NULL) {
1651 int n;
1652 char** pelements = split_path(pname, &n);
1653 if (pelements == NULL) {
1654 return false;
1655 }
1656 for (int i = 0 ; i < n ; i++) {
1657 // Really shouldn't be NULL, but check can't hurt
1658 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1659 continue; // skip the empty path values
1660 }
1661 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1662 if (file_exists(buffer)) {
1663 retval = true;
1664 break;
1665 }
1666 }
1667 // release the storage
1668 for (int i = 0 ; i < n ; i++) {
1669 if (pelements[i] != NULL) {
1670 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1671 }
1672 }
1673 if (pelements != NULL) {
1674 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1675 }
1676 } else {
1677 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1678 retval = true;
1679 }
1680 return retval;
1681 }
1683 // check if addr is inside libjvm.so
1684 bool os::address_is_in_vm(address addr) {
1685 static address libjvm_base_addr;
1686 Dl_info dlinfo;
1688 if (libjvm_base_addr == NULL) {
1689 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1690 libjvm_base_addr = (address)dlinfo.dli_fbase;
1691 }
1692 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1693 }
1695 if (dladdr((void *)addr, &dlinfo) != 0) {
1696 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1697 }
1699 return false;
1700 }
1702 bool os::dll_address_to_function_name(address addr, char *buf,
1703 int buflen, int *offset) {
1704 // buf is not optional, but offset is optional
1705 assert(buf != NULL, "sanity check");
1707 Dl_info dlinfo;
1709 if (dladdr((void*)addr, &dlinfo) != 0) {
1710 // see if we have a matching symbol
1711 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1712 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1713 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1714 }
1715 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1716 return true;
1717 }
1718 // no matching symbol so try for just file info
1719 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1720 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1721 buf, buflen, offset, dlinfo.dli_fname)) {
1722 return true;
1723 }
1724 }
1725 }
1727 buf[0] = '\0';
1728 if (offset != NULL) *offset = -1;
1729 return false;
1730 }
1732 struct _address_to_library_name {
1733 address addr; // input : memory address
1734 size_t buflen; // size of fname
1735 char* fname; // output: library name
1736 address base; // library base addr
1737 };
1739 static int address_to_library_name_callback(struct dl_phdr_info *info,
1740 size_t size, void *data) {
1741 int i;
1742 bool found = false;
1743 address libbase = NULL;
1744 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1746 // iterate through all loadable segments
1747 for (i = 0; i < info->dlpi_phnum; i++) {
1748 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1749 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1750 // base address of a library is the lowest address of its loaded
1751 // segments.
1752 if (libbase == NULL || libbase > segbase) {
1753 libbase = segbase;
1754 }
1755 // see if 'addr' is within current segment
1756 if (segbase <= d->addr &&
1757 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1758 found = true;
1759 }
1760 }
1761 }
1763 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1764 // so dll_address_to_library_name() can fall through to use dladdr() which
1765 // can figure out executable name from argv[0].
1766 if (found && info->dlpi_name && info->dlpi_name[0]) {
1767 d->base = libbase;
1768 if (d->fname) {
1769 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1770 }
1771 return 1;
1772 }
1773 return 0;
1774 }
1776 bool os::dll_address_to_library_name(address addr, char* buf,
1777 int buflen, int* offset) {
1778 // buf is not optional, but offset is optional
1779 assert(buf != NULL, "sanity check");
1781 Dl_info dlinfo;
1782 struct _address_to_library_name data;
1784 // There is a bug in old glibc dladdr() implementation that it could resolve
1785 // to wrong library name if the .so file has a base address != NULL. Here
1786 // we iterate through the program headers of all loaded libraries to find
1787 // out which library 'addr' really belongs to. This workaround can be
1788 // removed once the minimum requirement for glibc is moved to 2.3.x.
1789 data.addr = addr;
1790 data.fname = buf;
1791 data.buflen = buflen;
1792 data.base = NULL;
1793 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1795 if (rslt) {
1796 // buf already contains library name
1797 if (offset) *offset = addr - data.base;
1798 return true;
1799 }
1800 if (dladdr((void*)addr, &dlinfo) != 0) {
1801 if (dlinfo.dli_fname != NULL) {
1802 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1803 }
1804 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1805 *offset = addr - (address)dlinfo.dli_fbase;
1806 }
1807 return true;
1808 }
1810 buf[0] = '\0';
1811 if (offset) *offset = -1;
1812 return false;
1813 }
1815 // Loads .dll/.so and
1816 // in case of error it checks if .dll/.so was built for the
1817 // same architecture as Hotspot is running on
1820 // Remember the stack's state. The Linux dynamic linker will change
1821 // the stack to 'executable' at most once, so we must safepoint only once.
1822 bool os::Linux::_stack_is_executable = false;
1824 // VM operation that loads a library. This is necessary if stack protection
1825 // of the Java stacks can be lost during loading the library. If we
1826 // do not stop the Java threads, they can stack overflow before the stacks
1827 // are protected again.
1828 class VM_LinuxDllLoad: public VM_Operation {
1829 private:
1830 const char *_filename;
1831 char *_ebuf;
1832 int _ebuflen;
1833 void *_lib;
1834 public:
1835 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1836 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1837 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1838 void doit() {
1839 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1840 os::Linux::_stack_is_executable = true;
1841 }
1842 void* loaded_library() { return _lib; }
1843 };
1845 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1846 {
1847 void * result = NULL;
1848 bool load_attempted = false;
1850 // Check whether the library to load might change execution rights
1851 // of the stack. If they are changed, the protection of the stack
1852 // guard pages will be lost. We need a safepoint to fix this.
1853 //
1854 // See Linux man page execstack(8) for more info.
1855 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1856 ElfFile ef(filename);
1857 if (!ef.specifies_noexecstack()) {
1858 if (!is_init_completed()) {
1859 os::Linux::_stack_is_executable = true;
1860 // This is OK - No Java threads have been created yet, and hence no
1861 // stack guard pages to fix.
1862 //
1863 // This should happen only when you are building JDK7 using a very
1864 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1865 //
1866 // Dynamic loader will make all stacks executable after
1867 // this function returns, and will not do that again.
1868 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1869 } else {
1870 warning("You have loaded library %s which might have disabled stack guard. "
1871 "The VM will try to fix the stack guard now.\n"
1872 "It's highly recommended that you fix the library with "
1873 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1874 filename);
1876 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1877 JavaThread *jt = JavaThread::current();
1878 if (jt->thread_state() != _thread_in_native) {
1879 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1880 // that requires ExecStack. Cannot enter safe point. Let's give up.
1881 warning("Unable to fix stack guard. Giving up.");
1882 } else {
1883 if (!LoadExecStackDllInVMThread) {
1884 // This is for the case where the DLL has an static
1885 // constructor function that executes JNI code. We cannot
1886 // load such DLLs in the VMThread.
1887 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1888 }
1890 ThreadInVMfromNative tiv(jt);
1891 debug_only(VMNativeEntryWrapper vew;)
1893 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1894 VMThread::execute(&op);
1895 if (LoadExecStackDllInVMThread) {
1896 result = op.loaded_library();
1897 }
1898 load_attempted = true;
1899 }
1900 }
1901 }
1902 }
1904 if (!load_attempted) {
1905 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1906 }
1908 if (result != NULL) {
1909 // Successful loading
1910 return result;
1911 }
1913 Elf32_Ehdr elf_head;
1914 int diag_msg_max_length=ebuflen-strlen(ebuf);
1915 char* diag_msg_buf=ebuf+strlen(ebuf);
1917 if (diag_msg_max_length==0) {
1918 // No more space in ebuf for additional diagnostics message
1919 return NULL;
1920 }
1923 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1925 if (file_descriptor < 0) {
1926 // Can't open library, report dlerror() message
1927 return NULL;
1928 }
1930 bool failed_to_read_elf_head=
1931 (sizeof(elf_head)!=
1932 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1934 ::close(file_descriptor);
1935 if (failed_to_read_elf_head) {
1936 // file i/o error - report dlerror() msg
1937 return NULL;
1938 }
1940 typedef struct {
1941 Elf32_Half code; // Actual value as defined in elf.h
1942 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1943 char elf_class; // 32 or 64 bit
1944 char endianess; // MSB or LSB
1945 char* name; // String representation
1946 } arch_t;
1948 #ifndef EM_486
1949 #define EM_486 6 /* Intel 80486 */
1950 #endif
1952 static const arch_t arch_array[]={
1953 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1954 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1955 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1956 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1957 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1958 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1959 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1960 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1961 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1962 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1963 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1964 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1965 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1966 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1967 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1968 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1969 };
1971 #if (defined IA32)
1972 static Elf32_Half running_arch_code=EM_386;
1973 #elif (defined AMD64)
1974 static Elf32_Half running_arch_code=EM_X86_64;
1975 #elif (defined IA64)
1976 static Elf32_Half running_arch_code=EM_IA_64;
1977 #elif (defined __sparc) && (defined _LP64)
1978 static Elf32_Half running_arch_code=EM_SPARCV9;
1979 #elif (defined __sparc) && (!defined _LP64)
1980 static Elf32_Half running_arch_code=EM_SPARC;
1981 #elif (defined __powerpc64__)
1982 static Elf32_Half running_arch_code=EM_PPC64;
1983 #elif (defined __powerpc__)
1984 static Elf32_Half running_arch_code=EM_PPC;
1985 #elif (defined ARM)
1986 static Elf32_Half running_arch_code=EM_ARM;
1987 #elif (defined S390)
1988 static Elf32_Half running_arch_code=EM_S390;
1989 #elif (defined ALPHA)
1990 static Elf32_Half running_arch_code=EM_ALPHA;
1991 #elif (defined MIPSEL)
1992 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1993 #elif (defined PARISC)
1994 static Elf32_Half running_arch_code=EM_PARISC;
1995 #elif (defined MIPS)
1996 static Elf32_Half running_arch_code=EM_MIPS;
1997 #elif (defined M68K)
1998 static Elf32_Half running_arch_code=EM_68K;
1999 #else
2000 #error Method os::dll_load requires that one of following is defined:\
2001 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2002 #endif
2004 // Identify compatability class for VM's architecture and library's architecture
2005 // Obtain string descriptions for architectures
2007 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2008 int running_arch_index=-1;
2010 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2011 if (running_arch_code == arch_array[i].code) {
2012 running_arch_index = i;
2013 }
2014 if (lib_arch.code == arch_array[i].code) {
2015 lib_arch.compat_class = arch_array[i].compat_class;
2016 lib_arch.name = arch_array[i].name;
2017 }
2018 }
2020 assert(running_arch_index != -1,
2021 "Didn't find running architecture code (running_arch_code) in arch_array");
2022 if (running_arch_index == -1) {
2023 // Even though running architecture detection failed
2024 // we may still continue with reporting dlerror() message
2025 return NULL;
2026 }
2028 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2029 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2030 return NULL;
2031 }
2033 #ifndef S390
2034 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2035 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2036 return NULL;
2037 }
2038 #endif // !S390
2040 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2041 if ( lib_arch.name!=NULL ) {
2042 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2043 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2044 lib_arch.name, arch_array[running_arch_index].name);
2045 } else {
2046 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2047 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2048 lib_arch.code,
2049 arch_array[running_arch_index].name);
2050 }
2051 }
2053 return NULL;
2054 }
2056 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2057 void * result = ::dlopen(filename, RTLD_LAZY);
2058 if (result == NULL) {
2059 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2060 ebuf[ebuflen-1] = '\0';
2061 }
2062 return result;
2063 }
2065 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2066 void * result = NULL;
2067 if (LoadExecStackDllInVMThread) {
2068 result = dlopen_helper(filename, ebuf, ebuflen);
2069 }
2071 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2072 // library that requires an executable stack, or which does not have this
2073 // stack attribute set, dlopen changes the stack attribute to executable. The
2074 // read protection of the guard pages gets lost.
2075 //
2076 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2077 // may have been queued at the same time.
2079 if (!_stack_is_executable) {
2080 JavaThread *jt = Threads::first();
2082 while (jt) {
2083 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2084 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2085 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2086 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2087 warning("Attempt to reguard stack yellow zone failed.");
2088 }
2089 }
2090 jt = jt->next();
2091 }
2092 }
2094 return result;
2095 }
2097 /*
2098 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2099 * chances are you might want to run the generated bits against glibc-2.0
2100 * libdl.so, so always use locking for any version of glibc.
2101 */
2102 void* os::dll_lookup(void* handle, const char* name) {
2103 pthread_mutex_lock(&dl_mutex);
2104 void* res = dlsym(handle, name);
2105 pthread_mutex_unlock(&dl_mutex);
2106 return res;
2107 }
2109 void* os::get_default_process_handle() {
2110 return (void*)::dlopen(NULL, RTLD_LAZY);
2111 }
2113 static bool _print_ascii_file(const char* filename, outputStream* st) {
2114 int fd = ::open(filename, O_RDONLY);
2115 if (fd == -1) {
2116 return false;
2117 }
2119 char buf[32];
2120 int bytes;
2121 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2122 st->print_raw(buf, bytes);
2123 }
2125 ::close(fd);
2127 return true;
2128 }
2130 void os::print_dll_info(outputStream *st) {
2131 st->print_cr("Dynamic libraries:");
2133 char fname[32];
2134 pid_t pid = os::Linux::gettid();
2136 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2138 if (!_print_ascii_file(fname, st)) {
2139 st->print("Can not get library information for pid = %d\n", pid);
2140 }
2141 }
2143 void os::print_os_info_brief(outputStream* st) {
2144 os::Linux::print_distro_info(st);
2146 os::Posix::print_uname_info(st);
2148 os::Linux::print_libversion_info(st);
2150 }
2152 void os::print_os_info(outputStream* st) {
2153 st->print("OS:");
2155 os::Linux::print_distro_info(st);
2157 os::Posix::print_uname_info(st);
2159 // Print warning if unsafe chroot environment detected
2160 if (unsafe_chroot_detected) {
2161 st->print("WARNING!! ");
2162 st->print_cr(unstable_chroot_error);
2163 }
2165 os::Linux::print_libversion_info(st);
2167 os::Posix::print_rlimit_info(st);
2169 os::Posix::print_load_average(st);
2171 os::Linux::print_full_memory_info(st);
2172 }
2174 // Try to identify popular distros.
2175 // Most Linux distributions have a /etc/XXX-release file, which contains
2176 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2177 // file that also contains the OS version string. Some have more than one
2178 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2179 // /etc/redhat-release.), so the order is important.
2180 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2181 // their own specific XXX-release file as well as a redhat-release file.
2182 // Because of this the XXX-release file needs to be searched for before the
2183 // redhat-release file.
2184 // Since Red Hat has a lsb-release file that is not very descriptive the
2185 // search for redhat-release needs to be before lsb-release.
2186 // Since the lsb-release file is the new standard it needs to be searched
2187 // before the older style release files.
2188 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2189 // next to last resort. The os-release file is a new standard that contains
2190 // distribution information and the system-release file seems to be an old
2191 // standard that has been replaced by the lsb-release and os-release files.
2192 // Searching for the debian_version file is the last resort. It contains
2193 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2194 // "Debian " is printed before the contents of the debian_version file.
2195 void os::Linux::print_distro_info(outputStream* st) {
2196 if (!_print_ascii_file("/etc/oracle-release", st) &&
2197 !_print_ascii_file("/etc/mandriva-release", st) &&
2198 !_print_ascii_file("/etc/mandrake-release", st) &&
2199 !_print_ascii_file("/etc/sun-release", st) &&
2200 !_print_ascii_file("/etc/redhat-release", st) &&
2201 !_print_ascii_file("/etc/lsb-release", st) &&
2202 !_print_ascii_file("/etc/SuSE-release", st) &&
2203 !_print_ascii_file("/etc/turbolinux-release", st) &&
2204 !_print_ascii_file("/etc/gentoo-release", st) &&
2205 !_print_ascii_file("/etc/ltib-release", st) &&
2206 !_print_ascii_file("/etc/angstrom-version", st) &&
2207 !_print_ascii_file("/etc/system-release", st) &&
2208 !_print_ascii_file("/etc/os-release", st)) {
2210 if (file_exists("/etc/debian_version")) {
2211 st->print("Debian ");
2212 _print_ascii_file("/etc/debian_version", st);
2213 } else {
2214 st->print("Linux");
2215 }
2216 }
2217 st->cr();
2218 }
2220 void os::Linux::print_libversion_info(outputStream* st) {
2221 // libc, pthread
2222 st->print("libc:");
2223 st->print(os::Linux::glibc_version()); st->print(" ");
2224 st->print(os::Linux::libpthread_version()); st->print(" ");
2225 if (os::Linux::is_LinuxThreads()) {
2226 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2227 }
2228 st->cr();
2229 }
2231 void os::Linux::print_full_memory_info(outputStream* st) {
2232 st->print("\n/proc/meminfo:\n");
2233 _print_ascii_file("/proc/meminfo", st);
2234 st->cr();
2235 }
2237 void os::print_memory_info(outputStream* st) {
2239 st->print("Memory:");
2240 st->print(" %dk page", os::vm_page_size()>>10);
2242 // values in struct sysinfo are "unsigned long"
2243 struct sysinfo si;
2244 sysinfo(&si);
2246 st->print(", physical " UINT64_FORMAT "k",
2247 os::physical_memory() >> 10);
2248 st->print("(" UINT64_FORMAT "k free)",
2249 os::available_memory() >> 10);
2250 st->print(", swap " UINT64_FORMAT "k",
2251 ((jlong)si.totalswap * si.mem_unit) >> 10);
2252 st->print("(" UINT64_FORMAT "k free)",
2253 ((jlong)si.freeswap * si.mem_unit) >> 10);
2254 st->cr();
2255 }
2257 void os::pd_print_cpu_info(outputStream* st) {
2258 st->print("\n/proc/cpuinfo:\n");
2259 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2260 st->print(" <Not Available>");
2261 }
2262 st->cr();
2263 }
2265 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2266 // but they're the same for all the linux arch that we support
2267 // and they're the same for solaris but there's no common place to put this.
2268 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2269 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2270 "ILL_COPROC", "ILL_BADSTK" };
2272 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2273 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2274 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2276 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2278 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2280 void os::print_siginfo(outputStream* st, void* siginfo) {
2281 st->print("siginfo:");
2283 const int buflen = 100;
2284 char buf[buflen];
2285 siginfo_t *si = (siginfo_t*)siginfo;
2286 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2287 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2288 st->print("si_errno=%s", buf);
2289 } else {
2290 st->print("si_errno=%d", si->si_errno);
2291 }
2292 const int c = si->si_code;
2293 assert(c > 0, "unexpected si_code");
2294 switch (si->si_signo) {
2295 case SIGILL:
2296 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2297 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2298 break;
2299 case SIGFPE:
2300 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2301 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2302 break;
2303 case SIGSEGV:
2304 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2305 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2306 break;
2307 case SIGBUS:
2308 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2309 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2310 break;
2311 default:
2312 st->print(", si_code=%d", si->si_code);
2313 // no si_addr
2314 }
2316 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2317 UseSharedSpaces) {
2318 FileMapInfo* mapinfo = FileMapInfo::current_info();
2319 if (mapinfo->is_in_shared_space(si->si_addr)) {
2320 st->print("\n\nError accessing class data sharing archive." \
2321 " Mapped file inaccessible during execution, " \
2322 " possible disk/network problem.");
2323 }
2324 }
2325 st->cr();
2326 }
2329 static void print_signal_handler(outputStream* st, int sig,
2330 char* buf, size_t buflen);
2332 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2333 st->print_cr("Signal Handlers:");
2334 print_signal_handler(st, SIGSEGV, buf, buflen);
2335 print_signal_handler(st, SIGBUS , buf, buflen);
2336 print_signal_handler(st, SIGFPE , buf, buflen);
2337 print_signal_handler(st, SIGPIPE, buf, buflen);
2338 print_signal_handler(st, SIGXFSZ, buf, buflen);
2339 print_signal_handler(st, SIGILL , buf, buflen);
2340 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2341 print_signal_handler(st, SR_signum, buf, buflen);
2342 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2343 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2344 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2345 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2346 }
2348 static char saved_jvm_path[MAXPATHLEN] = {0};
2350 // Find the full path to the current module, libjvm.so
2351 void os::jvm_path(char *buf, jint buflen) {
2352 // Error checking.
2353 if (buflen < MAXPATHLEN) {
2354 assert(false, "must use a large-enough buffer");
2355 buf[0] = '\0';
2356 return;
2357 }
2358 // Lazy resolve the path to current module.
2359 if (saved_jvm_path[0] != 0) {
2360 strcpy(buf, saved_jvm_path);
2361 return;
2362 }
2364 char dli_fname[MAXPATHLEN];
2365 bool ret = dll_address_to_library_name(
2366 CAST_FROM_FN_PTR(address, os::jvm_path),
2367 dli_fname, sizeof(dli_fname), NULL);
2368 assert(ret, "cannot locate libjvm");
2369 char *rp = NULL;
2370 if (ret && dli_fname[0] != '\0') {
2371 rp = realpath(dli_fname, buf);
2372 }
2373 if (rp == NULL)
2374 return;
2376 if (Arguments::created_by_gamma_launcher()) {
2377 // Support for the gamma launcher. Typical value for buf is
2378 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2379 // the right place in the string, then assume we are installed in a JDK and
2380 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2381 // up the path so it looks like libjvm.so is installed there (append a
2382 // fake suffix hotspot/libjvm.so).
2383 const char *p = buf + strlen(buf) - 1;
2384 for (int count = 0; p > buf && count < 5; ++count) {
2385 for (--p; p > buf && *p != '/'; --p)
2386 /* empty */ ;
2387 }
2389 if (strncmp(p, "/jre/lib/", 9) != 0) {
2390 // Look for JAVA_HOME in the environment.
2391 char* java_home_var = ::getenv("JAVA_HOME");
2392 if (java_home_var != NULL && java_home_var[0] != 0) {
2393 char* jrelib_p;
2394 int len;
2396 // Check the current module name "libjvm.so".
2397 p = strrchr(buf, '/');
2398 assert(strstr(p, "/libjvm") == p, "invalid library name");
2400 rp = realpath(java_home_var, buf);
2401 if (rp == NULL)
2402 return;
2404 // determine if this is a legacy image or modules image
2405 // modules image doesn't have "jre" subdirectory
2406 len = strlen(buf);
2407 jrelib_p = buf + len;
2408 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2409 if (0 != access(buf, F_OK)) {
2410 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2411 }
2413 if (0 == access(buf, F_OK)) {
2414 // Use current module name "libjvm.so"
2415 len = strlen(buf);
2416 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2417 } else {
2418 // Go back to path of .so
2419 rp = realpath(dli_fname, buf);
2420 if (rp == NULL)
2421 return;
2422 }
2423 }
2424 }
2425 }
2427 strcpy(saved_jvm_path, buf);
2428 }
2430 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2431 // no prefix required, not even "_"
2432 }
2434 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2435 // no suffix required
2436 }
2438 ////////////////////////////////////////////////////////////////////////////////
2439 // sun.misc.Signal support
2441 static volatile jint sigint_count = 0;
2443 static void
2444 UserHandler(int sig, void *siginfo, void *context) {
2445 // 4511530 - sem_post is serialized and handled by the manager thread. When
2446 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2447 // don't want to flood the manager thread with sem_post requests.
2448 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2449 return;
2451 // Ctrl-C is pressed during error reporting, likely because the error
2452 // handler fails to abort. Let VM die immediately.
2453 if (sig == SIGINT && is_error_reported()) {
2454 os::die();
2455 }
2457 os::signal_notify(sig);
2458 }
2460 void* os::user_handler() {
2461 return CAST_FROM_FN_PTR(void*, UserHandler);
2462 }
2464 class Semaphore : public StackObj {
2465 public:
2466 Semaphore();
2467 ~Semaphore();
2468 void signal();
2469 void wait();
2470 bool trywait();
2471 bool timedwait(unsigned int sec, int nsec);
2472 private:
2473 sem_t _semaphore;
2474 };
2476 Semaphore::Semaphore() {
2477 sem_init(&_semaphore, 0, 0);
2478 }
2480 Semaphore::~Semaphore() {
2481 sem_destroy(&_semaphore);
2482 }
2484 void Semaphore::signal() {
2485 sem_post(&_semaphore);
2486 }
2488 void Semaphore::wait() {
2489 sem_wait(&_semaphore);
2490 }
2492 bool Semaphore::trywait() {
2493 return sem_trywait(&_semaphore) == 0;
2494 }
2496 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2498 struct timespec ts;
2499 // Semaphore's are always associated with CLOCK_REALTIME
2500 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2501 // see unpackTime for discussion on overflow checking
2502 if (sec >= MAX_SECS) {
2503 ts.tv_sec += MAX_SECS;
2504 ts.tv_nsec = 0;
2505 } else {
2506 ts.tv_sec += sec;
2507 ts.tv_nsec += nsec;
2508 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2509 ts.tv_nsec -= NANOSECS_PER_SEC;
2510 ++ts.tv_sec; // note: this must be <= max_secs
2511 }
2512 }
2514 while (1) {
2515 int result = sem_timedwait(&_semaphore, &ts);
2516 if (result == 0) {
2517 return true;
2518 } else if (errno == EINTR) {
2519 continue;
2520 } else if (errno == ETIMEDOUT) {
2521 return false;
2522 } else {
2523 return false;
2524 }
2525 }
2526 }
2528 extern "C" {
2529 typedef void (*sa_handler_t)(int);
2530 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2531 }
2533 void* os::signal(int signal_number, void* handler) {
2534 struct sigaction sigAct, oldSigAct;
2536 sigfillset(&(sigAct.sa_mask));
2537 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2538 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2540 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2541 // -1 means registration failed
2542 return (void *)-1;
2543 }
2545 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2546 }
2548 void os::signal_raise(int signal_number) {
2549 ::raise(signal_number);
2550 }
2552 /*
2553 * The following code is moved from os.cpp for making this
2554 * code platform specific, which it is by its very nature.
2555 */
2557 // Will be modified when max signal is changed to be dynamic
2558 int os::sigexitnum_pd() {
2559 return NSIG;
2560 }
2562 // a counter for each possible signal value
2563 static volatile jint pending_signals[NSIG+1] = { 0 };
2565 // Linux(POSIX) specific hand shaking semaphore.
2566 static sem_t sig_sem;
2567 static Semaphore sr_semaphore;
2569 void os::signal_init_pd() {
2570 // Initialize signal structures
2571 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2573 // Initialize signal semaphore
2574 ::sem_init(&sig_sem, 0, 0);
2575 }
2577 void os::signal_notify(int sig) {
2578 Atomic::inc(&pending_signals[sig]);
2579 ::sem_post(&sig_sem);
2580 }
2582 static int check_pending_signals(bool wait) {
2583 Atomic::store(0, &sigint_count);
2584 for (;;) {
2585 for (int i = 0; i < NSIG + 1; i++) {
2586 jint n = pending_signals[i];
2587 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2588 return i;
2589 }
2590 }
2591 if (!wait) {
2592 return -1;
2593 }
2594 JavaThread *thread = JavaThread::current();
2595 ThreadBlockInVM tbivm(thread);
2597 bool threadIsSuspended;
2598 do {
2599 thread->set_suspend_equivalent();
2600 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2601 ::sem_wait(&sig_sem);
2603 // were we externally suspended while we were waiting?
2604 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2605 if (threadIsSuspended) {
2606 //
2607 // The semaphore has been incremented, but while we were waiting
2608 // another thread suspended us. We don't want to continue running
2609 // while suspended because that would surprise the thread that
2610 // suspended us.
2611 //
2612 ::sem_post(&sig_sem);
2614 thread->java_suspend_self();
2615 }
2616 } while (threadIsSuspended);
2617 }
2618 }
2620 int os::signal_lookup() {
2621 return check_pending_signals(false);
2622 }
2624 int os::signal_wait() {
2625 return check_pending_signals(true);
2626 }
2628 ////////////////////////////////////////////////////////////////////////////////
2629 // Virtual Memory
2631 int os::vm_page_size() {
2632 // Seems redundant as all get out
2633 assert(os::Linux::page_size() != -1, "must call os::init");
2634 return os::Linux::page_size();
2635 }
2637 // Solaris allocates memory by pages.
2638 int os::vm_allocation_granularity() {
2639 assert(os::Linux::page_size() != -1, "must call os::init");
2640 return os::Linux::page_size();
2641 }
2643 // Rationale behind this function:
2644 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2645 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2646 // samples for JITted code. Here we create private executable mapping over the code cache
2647 // and then we can use standard (well, almost, as mapping can change) way to provide
2648 // info for the reporting script by storing timestamp and location of symbol
2649 void linux_wrap_code(char* base, size_t size) {
2650 static volatile jint cnt = 0;
2652 if (!UseOprofile) {
2653 return;
2654 }
2656 char buf[PATH_MAX+1];
2657 int num = Atomic::add(1, &cnt);
2659 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2660 os::get_temp_directory(), os::current_process_id(), num);
2661 unlink(buf);
2663 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2665 if (fd != -1) {
2666 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2667 if (rv != (off_t)-1) {
2668 if (::write(fd, "", 1) == 1) {
2669 mmap(base, size,
2670 PROT_READ|PROT_WRITE|PROT_EXEC,
2671 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2672 }
2673 }
2674 ::close(fd);
2675 unlink(buf);
2676 }
2677 }
2679 static bool recoverable_mmap_error(int err) {
2680 // See if the error is one we can let the caller handle. This
2681 // list of errno values comes from JBS-6843484. I can't find a
2682 // Linux man page that documents this specific set of errno
2683 // values so while this list currently matches Solaris, it may
2684 // change as we gain experience with this failure mode.
2685 switch (err) {
2686 case EBADF:
2687 case EINVAL:
2688 case ENOTSUP:
2689 // let the caller deal with these errors
2690 return true;
2692 default:
2693 // Any remaining errors on this OS can cause our reserved mapping
2694 // to be lost. That can cause confusion where different data
2695 // structures think they have the same memory mapped. The worst
2696 // scenario is if both the VM and a library think they have the
2697 // same memory mapped.
2698 return false;
2699 }
2700 }
2702 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2703 int err) {
2704 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2705 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2706 strerror(err), err);
2707 }
2709 static void warn_fail_commit_memory(char* addr, size_t size,
2710 size_t alignment_hint, bool exec,
2711 int err) {
2712 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2713 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2714 alignment_hint, exec, strerror(err), err);
2715 }
2717 // NOTE: Linux kernel does not really reserve the pages for us.
2718 // All it does is to check if there are enough free pages
2719 // left at the time of mmap(). This could be a potential
2720 // problem.
2721 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2722 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2723 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2724 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2725 if (res != (uintptr_t) MAP_FAILED) {
2726 if (UseNUMAInterleaving) {
2727 numa_make_global(addr, size);
2728 }
2729 return 0;
2730 }
2732 int err = errno; // save errno from mmap() call above
2734 if (!recoverable_mmap_error(err)) {
2735 warn_fail_commit_memory(addr, size, exec, err);
2736 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2737 }
2739 return err;
2740 }
2742 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2743 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2744 }
2746 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2747 const char* mesg) {
2748 assert(mesg != NULL, "mesg must be specified");
2749 int err = os::Linux::commit_memory_impl(addr, size, exec);
2750 if (err != 0) {
2751 // the caller wants all commit errors to exit with the specified mesg:
2752 warn_fail_commit_memory(addr, size, exec, err);
2753 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2754 }
2755 }
2757 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2758 #ifndef MAP_HUGETLB
2759 #define MAP_HUGETLB 0x40000
2760 #endif
2762 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2763 #ifndef MADV_HUGEPAGE
2764 #define MADV_HUGEPAGE 14
2765 #endif
2767 int os::Linux::commit_memory_impl(char* addr, size_t size,
2768 size_t alignment_hint, bool exec) {
2769 int err = os::Linux::commit_memory_impl(addr, size, exec);
2770 if (err == 0) {
2771 realign_memory(addr, size, alignment_hint);
2772 }
2773 return err;
2774 }
2776 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2777 bool exec) {
2778 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2779 }
2781 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2782 size_t alignment_hint, bool exec,
2783 const char* mesg) {
2784 assert(mesg != NULL, "mesg must be specified");
2785 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2786 if (err != 0) {
2787 // the caller wants all commit errors to exit with the specified mesg:
2788 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2789 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2790 }
2791 }
2793 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2794 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2795 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2796 // be supported or the memory may already be backed by huge pages.
2797 ::madvise(addr, bytes, MADV_HUGEPAGE);
2798 }
2799 }
2801 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2802 // This method works by doing an mmap over an existing mmaping and effectively discarding
2803 // the existing pages. However it won't work for SHM-based large pages that cannot be
2804 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2805 // small pages on top of the SHM segment. This method always works for small pages, so we
2806 // allow that in any case.
2807 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2808 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2809 }
2810 }
2812 void os::numa_make_global(char *addr, size_t bytes) {
2813 Linux::numa_interleave_memory(addr, bytes);
2814 }
2816 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2817 // bind policy to MPOL_PREFERRED for the current thread.
2818 #define USE_MPOL_PREFERRED 0
2820 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2821 // To make NUMA and large pages more robust when both enabled, we need to ease
2822 // the requirements on where the memory should be allocated. MPOL_BIND is the
2823 // default policy and it will force memory to be allocated on the specified
2824 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2825 // the specified node, but will not force it. Using this policy will prevent
2826 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2827 // free large pages.
2828 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2829 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2830 }
2832 bool os::numa_topology_changed() { return false; }
2834 size_t os::numa_get_groups_num() {
2835 int max_node = Linux::numa_max_node();
2836 return max_node > 0 ? max_node + 1 : 1;
2837 }
2839 int os::numa_get_group_id() {
2840 int cpu_id = Linux::sched_getcpu();
2841 if (cpu_id != -1) {
2842 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2843 if (lgrp_id != -1) {
2844 return lgrp_id;
2845 }
2846 }
2847 return 0;
2848 }
2850 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2851 for (size_t i = 0; i < size; i++) {
2852 ids[i] = i;
2853 }
2854 return size;
2855 }
2857 bool os::get_page_info(char *start, page_info* info) {
2858 return false;
2859 }
2861 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2862 return end;
2863 }
2866 int os::Linux::sched_getcpu_syscall(void) {
2867 unsigned int cpu;
2868 int retval = -1;
2870 #if defined(IA32)
2871 # ifndef SYS_getcpu
2872 # define SYS_getcpu 318
2873 # endif
2874 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2875 #elif defined(AMD64)
2876 // Unfortunately we have to bring all these macros here from vsyscall.h
2877 // to be able to compile on old linuxes.
2878 # define __NR_vgetcpu 2
2879 # define VSYSCALL_START (-10UL << 20)
2880 # define VSYSCALL_SIZE 1024
2881 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2882 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2883 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2884 retval = vgetcpu(&cpu, NULL, NULL);
2885 #endif
2887 return (retval == -1) ? retval : cpu;
2888 }
2890 // Something to do with the numa-aware allocator needs these symbols
2891 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2892 extern "C" JNIEXPORT void numa_error(char *where) { }
2893 extern "C" JNIEXPORT int fork1() { return fork(); }
2896 // If we are running with libnuma version > 2, then we should
2897 // be trying to use symbols with versions 1.1
2898 // If we are running with earlier version, which did not have symbol versions,
2899 // we should use the base version.
2900 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2901 void *f = dlvsym(handle, name, "libnuma_1.1");
2902 if (f == NULL) {
2903 f = dlsym(handle, name);
2904 }
2905 return f;
2906 }
2908 bool os::Linux::libnuma_init() {
2909 // sched_getcpu() should be in libc.
2910 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2911 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2913 // If it's not, try a direct syscall.
2914 if (sched_getcpu() == -1)
2915 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2917 if (sched_getcpu() != -1) { // Does it work?
2918 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2919 if (handle != NULL) {
2920 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2921 libnuma_dlsym(handle, "numa_node_to_cpus")));
2922 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2923 libnuma_dlsym(handle, "numa_max_node")));
2924 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2925 libnuma_dlsym(handle, "numa_available")));
2926 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2927 libnuma_dlsym(handle, "numa_tonode_memory")));
2928 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2929 libnuma_dlsym(handle, "numa_interleave_memory")));
2930 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2931 libnuma_dlsym(handle, "numa_set_bind_policy")));
2934 if (numa_available() != -1) {
2935 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2936 // Create a cpu -> node mapping
2937 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2938 rebuild_cpu_to_node_map();
2939 return true;
2940 }
2941 }
2942 }
2943 return false;
2944 }
2946 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2947 // The table is later used in get_node_by_cpu().
2948 void os::Linux::rebuild_cpu_to_node_map() {
2949 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2950 // in libnuma (possible values are starting from 16,
2951 // and continuing up with every other power of 2, but less
2952 // than the maximum number of CPUs supported by kernel), and
2953 // is a subject to change (in libnuma version 2 the requirements
2954 // are more reasonable) we'll just hardcode the number they use
2955 // in the library.
2956 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2958 size_t cpu_num = os::active_processor_count();
2959 size_t cpu_map_size = NCPUS / BitsPerCLong;
2960 size_t cpu_map_valid_size =
2961 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2963 cpu_to_node()->clear();
2964 cpu_to_node()->at_grow(cpu_num - 1);
2965 size_t node_num = numa_get_groups_num();
2967 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2968 for (size_t i = 0; i < node_num; i++) {
2969 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2970 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2971 if (cpu_map[j] != 0) {
2972 for (size_t k = 0; k < BitsPerCLong; k++) {
2973 if (cpu_map[j] & (1UL << k)) {
2974 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2975 }
2976 }
2977 }
2978 }
2979 }
2980 }
2981 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2982 }
2984 int os::Linux::get_node_by_cpu(int cpu_id) {
2985 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2986 return cpu_to_node()->at(cpu_id);
2987 }
2988 return -1;
2989 }
2991 GrowableArray<int>* os::Linux::_cpu_to_node;
2992 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2993 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2994 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2995 os::Linux::numa_available_func_t os::Linux::_numa_available;
2996 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2997 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2998 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2999 unsigned long* os::Linux::_numa_all_nodes;
3001 bool os::pd_uncommit_memory(char* addr, size_t size) {
3002 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3003 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3004 return res != (uintptr_t) MAP_FAILED;
3005 }
3007 static
3008 address get_stack_commited_bottom(address bottom, size_t size) {
3009 address nbot = bottom;
3010 address ntop = bottom + size;
3012 size_t page_sz = os::vm_page_size();
3013 unsigned pages = size / page_sz;
3015 unsigned char vec[1];
3016 unsigned imin = 1, imax = pages + 1, imid;
3017 int mincore_return_value = 0;
3019 assert(imin <= imax, "Unexpected page size");
3021 while (imin < imax) {
3022 imid = (imax + imin) / 2;
3023 nbot = ntop - (imid * page_sz);
3025 // Use a trick with mincore to check whether the page is mapped or not.
3026 // mincore sets vec to 1 if page resides in memory and to 0 if page
3027 // is swapped output but if page we are asking for is unmapped
3028 // it returns -1,ENOMEM
3029 mincore_return_value = mincore(nbot, page_sz, vec);
3031 if (mincore_return_value == -1) {
3032 // Page is not mapped go up
3033 // to find first mapped page
3034 if (errno != EAGAIN) {
3035 assert(errno == ENOMEM, "Unexpected mincore errno");
3036 imax = imid;
3037 }
3038 } else {
3039 // Page is mapped go down
3040 // to find first not mapped page
3041 imin = imid + 1;
3042 }
3043 }
3045 nbot = nbot + page_sz;
3047 // Adjust stack bottom one page up if last checked page is not mapped
3048 if (mincore_return_value == -1) {
3049 nbot = nbot + page_sz;
3050 }
3052 return nbot;
3053 }
3056 // Linux uses a growable mapping for the stack, and if the mapping for
3057 // the stack guard pages is not removed when we detach a thread the
3058 // stack cannot grow beyond the pages where the stack guard was
3059 // mapped. If at some point later in the process the stack expands to
3060 // that point, the Linux kernel cannot expand the stack any further
3061 // because the guard pages are in the way, and a segfault occurs.
3062 //
3063 // However, it's essential not to split the stack region by unmapping
3064 // a region (leaving a hole) that's already part of the stack mapping,
3065 // so if the stack mapping has already grown beyond the guard pages at
3066 // the time we create them, we have to truncate the stack mapping.
3067 // So, we need to know the extent of the stack mapping when
3068 // create_stack_guard_pages() is called.
3070 // We only need this for stacks that are growable: at the time of
3071 // writing thread stacks don't use growable mappings (i.e. those
3072 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3073 // only applies to the main thread.
3075 // If the (growable) stack mapping already extends beyond the point
3076 // where we're going to put our guard pages, truncate the mapping at
3077 // that point by munmap()ping it. This ensures that when we later
3078 // munmap() the guard pages we don't leave a hole in the stack
3079 // mapping. This only affects the main/initial thread
3081 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3083 if (os::Linux::is_initial_thread()) {
3084 // As we manually grow stack up to bottom inside create_attached_thread(),
3085 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3086 // we don't need to do anything special.
3087 // Check it first, before calling heavy function.
3088 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3089 unsigned char vec[1];
3091 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3092 // Fallback to slow path on all errors, including EAGAIN
3093 stack_extent = (uintptr_t) get_stack_commited_bottom(
3094 os::Linux::initial_thread_stack_bottom(),
3095 (size_t)addr - stack_extent);
3096 }
3098 if (stack_extent < (uintptr_t)addr) {
3099 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3100 }
3101 }
3103 return os::commit_memory(addr, size, !ExecMem);
3104 }
3106 // If this is a growable mapping, remove the guard pages entirely by
3107 // munmap()ping them. If not, just call uncommit_memory(). This only
3108 // affects the main/initial thread, but guard against future OS changes
3109 // It's safe to always unmap guard pages for initial thread because we
3110 // always place it right after end of the mapped region
3112 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3113 uintptr_t stack_extent, stack_base;
3115 if (os::Linux::is_initial_thread()) {
3116 return ::munmap(addr, size) == 0;
3117 }
3119 return os::uncommit_memory(addr, size);
3120 }
3122 static address _highest_vm_reserved_address = NULL;
3124 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3125 // at 'requested_addr'. If there are existing memory mappings at the same
3126 // location, however, they will be overwritten. If 'fixed' is false,
3127 // 'requested_addr' is only treated as a hint, the return value may or
3128 // may not start from the requested address. Unlike Linux mmap(), this
3129 // function returns NULL to indicate failure.
3130 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3131 char * addr;
3132 int flags;
3134 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3135 if (fixed) {
3136 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3137 flags |= MAP_FIXED;
3138 }
3140 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3141 // touch an uncommitted page. Otherwise, the read/write might
3142 // succeed if we have enough swap space to back the physical page.
3143 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3144 flags, -1, 0);
3146 if (addr != MAP_FAILED) {
3147 // anon_mmap() should only get called during VM initialization,
3148 // don't need lock (actually we can skip locking even it can be called
3149 // from multiple threads, because _highest_vm_reserved_address is just a
3150 // hint about the upper limit of non-stack memory regions.)
3151 if ((address)addr + bytes > _highest_vm_reserved_address) {
3152 _highest_vm_reserved_address = (address)addr + bytes;
3153 }
3154 }
3156 return addr == MAP_FAILED ? NULL : addr;
3157 }
3159 // Don't update _highest_vm_reserved_address, because there might be memory
3160 // regions above addr + size. If so, releasing a memory region only creates
3161 // a hole in the address space, it doesn't help prevent heap-stack collision.
3162 //
3163 static int anon_munmap(char * addr, size_t size) {
3164 return ::munmap(addr, size) == 0;
3165 }
3167 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3168 size_t alignment_hint) {
3169 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3170 }
3172 bool os::pd_release_memory(char* addr, size_t size) {
3173 return anon_munmap(addr, size);
3174 }
3176 static address highest_vm_reserved_address() {
3177 return _highest_vm_reserved_address;
3178 }
3180 static bool linux_mprotect(char* addr, size_t size, int prot) {
3181 // Linux wants the mprotect address argument to be page aligned.
3182 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3184 // According to SUSv3, mprotect() should only be used with mappings
3185 // established by mmap(), and mmap() always maps whole pages. Unaligned
3186 // 'addr' likely indicates problem in the VM (e.g. trying to change
3187 // protection of malloc'ed or statically allocated memory). Check the
3188 // caller if you hit this assert.
3189 assert(addr == bottom, "sanity check");
3191 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3192 return ::mprotect(bottom, size, prot) == 0;
3193 }
3195 // Set protections specified
3196 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3197 bool is_committed) {
3198 unsigned int p = 0;
3199 switch (prot) {
3200 case MEM_PROT_NONE: p = PROT_NONE; break;
3201 case MEM_PROT_READ: p = PROT_READ; break;
3202 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3203 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3204 default:
3205 ShouldNotReachHere();
3206 }
3207 // is_committed is unused.
3208 return linux_mprotect(addr, bytes, p);
3209 }
3211 bool os::guard_memory(char* addr, size_t size) {
3212 return linux_mprotect(addr, size, PROT_NONE);
3213 }
3215 bool os::unguard_memory(char* addr, size_t size) {
3216 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3217 }
3219 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3220 bool result = false;
3221 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3222 MAP_ANONYMOUS|MAP_PRIVATE,
3223 -1, 0);
3224 if (p != MAP_FAILED) {
3225 void *aligned_p = align_ptr_up(p, page_size);
3227 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3229 munmap(p, page_size * 2);
3230 }
3232 if (warn && !result) {
3233 warning("TransparentHugePages is not supported by the operating system.");
3234 }
3236 return result;
3237 }
3239 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3240 bool result = false;
3241 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3242 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3243 -1, 0);
3245 if (p != MAP_FAILED) {
3246 // We don't know if this really is a huge page or not.
3247 FILE *fp = fopen("/proc/self/maps", "r");
3248 if (fp) {
3249 while (!feof(fp)) {
3250 char chars[257];
3251 long x = 0;
3252 if (fgets(chars, sizeof(chars), fp)) {
3253 if (sscanf(chars, "%lx-%*x", &x) == 1
3254 && x == (long)p) {
3255 if (strstr (chars, "hugepage")) {
3256 result = true;
3257 break;
3258 }
3259 }
3260 }
3261 }
3262 fclose(fp);
3263 }
3264 munmap(p, page_size);
3265 }
3267 if (warn && !result) {
3268 warning("HugeTLBFS is not supported by the operating system.");
3269 }
3271 return result;
3272 }
3274 /*
3275 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3276 *
3277 * From the coredump_filter documentation:
3278 *
3279 * - (bit 0) anonymous private memory
3280 * - (bit 1) anonymous shared memory
3281 * - (bit 2) file-backed private memory
3282 * - (bit 3) file-backed shared memory
3283 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3284 * effective only if the bit 2 is cleared)
3285 * - (bit 5) hugetlb private memory
3286 * - (bit 6) hugetlb shared memory
3287 */
3288 static void set_coredump_filter(void) {
3289 FILE *f;
3290 long cdm;
3292 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3293 return;
3294 }
3296 if (fscanf(f, "%lx", &cdm) != 1) {
3297 fclose(f);
3298 return;
3299 }
3301 rewind(f);
3303 if ((cdm & LARGEPAGES_BIT) == 0) {
3304 cdm |= LARGEPAGES_BIT;
3305 fprintf(f, "%#lx", cdm);
3306 }
3308 fclose(f);
3309 }
3311 // Large page support
3313 static size_t _large_page_size = 0;
3315 size_t os::Linux::find_large_page_size() {
3316 size_t large_page_size = 0;
3318 // large_page_size on Linux is used to round up heap size. x86 uses either
3319 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3320 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3321 // page as large as 256M.
3322 //
3323 // Here we try to figure out page size by parsing /proc/meminfo and looking
3324 // for a line with the following format:
3325 // Hugepagesize: 2048 kB
3326 //
3327 // If we can't determine the value (e.g. /proc is not mounted, or the text
3328 // format has been changed), we'll use the largest page size supported by
3329 // the processor.
3331 #ifndef ZERO
3332 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3333 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3334 #endif // ZERO
3336 FILE *fp = fopen("/proc/meminfo", "r");
3337 if (fp) {
3338 while (!feof(fp)) {
3339 int x = 0;
3340 char buf[16];
3341 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3342 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3343 large_page_size = x * K;
3344 break;
3345 }
3346 } else {
3347 // skip to next line
3348 for (;;) {
3349 int ch = fgetc(fp);
3350 if (ch == EOF || ch == (int)'\n') break;
3351 }
3352 }
3353 }
3354 fclose(fp);
3355 }
3357 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3358 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3359 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3360 proper_unit_for_byte_size(large_page_size));
3361 }
3363 return large_page_size;
3364 }
3366 size_t os::Linux::setup_large_page_size() {
3367 _large_page_size = Linux::find_large_page_size();
3368 const size_t default_page_size = (size_t)Linux::page_size();
3369 if (_large_page_size > default_page_size) {
3370 _page_sizes[0] = _large_page_size;
3371 _page_sizes[1] = default_page_size;
3372 _page_sizes[2] = 0;
3373 }
3375 return _large_page_size;
3376 }
3378 bool os::Linux::setup_large_page_type(size_t page_size) {
3379 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3380 FLAG_IS_DEFAULT(UseSHM) &&
3381 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3383 // The type of large pages has not been specified by the user.
3385 // Try UseHugeTLBFS and then UseSHM.
3386 UseHugeTLBFS = UseSHM = true;
3388 // Don't try UseTransparentHugePages since there are known
3389 // performance issues with it turned on. This might change in the future.
3390 UseTransparentHugePages = false;
3391 }
3393 if (UseTransparentHugePages) {
3394 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3395 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3396 UseHugeTLBFS = false;
3397 UseSHM = false;
3398 return true;
3399 }
3400 UseTransparentHugePages = false;
3401 }
3403 if (UseHugeTLBFS) {
3404 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3405 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3406 UseSHM = false;
3407 return true;
3408 }
3409 UseHugeTLBFS = false;
3410 }
3412 return UseSHM;
3413 }
3415 void os::large_page_init() {
3416 if (!UseLargePages &&
3417 !UseTransparentHugePages &&
3418 !UseHugeTLBFS &&
3419 !UseSHM) {
3420 // Not using large pages.
3421 return;
3422 }
3424 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3425 // The user explicitly turned off large pages.
3426 // Ignore the rest of the large pages flags.
3427 UseTransparentHugePages = false;
3428 UseHugeTLBFS = false;
3429 UseSHM = false;
3430 return;
3431 }
3433 size_t large_page_size = Linux::setup_large_page_size();
3434 UseLargePages = Linux::setup_large_page_type(large_page_size);
3436 set_coredump_filter();
3437 }
3439 #ifndef SHM_HUGETLB
3440 #define SHM_HUGETLB 04000
3441 #endif
3443 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3444 // "exec" is passed in but not used. Creating the shared image for
3445 // the code cache doesn't have an SHM_X executable permission to check.
3446 assert(UseLargePages && UseSHM, "only for SHM large pages");
3447 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3449 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3450 return NULL; // Fallback to small pages.
3451 }
3453 key_t key = IPC_PRIVATE;
3454 char *addr;
3456 bool warn_on_failure = UseLargePages &&
3457 (!FLAG_IS_DEFAULT(UseLargePages) ||
3458 !FLAG_IS_DEFAULT(UseSHM) ||
3459 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3460 );
3461 char msg[128];
3463 // Create a large shared memory region to attach to based on size.
3464 // Currently, size is the total size of the heap
3465 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3466 if (shmid == -1) {
3467 // Possible reasons for shmget failure:
3468 // 1. shmmax is too small for Java heap.
3469 // > check shmmax value: cat /proc/sys/kernel/shmmax
3470 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3471 // 2. not enough large page memory.
3472 // > check available large pages: cat /proc/meminfo
3473 // > increase amount of large pages:
3474 // echo new_value > /proc/sys/vm/nr_hugepages
3475 // Note 1: different Linux may use different name for this property,
3476 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3477 // Note 2: it's possible there's enough physical memory available but
3478 // they are so fragmented after a long run that they can't
3479 // coalesce into large pages. Try to reserve large pages when
3480 // the system is still "fresh".
3481 if (warn_on_failure) {
3482 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3483 warning(msg);
3484 }
3485 return NULL;
3486 }
3488 // attach to the region
3489 addr = (char*)shmat(shmid, req_addr, 0);
3490 int err = errno;
3492 // Remove shmid. If shmat() is successful, the actual shared memory segment
3493 // will be deleted when it's detached by shmdt() or when the process
3494 // terminates. If shmat() is not successful this will remove the shared
3495 // segment immediately.
3496 shmctl(shmid, IPC_RMID, NULL);
3498 if ((intptr_t)addr == -1) {
3499 if (warn_on_failure) {
3500 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3501 warning(msg);
3502 }
3503 return NULL;
3504 }
3506 return addr;
3507 }
3509 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3510 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3512 bool warn_on_failure = UseLargePages &&
3513 (!FLAG_IS_DEFAULT(UseLargePages) ||
3514 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3515 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3517 if (warn_on_failure) {
3518 char msg[128];
3519 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3520 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3521 warning(msg);
3522 }
3523 }
3525 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3526 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3527 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3528 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3530 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3531 char* addr = (char*)::mmap(req_addr, bytes, prot,
3532 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3533 -1, 0);
3535 if (addr == MAP_FAILED) {
3536 warn_on_large_pages_failure(req_addr, bytes, errno);
3537 return NULL;
3538 }
3540 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3542 return addr;
3543 }
3545 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3546 size_t large_page_size = os::large_page_size();
3548 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3550 // Allocate small pages.
3552 char* start;
3553 if (req_addr != NULL) {
3554 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3555 assert(is_size_aligned(bytes, alignment), "Must be");
3556 start = os::reserve_memory(bytes, req_addr);
3557 assert(start == NULL || start == req_addr, "Must be");
3558 } else {
3559 start = os::reserve_memory_aligned(bytes, alignment);
3560 }
3562 if (start == NULL) {
3563 return NULL;
3564 }
3566 assert(is_ptr_aligned(start, alignment), "Must be");
3568 // os::reserve_memory_special will record this memory area.
3569 // Need to release it here to prevent overlapping reservations.
3570 MemTracker::record_virtual_memory_release((address)start, bytes);
3572 char* end = start + bytes;
3574 // Find the regions of the allocated chunk that can be promoted to large pages.
3575 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3576 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3578 size_t lp_bytes = lp_end - lp_start;
3580 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3582 if (lp_bytes == 0) {
3583 // The mapped region doesn't even span the start and the end of a large page.
3584 // Fall back to allocate a non-special area.
3585 ::munmap(start, end - start);
3586 return NULL;
3587 }
3589 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3592 void* result;
3594 if (start != lp_start) {
3595 result = ::mmap(start, lp_start - start, prot,
3596 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3597 -1, 0);
3598 if (result == MAP_FAILED) {
3599 ::munmap(lp_start, end - lp_start);
3600 return NULL;
3601 }
3602 }
3604 result = ::mmap(lp_start, lp_bytes, prot,
3605 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3606 -1, 0);
3607 if (result == MAP_FAILED) {
3608 warn_on_large_pages_failure(req_addr, bytes, errno);
3609 // If the mmap above fails, the large pages region will be unmapped and we
3610 // have regions before and after with small pages. Release these regions.
3611 //
3612 // | mapped | unmapped | mapped |
3613 // ^ ^ ^ ^
3614 // start lp_start lp_end end
3615 //
3616 ::munmap(start, lp_start - start);
3617 ::munmap(lp_end, end - lp_end);
3618 return NULL;
3619 }
3621 if (lp_end != end) {
3622 result = ::mmap(lp_end, end - lp_end, prot,
3623 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3624 -1, 0);
3625 if (result == MAP_FAILED) {
3626 ::munmap(start, lp_end - start);
3627 return NULL;
3628 }
3629 }
3631 return start;
3632 }
3634 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3635 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3636 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3637 assert(is_power_of_2(alignment), "Must be");
3638 assert(is_power_of_2(os::large_page_size()), "Must be");
3639 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3641 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3642 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3643 } else {
3644 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3645 }
3646 }
3648 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3649 assert(UseLargePages, "only for large pages");
3651 char* addr;
3652 if (UseSHM) {
3653 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3654 } else {
3655 assert(UseHugeTLBFS, "must be");
3656 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3657 }
3659 if (addr != NULL) {
3660 if (UseNUMAInterleaving) {
3661 numa_make_global(addr, bytes);
3662 }
3664 // The memory is committed
3665 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3666 }
3668 return addr;
3669 }
3671 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3672 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3673 return shmdt(base) == 0;
3674 }
3676 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3677 return pd_release_memory(base, bytes);
3678 }
3680 bool os::release_memory_special(char* base, size_t bytes) {
3681 assert(UseLargePages, "only for large pages");
3683 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3685 bool res;
3686 if (UseSHM) {
3687 res = os::Linux::release_memory_special_shm(base, bytes);
3688 } else {
3689 assert(UseHugeTLBFS, "must be");
3690 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3691 }
3693 if (res) {
3694 tkr.record((address)base, bytes);
3695 } else {
3696 tkr.discard();
3697 }
3699 return res;
3700 }
3702 size_t os::large_page_size() {
3703 return _large_page_size;
3704 }
3706 // With SysV SHM the entire memory region must be allocated as shared
3707 // memory.
3708 // HugeTLBFS allows application to commit large page memory on demand.
3709 // However, when committing memory with HugeTLBFS fails, the region
3710 // that was supposed to be committed will lose the old reservation
3711 // and allow other threads to steal that memory region. Because of this
3712 // behavior we can't commit HugeTLBFS memory.
3713 bool os::can_commit_large_page_memory() {
3714 return UseTransparentHugePages;
3715 }
3717 bool os::can_execute_large_page_memory() {
3718 return UseTransparentHugePages || UseHugeTLBFS;
3719 }
3721 // Reserve memory at an arbitrary address, only if that area is
3722 // available (and not reserved for something else).
3724 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3725 const int max_tries = 10;
3726 char* base[max_tries];
3727 size_t size[max_tries];
3728 const size_t gap = 0x000000;
3730 // Assert only that the size is a multiple of the page size, since
3731 // that's all that mmap requires, and since that's all we really know
3732 // about at this low abstraction level. If we need higher alignment,
3733 // we can either pass an alignment to this method or verify alignment
3734 // in one of the methods further up the call chain. See bug 5044738.
3735 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3737 // Repeatedly allocate blocks until the block is allocated at the
3738 // right spot. Give up after max_tries. Note that reserve_memory() will
3739 // automatically update _highest_vm_reserved_address if the call is
3740 // successful. The variable tracks the highest memory address every reserved
3741 // by JVM. It is used to detect heap-stack collision if running with
3742 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3743 // space than needed, it could confuse the collision detecting code. To
3744 // solve the problem, save current _highest_vm_reserved_address and
3745 // calculate the correct value before return.
3746 address old_highest = _highest_vm_reserved_address;
3748 // Linux mmap allows caller to pass an address as hint; give it a try first,
3749 // if kernel honors the hint then we can return immediately.
3750 char * addr = anon_mmap(requested_addr, bytes, false);
3751 if (addr == requested_addr) {
3752 return requested_addr;
3753 }
3755 if (addr != NULL) {
3756 // mmap() is successful but it fails to reserve at the requested address
3757 anon_munmap(addr, bytes);
3758 }
3760 int i;
3761 for (i = 0; i < max_tries; ++i) {
3762 base[i] = reserve_memory(bytes);
3764 if (base[i] != NULL) {
3765 // Is this the block we wanted?
3766 if (base[i] == requested_addr) {
3767 size[i] = bytes;
3768 break;
3769 }
3771 // Does this overlap the block we wanted? Give back the overlapped
3772 // parts and try again.
3774 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3775 if (top_overlap >= 0 && top_overlap < bytes) {
3776 unmap_memory(base[i], top_overlap);
3777 base[i] += top_overlap;
3778 size[i] = bytes - top_overlap;
3779 } else {
3780 size_t bottom_overlap = base[i] + bytes - requested_addr;
3781 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3782 unmap_memory(requested_addr, bottom_overlap);
3783 size[i] = bytes - bottom_overlap;
3784 } else {
3785 size[i] = bytes;
3786 }
3787 }
3788 }
3789 }
3791 // Give back the unused reserved pieces.
3793 for (int j = 0; j < i; ++j) {
3794 if (base[j] != NULL) {
3795 unmap_memory(base[j], size[j]);
3796 }
3797 }
3799 if (i < max_tries) {
3800 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3801 return requested_addr;
3802 } else {
3803 _highest_vm_reserved_address = old_highest;
3804 return NULL;
3805 }
3806 }
3808 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3809 return ::read(fd, buf, nBytes);
3810 }
3812 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3813 // Solaris uses poll(), linux uses park().
3814 // Poll() is likely a better choice, assuming that Thread.interrupt()
3815 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3816 // SIGSEGV, see 4355769.
3818 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3819 assert(thread == Thread::current(), "thread consistency check");
3821 ParkEvent * const slp = thread->_SleepEvent ;
3822 slp->reset() ;
3823 OrderAccess::fence() ;
3825 if (interruptible) {
3826 jlong prevtime = javaTimeNanos();
3828 for (;;) {
3829 if (os::is_interrupted(thread, true)) {
3830 return OS_INTRPT;
3831 }
3833 jlong newtime = javaTimeNanos();
3835 if (newtime - prevtime < 0) {
3836 // time moving backwards, should only happen if no monotonic clock
3837 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3838 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3839 } else {
3840 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3841 }
3843 if(millis <= 0) {
3844 return OS_OK;
3845 }
3847 prevtime = newtime;
3849 {
3850 assert(thread->is_Java_thread(), "sanity check");
3851 JavaThread *jt = (JavaThread *) thread;
3852 ThreadBlockInVM tbivm(jt);
3853 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3855 jt->set_suspend_equivalent();
3856 // cleared by handle_special_suspend_equivalent_condition() or
3857 // java_suspend_self() via check_and_wait_while_suspended()
3859 slp->park(millis);
3861 // were we externally suspended while we were waiting?
3862 jt->check_and_wait_while_suspended();
3863 }
3864 }
3865 } else {
3866 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3867 jlong prevtime = javaTimeNanos();
3869 for (;;) {
3870 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3871 // the 1st iteration ...
3872 jlong newtime = javaTimeNanos();
3874 if (newtime - prevtime < 0) {
3875 // time moving backwards, should only happen if no monotonic clock
3876 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3877 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3878 } else {
3879 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3880 }
3882 if(millis <= 0) break ;
3884 prevtime = newtime;
3885 slp->park(millis);
3886 }
3887 return OS_OK ;
3888 }
3889 }
3891 //
3892 // Short sleep, direct OS call.
3893 //
3894 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3895 // sched_yield(2) will actually give up the CPU:
3896 //
3897 // * Alone on this pariticular CPU, keeps running.
3898 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3899 // (pre 2.6.39).
3900 //
3901 // So calling this with 0 is an alternative.
3902 //
3903 void os::naked_short_sleep(jlong ms) {
3904 struct timespec req;
3906 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3907 req.tv_sec = 0;
3908 if (ms > 0) {
3909 req.tv_nsec = (ms % 1000) * 1000000;
3910 }
3911 else {
3912 req.tv_nsec = 1;
3913 }
3915 nanosleep(&req, NULL);
3917 return;
3918 }
3920 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3921 void os::infinite_sleep() {
3922 while (true) { // sleep forever ...
3923 ::sleep(100); // ... 100 seconds at a time
3924 }
3925 }
3927 // Used to convert frequent JVM_Yield() to nops
3928 bool os::dont_yield() {
3929 return DontYieldALot;
3930 }
3932 void os::yield() {
3933 sched_yield();
3934 }
3936 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3938 void os::yield_all(int attempts) {
3939 // Yields to all threads, including threads with lower priorities
3940 // Threads on Linux are all with same priority. The Solaris style
3941 // os::yield_all() with nanosleep(1ms) is not necessary.
3942 sched_yield();
3943 }
3945 // Called from the tight loops to possibly influence time-sharing heuristics
3946 void os::loop_breaker(int attempts) {
3947 os::yield_all(attempts);
3948 }
3950 ////////////////////////////////////////////////////////////////////////////////
3951 // thread priority support
3953 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3954 // only supports dynamic priority, static priority must be zero. For real-time
3955 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3956 // However, for large multi-threaded applications, SCHED_RR is not only slower
3957 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3958 // of 5 runs - Sep 2005).
3959 //
3960 // The following code actually changes the niceness of kernel-thread/LWP. It
3961 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3962 // not the entire user process, and user level threads are 1:1 mapped to kernel
3963 // threads. It has always been the case, but could change in the future. For
3964 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3965 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3967 int os::java_to_os_priority[CriticalPriority + 1] = {
3968 19, // 0 Entry should never be used
3970 4, // 1 MinPriority
3971 3, // 2
3972 2, // 3
3974 1, // 4
3975 0, // 5 NormPriority
3976 -1, // 6
3978 -2, // 7
3979 -3, // 8
3980 -4, // 9 NearMaxPriority
3982 -5, // 10 MaxPriority
3984 -5 // 11 CriticalPriority
3985 };
3987 static int prio_init() {
3988 if (ThreadPriorityPolicy == 1) {
3989 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3990 // if effective uid is not root. Perhaps, a more elegant way of doing
3991 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3992 if (geteuid() != 0) {
3993 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3994 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3995 }
3996 ThreadPriorityPolicy = 0;
3997 }
3998 }
3999 if (UseCriticalJavaThreadPriority) {
4000 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4001 }
4002 return 0;
4003 }
4005 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4006 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4008 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4009 return (ret == 0) ? OS_OK : OS_ERR;
4010 }
4012 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4013 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4014 *priority_ptr = java_to_os_priority[NormPriority];
4015 return OS_OK;
4016 }
4018 errno = 0;
4019 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4020 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4021 }
4023 // Hint to the underlying OS that a task switch would not be good.
4024 // Void return because it's a hint and can fail.
4025 void os::hint_no_preempt() {}
4027 ////////////////////////////////////////////////////////////////////////////////
4028 // suspend/resume support
4030 // the low-level signal-based suspend/resume support is a remnant from the
4031 // old VM-suspension that used to be for java-suspension, safepoints etc,
4032 // within hotspot. Now there is a single use-case for this:
4033 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4034 // that runs in the watcher thread.
4035 // The remaining code is greatly simplified from the more general suspension
4036 // code that used to be used.
4037 //
4038 // The protocol is quite simple:
4039 // - suspend:
4040 // - sends a signal to the target thread
4041 // - polls the suspend state of the osthread using a yield loop
4042 // - target thread signal handler (SR_handler) sets suspend state
4043 // and blocks in sigsuspend until continued
4044 // - resume:
4045 // - sets target osthread state to continue
4046 // - sends signal to end the sigsuspend loop in the SR_handler
4047 //
4048 // Note that the SR_lock plays no role in this suspend/resume protocol.
4049 //
4051 static void resume_clear_context(OSThread *osthread) {
4052 osthread->set_ucontext(NULL);
4053 osthread->set_siginfo(NULL);
4054 }
4056 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4057 osthread->set_ucontext(context);
4058 osthread->set_siginfo(siginfo);
4059 }
4061 //
4062 // Handler function invoked when a thread's execution is suspended or
4063 // resumed. We have to be careful that only async-safe functions are
4064 // called here (Note: most pthread functions are not async safe and
4065 // should be avoided.)
4066 //
4067 // Note: sigwait() is a more natural fit than sigsuspend() from an
4068 // interface point of view, but sigwait() prevents the signal hander
4069 // from being run. libpthread would get very confused by not having
4070 // its signal handlers run and prevents sigwait()'s use with the
4071 // mutex granting granting signal.
4072 //
4073 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4074 //
4075 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4076 // Save and restore errno to avoid confusing native code with EINTR
4077 // after sigsuspend.
4078 int old_errno = errno;
4080 Thread* thread = Thread::current();
4081 OSThread* osthread = thread->osthread();
4082 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4084 os::SuspendResume::State current = osthread->sr.state();
4085 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4086 suspend_save_context(osthread, siginfo, context);
4088 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4089 os::SuspendResume::State state = osthread->sr.suspended();
4090 if (state == os::SuspendResume::SR_SUSPENDED) {
4091 sigset_t suspend_set; // signals for sigsuspend()
4093 // get current set of blocked signals and unblock resume signal
4094 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4095 sigdelset(&suspend_set, SR_signum);
4097 sr_semaphore.signal();
4098 // wait here until we are resumed
4099 while (1) {
4100 sigsuspend(&suspend_set);
4102 os::SuspendResume::State result = osthread->sr.running();
4103 if (result == os::SuspendResume::SR_RUNNING) {
4104 sr_semaphore.signal();
4105 break;
4106 }
4107 }
4109 } else if (state == os::SuspendResume::SR_RUNNING) {
4110 // request was cancelled, continue
4111 } else {
4112 ShouldNotReachHere();
4113 }
4115 resume_clear_context(osthread);
4116 } else if (current == os::SuspendResume::SR_RUNNING) {
4117 // request was cancelled, continue
4118 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4119 // ignore
4120 } else {
4121 // ignore
4122 }
4124 errno = old_errno;
4125 }
4128 static int SR_initialize() {
4129 struct sigaction act;
4130 char *s;
4131 /* Get signal number to use for suspend/resume */
4132 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4133 int sig = ::strtol(s, 0, 10);
4134 if (sig > 0 || sig < _NSIG) {
4135 SR_signum = sig;
4136 }
4137 }
4139 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4140 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4142 sigemptyset(&SR_sigset);
4143 sigaddset(&SR_sigset, SR_signum);
4145 /* Set up signal handler for suspend/resume */
4146 act.sa_flags = SA_RESTART|SA_SIGINFO;
4147 act.sa_handler = (void (*)(int)) SR_handler;
4149 // SR_signum is blocked by default.
4150 // 4528190 - We also need to block pthread restart signal (32 on all
4151 // supported Linux platforms). Note that LinuxThreads need to block
4152 // this signal for all threads to work properly. So we don't have
4153 // to use hard-coded signal number when setting up the mask.
4154 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4156 if (sigaction(SR_signum, &act, 0) == -1) {
4157 return -1;
4158 }
4160 // Save signal flag
4161 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4162 return 0;
4163 }
4165 static int sr_notify(OSThread* osthread) {
4166 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4167 assert_status(status == 0, status, "pthread_kill");
4168 return status;
4169 }
4171 // "Randomly" selected value for how long we want to spin
4172 // before bailing out on suspending a thread, also how often
4173 // we send a signal to a thread we want to resume
4174 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4175 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4177 // returns true on success and false on error - really an error is fatal
4178 // but this seems the normal response to library errors
4179 static bool do_suspend(OSThread* osthread) {
4180 assert(osthread->sr.is_running(), "thread should be running");
4181 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4183 // mark as suspended and send signal
4184 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4185 // failed to switch, state wasn't running?
4186 ShouldNotReachHere();
4187 return false;
4188 }
4190 if (sr_notify(osthread) != 0) {
4191 ShouldNotReachHere();
4192 }
4194 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4195 while (true) {
4196 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4197 break;
4198 } else {
4199 // timeout
4200 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4201 if (cancelled == os::SuspendResume::SR_RUNNING) {
4202 return false;
4203 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4204 // make sure that we consume the signal on the semaphore as well
4205 sr_semaphore.wait();
4206 break;
4207 } else {
4208 ShouldNotReachHere();
4209 return false;
4210 }
4211 }
4212 }
4214 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4215 return true;
4216 }
4218 static void do_resume(OSThread* osthread) {
4219 assert(osthread->sr.is_suspended(), "thread should be suspended");
4220 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4222 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4223 // failed to switch to WAKEUP_REQUEST
4224 ShouldNotReachHere();
4225 return;
4226 }
4228 while (true) {
4229 if (sr_notify(osthread) == 0) {
4230 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4231 if (osthread->sr.is_running()) {
4232 return;
4233 }
4234 }
4235 } else {
4236 ShouldNotReachHere();
4237 }
4238 }
4240 guarantee(osthread->sr.is_running(), "Must be running!");
4241 }
4243 ////////////////////////////////////////////////////////////////////////////////
4244 // interrupt support
4246 void os::interrupt(Thread* thread) {
4247 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4248 "possibility of dangling Thread pointer");
4250 OSThread* osthread = thread->osthread();
4252 if (!osthread->interrupted()) {
4253 osthread->set_interrupted(true);
4254 // More than one thread can get here with the same value of osthread,
4255 // resulting in multiple notifications. We do, however, want the store
4256 // to interrupted() to be visible to other threads before we execute unpark().
4257 OrderAccess::fence();
4258 ParkEvent * const slp = thread->_SleepEvent ;
4259 if (slp != NULL) slp->unpark() ;
4260 }
4262 // For JSR166. Unpark even if interrupt status already was set
4263 if (thread->is_Java_thread())
4264 ((JavaThread*)thread)->parker()->unpark();
4266 ParkEvent * ev = thread->_ParkEvent ;
4267 if (ev != NULL) ev->unpark() ;
4269 }
4271 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4272 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4273 "possibility of dangling Thread pointer");
4275 OSThread* osthread = thread->osthread();
4277 bool interrupted = osthread->interrupted();
4279 if (interrupted && clear_interrupted) {
4280 osthread->set_interrupted(false);
4281 // consider thread->_SleepEvent->reset() ... optional optimization
4282 }
4284 return interrupted;
4285 }
4287 ///////////////////////////////////////////////////////////////////////////////////
4288 // signal handling (except suspend/resume)
4290 // This routine may be used by user applications as a "hook" to catch signals.
4291 // The user-defined signal handler must pass unrecognized signals to this
4292 // routine, and if it returns true (non-zero), then the signal handler must
4293 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4294 // routine will never retun false (zero), but instead will execute a VM panic
4295 // routine kill the process.
4296 //
4297 // If this routine returns false, it is OK to call it again. This allows
4298 // the user-defined signal handler to perform checks either before or after
4299 // the VM performs its own checks. Naturally, the user code would be making
4300 // a serious error if it tried to handle an exception (such as a null check
4301 // or breakpoint) that the VM was generating for its own correct operation.
4302 //
4303 // This routine may recognize any of the following kinds of signals:
4304 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4305 // It should be consulted by handlers for any of those signals.
4306 //
4307 // The caller of this routine must pass in the three arguments supplied
4308 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4309 // field of the structure passed to sigaction(). This routine assumes that
4310 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4311 //
4312 // Note that the VM will print warnings if it detects conflicting signal
4313 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4314 //
4315 extern "C" JNIEXPORT int
4316 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4317 void* ucontext, int abort_if_unrecognized);
4319 void signalHandler(int sig, siginfo_t* info, void* uc) {
4320 assert(info != NULL && uc != NULL, "it must be old kernel");
4321 int orig_errno = errno; // Preserve errno value over signal handler.
4322 JVM_handle_linux_signal(sig, info, uc, true);
4323 errno = orig_errno;
4324 }
4327 // This boolean allows users to forward their own non-matching signals
4328 // to JVM_handle_linux_signal, harmlessly.
4329 bool os::Linux::signal_handlers_are_installed = false;
4331 // For signal-chaining
4332 struct sigaction os::Linux::sigact[MAXSIGNUM];
4333 unsigned int os::Linux::sigs = 0;
4334 bool os::Linux::libjsig_is_loaded = false;
4335 typedef struct sigaction *(*get_signal_t)(int);
4336 get_signal_t os::Linux::get_signal_action = NULL;
4338 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4339 struct sigaction *actp = NULL;
4341 if (libjsig_is_loaded) {
4342 // Retrieve the old signal handler from libjsig
4343 actp = (*get_signal_action)(sig);
4344 }
4345 if (actp == NULL) {
4346 // Retrieve the preinstalled signal handler from jvm
4347 actp = get_preinstalled_handler(sig);
4348 }
4350 return actp;
4351 }
4353 static bool call_chained_handler(struct sigaction *actp, int sig,
4354 siginfo_t *siginfo, void *context) {
4355 // Call the old signal handler
4356 if (actp->sa_handler == SIG_DFL) {
4357 // It's more reasonable to let jvm treat it as an unexpected exception
4358 // instead of taking the default action.
4359 return false;
4360 } else if (actp->sa_handler != SIG_IGN) {
4361 if ((actp->sa_flags & SA_NODEFER) == 0) {
4362 // automaticlly block the signal
4363 sigaddset(&(actp->sa_mask), sig);
4364 }
4366 sa_handler_t hand;
4367 sa_sigaction_t sa;
4368 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4369 // retrieve the chained handler
4370 if (siginfo_flag_set) {
4371 sa = actp->sa_sigaction;
4372 } else {
4373 hand = actp->sa_handler;
4374 }
4376 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4377 actp->sa_handler = SIG_DFL;
4378 }
4380 // try to honor the signal mask
4381 sigset_t oset;
4382 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4384 // call into the chained handler
4385 if (siginfo_flag_set) {
4386 (*sa)(sig, siginfo, context);
4387 } else {
4388 (*hand)(sig);
4389 }
4391 // restore the signal mask
4392 pthread_sigmask(SIG_SETMASK, &oset, 0);
4393 }
4394 // Tell jvm's signal handler the signal is taken care of.
4395 return true;
4396 }
4398 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4399 bool chained = false;
4400 // signal-chaining
4401 if (UseSignalChaining) {
4402 struct sigaction *actp = get_chained_signal_action(sig);
4403 if (actp != NULL) {
4404 chained = call_chained_handler(actp, sig, siginfo, context);
4405 }
4406 }
4407 return chained;
4408 }
4410 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4411 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4412 return &sigact[sig];
4413 }
4414 return NULL;
4415 }
4417 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4418 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4419 sigact[sig] = oldAct;
4420 sigs |= (unsigned int)1 << sig;
4421 }
4423 // for diagnostic
4424 int os::Linux::sigflags[MAXSIGNUM];
4426 int os::Linux::get_our_sigflags(int sig) {
4427 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4428 return sigflags[sig];
4429 }
4431 void os::Linux::set_our_sigflags(int sig, int flags) {
4432 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4433 sigflags[sig] = flags;
4434 }
4436 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4437 // Check for overwrite.
4438 struct sigaction oldAct;
4439 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4441 void* oldhand = oldAct.sa_sigaction
4442 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4443 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4444 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4445 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4446 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4447 if (AllowUserSignalHandlers || !set_installed) {
4448 // Do not overwrite; user takes responsibility to forward to us.
4449 return;
4450 } else if (UseSignalChaining) {
4451 // save the old handler in jvm
4452 save_preinstalled_handler(sig, oldAct);
4453 // libjsig also interposes the sigaction() call below and saves the
4454 // old sigaction on it own.
4455 } else {
4456 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4457 "%#lx for signal %d.", (long)oldhand, sig));
4458 }
4459 }
4461 struct sigaction sigAct;
4462 sigfillset(&(sigAct.sa_mask));
4463 sigAct.sa_handler = SIG_DFL;
4464 if (!set_installed) {
4465 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4466 } else {
4467 sigAct.sa_sigaction = signalHandler;
4468 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4469 }
4470 // Save flags, which are set by ours
4471 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4472 sigflags[sig] = sigAct.sa_flags;
4474 int ret = sigaction(sig, &sigAct, &oldAct);
4475 assert(ret == 0, "check");
4477 void* oldhand2 = oldAct.sa_sigaction
4478 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4479 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4480 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4481 }
4483 // install signal handlers for signals that HotSpot needs to
4484 // handle in order to support Java-level exception handling.
4486 void os::Linux::install_signal_handlers() {
4487 if (!signal_handlers_are_installed) {
4488 signal_handlers_are_installed = true;
4490 // signal-chaining
4491 typedef void (*signal_setting_t)();
4492 signal_setting_t begin_signal_setting = NULL;
4493 signal_setting_t end_signal_setting = NULL;
4494 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4495 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4496 if (begin_signal_setting != NULL) {
4497 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4498 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4499 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4500 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4501 libjsig_is_loaded = true;
4502 assert(UseSignalChaining, "should enable signal-chaining");
4503 }
4504 if (libjsig_is_loaded) {
4505 // Tell libjsig jvm is setting signal handlers
4506 (*begin_signal_setting)();
4507 }
4509 set_signal_handler(SIGSEGV, true);
4510 set_signal_handler(SIGPIPE, true);
4511 set_signal_handler(SIGBUS, true);
4512 set_signal_handler(SIGILL, true);
4513 set_signal_handler(SIGFPE, true);
4514 set_signal_handler(SIGXFSZ, true);
4516 if (libjsig_is_loaded) {
4517 // Tell libjsig jvm finishes setting signal handlers
4518 (*end_signal_setting)();
4519 }
4521 // We don't activate signal checker if libjsig is in place, we trust ourselves
4522 // and if UserSignalHandler is installed all bets are off.
4523 // Log that signal checking is off only if -verbose:jni is specified.
4524 if (CheckJNICalls) {
4525 if (libjsig_is_loaded) {
4526 if (PrintJNIResolving) {
4527 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4528 }
4529 check_signals = false;
4530 }
4531 if (AllowUserSignalHandlers) {
4532 if (PrintJNIResolving) {
4533 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4534 }
4535 check_signals = false;
4536 }
4537 }
4538 }
4539 }
4541 // This is the fastest way to get thread cpu time on Linux.
4542 // Returns cpu time (user+sys) for any thread, not only for current.
4543 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4544 // It might work on 2.6.10+ with a special kernel/glibc patch.
4545 // For reference, please, see IEEE Std 1003.1-2004:
4546 // http://www.unix.org/single_unix_specification
4548 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4549 struct timespec tp;
4550 int rc = os::Linux::clock_gettime(clockid, &tp);
4551 assert(rc == 0, "clock_gettime is expected to return 0 code");
4553 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4554 }
4556 /////
4557 // glibc on Linux platform uses non-documented flag
4558 // to indicate, that some special sort of signal
4559 // trampoline is used.
4560 // We will never set this flag, and we should
4561 // ignore this flag in our diagnostic
4562 #ifdef SIGNIFICANT_SIGNAL_MASK
4563 #undef SIGNIFICANT_SIGNAL_MASK
4564 #endif
4565 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4567 static const char* get_signal_handler_name(address handler,
4568 char* buf, int buflen) {
4569 int offset;
4570 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4571 if (found) {
4572 // skip directory names
4573 const char *p1, *p2;
4574 p1 = buf;
4575 size_t len = strlen(os::file_separator());
4576 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4577 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4578 } else {
4579 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4580 }
4581 return buf;
4582 }
4584 static void print_signal_handler(outputStream* st, int sig,
4585 char* buf, size_t buflen) {
4586 struct sigaction sa;
4588 sigaction(sig, NULL, &sa);
4590 // See comment for SIGNIFICANT_SIGNAL_MASK define
4591 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4593 st->print("%s: ", os::exception_name(sig, buf, buflen));
4595 address handler = (sa.sa_flags & SA_SIGINFO)
4596 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4597 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4599 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4600 st->print("SIG_DFL");
4601 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4602 st->print("SIG_IGN");
4603 } else {
4604 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4605 }
4607 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
4609 address rh = VMError::get_resetted_sighandler(sig);
4610 // May be, handler was resetted by VMError?
4611 if(rh != NULL) {
4612 handler = rh;
4613 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4614 }
4616 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
4618 // Check: is it our handler?
4619 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4620 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4621 // It is our signal handler
4622 // check for flags, reset system-used one!
4623 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4624 st->print(
4625 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4626 os::Linux::get_our_sigflags(sig));
4627 }
4628 }
4629 st->cr();
4630 }
4633 #define DO_SIGNAL_CHECK(sig) \
4634 if (!sigismember(&check_signal_done, sig)) \
4635 os::Linux::check_signal_handler(sig)
4637 // This method is a periodic task to check for misbehaving JNI applications
4638 // under CheckJNI, we can add any periodic checks here
4640 void os::run_periodic_checks() {
4642 if (check_signals == false) return;
4644 // SEGV and BUS if overridden could potentially prevent
4645 // generation of hs*.log in the event of a crash, debugging
4646 // such a case can be very challenging, so we absolutely
4647 // check the following for a good measure:
4648 DO_SIGNAL_CHECK(SIGSEGV);
4649 DO_SIGNAL_CHECK(SIGILL);
4650 DO_SIGNAL_CHECK(SIGFPE);
4651 DO_SIGNAL_CHECK(SIGBUS);
4652 DO_SIGNAL_CHECK(SIGPIPE);
4653 DO_SIGNAL_CHECK(SIGXFSZ);
4656 // ReduceSignalUsage allows the user to override these handlers
4657 // see comments at the very top and jvm_solaris.h
4658 if (!ReduceSignalUsage) {
4659 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4660 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4661 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4662 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4663 }
4665 DO_SIGNAL_CHECK(SR_signum);
4666 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4667 }
4669 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4671 static os_sigaction_t os_sigaction = NULL;
4673 void os::Linux::check_signal_handler(int sig) {
4674 char buf[O_BUFLEN];
4675 address jvmHandler = NULL;
4678 struct sigaction act;
4679 if (os_sigaction == NULL) {
4680 // only trust the default sigaction, in case it has been interposed
4681 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4682 if (os_sigaction == NULL) return;
4683 }
4685 os_sigaction(sig, (struct sigaction*)NULL, &act);
4688 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4690 address thisHandler = (act.sa_flags & SA_SIGINFO)
4691 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4692 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4695 switch(sig) {
4696 case SIGSEGV:
4697 case SIGBUS:
4698 case SIGFPE:
4699 case SIGPIPE:
4700 case SIGILL:
4701 case SIGXFSZ:
4702 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4703 break;
4705 case SHUTDOWN1_SIGNAL:
4706 case SHUTDOWN2_SIGNAL:
4707 case SHUTDOWN3_SIGNAL:
4708 case BREAK_SIGNAL:
4709 jvmHandler = (address)user_handler();
4710 break;
4712 case INTERRUPT_SIGNAL:
4713 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4714 break;
4716 default:
4717 if (sig == SR_signum) {
4718 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4719 } else {
4720 return;
4721 }
4722 break;
4723 }
4725 if (thisHandler != jvmHandler) {
4726 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4727 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4728 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4729 // No need to check this sig any longer
4730 sigaddset(&check_signal_done, sig);
4731 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4732 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4733 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4734 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4735 // No need to check this sig any longer
4736 sigaddset(&check_signal_done, sig);
4737 }
4739 // Dump all the signal
4740 if (sigismember(&check_signal_done, sig)) {
4741 print_signal_handlers(tty, buf, O_BUFLEN);
4742 }
4743 }
4745 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4747 extern bool signal_name(int signo, char* buf, size_t len);
4749 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4750 if (0 < exception_code && exception_code <= SIGRTMAX) {
4751 // signal
4752 if (!signal_name(exception_code, buf, size)) {
4753 jio_snprintf(buf, size, "SIG%d", exception_code);
4754 }
4755 return buf;
4756 } else {
4757 return NULL;
4758 }
4759 }
4761 // this is called _before_ the most of global arguments have been parsed
4762 void os::init(void) {
4763 char dummy; /* used to get a guess on initial stack address */
4764 // first_hrtime = gethrtime();
4766 // With LinuxThreads the JavaMain thread pid (primordial thread)
4767 // is different than the pid of the java launcher thread.
4768 // So, on Linux, the launcher thread pid is passed to the VM
4769 // via the sun.java.launcher.pid property.
4770 // Use this property instead of getpid() if it was correctly passed.
4771 // See bug 6351349.
4772 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4774 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4776 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4778 init_random(1234567);
4780 ThreadCritical::initialize();
4782 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4783 if (Linux::page_size() == -1) {
4784 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4785 strerror(errno)));
4786 }
4787 init_page_sizes((size_t) Linux::page_size());
4789 Linux::initialize_system_info();
4791 // main_thread points to the aboriginal thread
4792 Linux::_main_thread = pthread_self();
4794 Linux::clock_init();
4795 initial_time_count = javaTimeNanos();
4797 // pthread_condattr initialization for monotonic clock
4798 int status;
4799 pthread_condattr_t* _condattr = os::Linux::condAttr();
4800 if ((status = pthread_condattr_init(_condattr)) != 0) {
4801 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4802 }
4803 // Only set the clock if CLOCK_MONOTONIC is available
4804 if (Linux::supports_monotonic_clock()) {
4805 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4806 if (status == EINVAL) {
4807 warning("Unable to use monotonic clock with relative timed-waits" \
4808 " - changes to the time-of-day clock may have adverse affects");
4809 } else {
4810 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4811 }
4812 }
4813 }
4814 // else it defaults to CLOCK_REALTIME
4816 pthread_mutex_init(&dl_mutex, NULL);
4818 // If the pagesize of the VM is greater than 8K determine the appropriate
4819 // number of initial guard pages. The user can change this with the
4820 // command line arguments, if needed.
4821 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4822 StackYellowPages = 1;
4823 StackRedPages = 1;
4824 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4825 }
4826 }
4828 // To install functions for atexit system call
4829 extern "C" {
4830 static void perfMemory_exit_helper() {
4831 perfMemory_exit();
4832 }
4833 }
4835 // this is called _after_ the global arguments have been parsed
4836 jint os::init_2(void)
4837 {
4838 Linux::fast_thread_clock_init();
4840 // Allocate a single page and mark it as readable for safepoint polling
4841 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4842 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4844 os::set_polling_page( polling_page );
4846 #ifndef PRODUCT
4847 if(Verbose && PrintMiscellaneous)
4848 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4849 #endif
4851 if (!UseMembar) {
4852 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4853 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4854 os::set_memory_serialize_page( mem_serialize_page );
4856 #ifndef PRODUCT
4857 if(Verbose && PrintMiscellaneous)
4858 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4859 #endif
4860 }
4862 // initialize suspend/resume support - must do this before signal_sets_init()
4863 if (SR_initialize() != 0) {
4864 perror("SR_initialize failed");
4865 return JNI_ERR;
4866 }
4868 Linux::signal_sets_init();
4869 Linux::install_signal_handlers();
4871 // Check minimum allowable stack size for thread creation and to initialize
4872 // the java system classes, including StackOverflowError - depends on page
4873 // size. Add a page for compiler2 recursion in main thread.
4874 // Add in 2*BytesPerWord times page size to account for VM stack during
4875 // class initialization depending on 32 or 64 bit VM.
4876 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4877 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4878 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4880 size_t threadStackSizeInBytes = ThreadStackSize * K;
4881 if (threadStackSizeInBytes != 0 &&
4882 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4883 tty->print_cr("\nThe stack size specified is too small, "
4884 "Specify at least %dk",
4885 os::Linux::min_stack_allowed/ K);
4886 return JNI_ERR;
4887 }
4889 // Make the stack size a multiple of the page size so that
4890 // the yellow/red zones can be guarded.
4891 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4892 vm_page_size()));
4894 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4896 #if defined(IA32)
4897 workaround_expand_exec_shield_cs_limit();
4898 #endif
4900 Linux::libpthread_init();
4901 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4902 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4903 Linux::glibc_version(), Linux::libpthread_version(),
4904 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4905 }
4907 if (UseNUMA) {
4908 if (!Linux::libnuma_init()) {
4909 UseNUMA = false;
4910 } else {
4911 if ((Linux::numa_max_node() < 1)) {
4912 // There's only one node(they start from 0), disable NUMA.
4913 UseNUMA = false;
4914 }
4915 }
4916 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4917 // we can make the adaptive lgrp chunk resizing work. If the user specified
4918 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4919 // disable adaptive resizing.
4920 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4921 if (FLAG_IS_DEFAULT(UseNUMA)) {
4922 UseNUMA = false;
4923 } else {
4924 if (FLAG_IS_DEFAULT(UseLargePages) &&
4925 FLAG_IS_DEFAULT(UseSHM) &&
4926 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4927 UseLargePages = false;
4928 } else {
4929 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4930 UseAdaptiveSizePolicy = false;
4931 UseAdaptiveNUMAChunkSizing = false;
4932 }
4933 }
4934 }
4935 if (!UseNUMA && ForceNUMA) {
4936 UseNUMA = true;
4937 }
4938 }
4940 if (MaxFDLimit) {
4941 // set the number of file descriptors to max. print out error
4942 // if getrlimit/setrlimit fails but continue regardless.
4943 struct rlimit nbr_files;
4944 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4945 if (status != 0) {
4946 if (PrintMiscellaneous && (Verbose || WizardMode))
4947 perror("os::init_2 getrlimit failed");
4948 } else {
4949 nbr_files.rlim_cur = nbr_files.rlim_max;
4950 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4951 if (status != 0) {
4952 if (PrintMiscellaneous && (Verbose || WizardMode))
4953 perror("os::init_2 setrlimit failed");
4954 }
4955 }
4956 }
4958 // Initialize lock used to serialize thread creation (see os::create_thread)
4959 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4961 // at-exit methods are called in the reverse order of their registration.
4962 // atexit functions are called on return from main or as a result of a
4963 // call to exit(3C). There can be only 32 of these functions registered
4964 // and atexit() does not set errno.
4966 if (PerfAllowAtExitRegistration) {
4967 // only register atexit functions if PerfAllowAtExitRegistration is set.
4968 // atexit functions can be delayed until process exit time, which
4969 // can be problematic for embedded VM situations. Embedded VMs should
4970 // call DestroyJavaVM() to assure that VM resources are released.
4972 // note: perfMemory_exit_helper atexit function may be removed in
4973 // the future if the appropriate cleanup code can be added to the
4974 // VM_Exit VMOperation's doit method.
4975 if (atexit(perfMemory_exit_helper) != 0) {
4976 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4977 }
4978 }
4980 // initialize thread priority policy
4981 prio_init();
4983 return JNI_OK;
4984 }
4986 // this is called at the end of vm_initialization
4987 void os::init_3(void)
4988 {
4989 #ifdef JAVASE_EMBEDDED
4990 // Start the MemNotifyThread
4991 if (LowMemoryProtection) {
4992 MemNotifyThread::start();
4993 }
4994 return;
4995 #endif
4996 }
4998 // Mark the polling page as unreadable
4999 void os::make_polling_page_unreadable(void) {
5000 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5001 fatal("Could not disable polling page");
5002 };
5004 // Mark the polling page as readable
5005 void os::make_polling_page_readable(void) {
5006 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5007 fatal("Could not enable polling page");
5008 }
5009 };
5011 int os::active_processor_count() {
5012 // Linux doesn't yet have a (official) notion of processor sets,
5013 // so just return the number of online processors.
5014 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
5015 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
5016 return online_cpus;
5017 }
5019 void os::set_native_thread_name(const char *name) {
5020 // Not yet implemented.
5021 return;
5022 }
5024 bool os::distribute_processes(uint length, uint* distribution) {
5025 // Not yet implemented.
5026 return false;
5027 }
5029 bool os::bind_to_processor(uint processor_id) {
5030 // Not yet implemented.
5031 return false;
5032 }
5034 ///
5036 void os::SuspendedThreadTask::internal_do_task() {
5037 if (do_suspend(_thread->osthread())) {
5038 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5039 do_task(context);
5040 do_resume(_thread->osthread());
5041 }
5042 }
5044 class PcFetcher : public os::SuspendedThreadTask {
5045 public:
5046 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5047 ExtendedPC result();
5048 protected:
5049 void do_task(const os::SuspendedThreadTaskContext& context);
5050 private:
5051 ExtendedPC _epc;
5052 };
5054 ExtendedPC PcFetcher::result() {
5055 guarantee(is_done(), "task is not done yet.");
5056 return _epc;
5057 }
5059 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5060 Thread* thread = context.thread();
5061 OSThread* osthread = thread->osthread();
5062 if (osthread->ucontext() != NULL) {
5063 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5064 } else {
5065 // NULL context is unexpected, double-check this is the VMThread
5066 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5067 }
5068 }
5070 // Suspends the target using the signal mechanism and then grabs the PC before
5071 // resuming the target. Used by the flat-profiler only
5072 ExtendedPC os::get_thread_pc(Thread* thread) {
5073 // Make sure that it is called by the watcher for the VMThread
5074 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5075 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5077 PcFetcher fetcher(thread);
5078 fetcher.run();
5079 return fetcher.result();
5080 }
5082 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5083 {
5084 if (is_NPTL()) {
5085 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5086 } else {
5087 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5088 // word back to default 64bit precision if condvar is signaled. Java
5089 // wants 53bit precision. Save and restore current value.
5090 int fpu = get_fpu_control_word();
5091 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5092 set_fpu_control_word(fpu);
5093 return status;
5094 }
5095 }
5097 ////////////////////////////////////////////////////////////////////////////////
5098 // debug support
5100 bool os::find(address addr, outputStream* st) {
5101 Dl_info dlinfo;
5102 memset(&dlinfo, 0, sizeof(dlinfo));
5103 if (dladdr(addr, &dlinfo) != 0) {
5104 st->print(PTR_FORMAT ": ", addr);
5105 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5106 st->print("%s+%#x", dlinfo.dli_sname,
5107 addr - (intptr_t)dlinfo.dli_saddr);
5108 } else if (dlinfo.dli_fbase != NULL) {
5109 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5110 } else {
5111 st->print("<absolute address>");
5112 }
5113 if (dlinfo.dli_fname != NULL) {
5114 st->print(" in %s", dlinfo.dli_fname);
5115 }
5116 if (dlinfo.dli_fbase != NULL) {
5117 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5118 }
5119 st->cr();
5121 if (Verbose) {
5122 // decode some bytes around the PC
5123 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5124 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5125 address lowest = (address) dlinfo.dli_sname;
5126 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5127 if (begin < lowest) begin = lowest;
5128 Dl_info dlinfo2;
5129 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5130 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5131 end = (address) dlinfo2.dli_saddr;
5132 Disassembler::decode(begin, end, st);
5133 }
5134 return true;
5135 }
5136 return false;
5137 }
5139 ////////////////////////////////////////////////////////////////////////////////
5140 // misc
5142 // This does not do anything on Linux. This is basically a hook for being
5143 // able to use structured exception handling (thread-local exception filters)
5144 // on, e.g., Win32.
5145 void
5146 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5147 JavaCallArguments* args, Thread* thread) {
5148 f(value, method, args, thread);
5149 }
5151 void os::print_statistics() {
5152 }
5154 int os::message_box(const char* title, const char* message) {
5155 int i;
5156 fdStream err(defaultStream::error_fd());
5157 for (i = 0; i < 78; i++) err.print_raw("=");
5158 err.cr();
5159 err.print_raw_cr(title);
5160 for (i = 0; i < 78; i++) err.print_raw("-");
5161 err.cr();
5162 err.print_raw_cr(message);
5163 for (i = 0; i < 78; i++) err.print_raw("=");
5164 err.cr();
5166 char buf[16];
5167 // Prevent process from exiting upon "read error" without consuming all CPU
5168 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5170 return buf[0] == 'y' || buf[0] == 'Y';
5171 }
5173 int os::stat(const char *path, struct stat *sbuf) {
5174 char pathbuf[MAX_PATH];
5175 if (strlen(path) > MAX_PATH - 1) {
5176 errno = ENAMETOOLONG;
5177 return -1;
5178 }
5179 os::native_path(strcpy(pathbuf, path));
5180 return ::stat(pathbuf, sbuf);
5181 }
5183 bool os::check_heap(bool force) {
5184 return true;
5185 }
5187 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5188 return ::vsnprintf(buf, count, format, args);
5189 }
5191 // Is a (classpath) directory empty?
5192 bool os::dir_is_empty(const char* path) {
5193 DIR *dir = NULL;
5194 struct dirent *ptr;
5196 dir = opendir(path);
5197 if (dir == NULL) return true;
5199 /* Scan the directory */
5200 bool result = true;
5201 char buf[sizeof(struct dirent) + MAX_PATH];
5202 while (result && (ptr = ::readdir(dir)) != NULL) {
5203 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5204 result = false;
5205 }
5206 }
5207 closedir(dir);
5208 return result;
5209 }
5211 // This code originates from JDK's sysOpen and open64_w
5212 // from src/solaris/hpi/src/system_md.c
5214 #ifndef O_DELETE
5215 #define O_DELETE 0x10000
5216 #endif
5218 // Open a file. Unlink the file immediately after open returns
5219 // if the specified oflag has the O_DELETE flag set.
5220 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5222 int os::open(const char *path, int oflag, int mode) {
5224 if (strlen(path) > MAX_PATH - 1) {
5225 errno = ENAMETOOLONG;
5226 return -1;
5227 }
5228 int fd;
5229 int o_delete = (oflag & O_DELETE);
5230 oflag = oflag & ~O_DELETE;
5232 fd = ::open64(path, oflag, mode);
5233 if (fd == -1) return -1;
5235 //If the open succeeded, the file might still be a directory
5236 {
5237 struct stat64 buf64;
5238 int ret = ::fstat64(fd, &buf64);
5239 int st_mode = buf64.st_mode;
5241 if (ret != -1) {
5242 if ((st_mode & S_IFMT) == S_IFDIR) {
5243 errno = EISDIR;
5244 ::close(fd);
5245 return -1;
5246 }
5247 } else {
5248 ::close(fd);
5249 return -1;
5250 }
5251 }
5253 /*
5254 * All file descriptors that are opened in the JVM and not
5255 * specifically destined for a subprocess should have the
5256 * close-on-exec flag set. If we don't set it, then careless 3rd
5257 * party native code might fork and exec without closing all
5258 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5259 * UNIXProcess.c), and this in turn might:
5260 *
5261 * - cause end-of-file to fail to be detected on some file
5262 * descriptors, resulting in mysterious hangs, or
5263 *
5264 * - might cause an fopen in the subprocess to fail on a system
5265 * suffering from bug 1085341.
5266 *
5267 * (Yes, the default setting of the close-on-exec flag is a Unix
5268 * design flaw)
5269 *
5270 * See:
5271 * 1085341: 32-bit stdio routines should support file descriptors >255
5272 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5273 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5274 */
5275 #ifdef FD_CLOEXEC
5276 {
5277 int flags = ::fcntl(fd, F_GETFD);
5278 if (flags != -1)
5279 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5280 }
5281 #endif
5283 if (o_delete != 0) {
5284 ::unlink(path);
5285 }
5286 return fd;
5287 }
5290 // create binary file, rewriting existing file if required
5291 int os::create_binary_file(const char* path, bool rewrite_existing) {
5292 int oflags = O_WRONLY | O_CREAT;
5293 if (!rewrite_existing) {
5294 oflags |= O_EXCL;
5295 }
5296 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5297 }
5299 // return current position of file pointer
5300 jlong os::current_file_offset(int fd) {
5301 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5302 }
5304 // move file pointer to the specified offset
5305 jlong os::seek_to_file_offset(int fd, jlong offset) {
5306 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5307 }
5309 // This code originates from JDK's sysAvailable
5310 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5312 int os::available(int fd, jlong *bytes) {
5313 jlong cur, end;
5314 int mode;
5315 struct stat64 buf64;
5317 if (::fstat64(fd, &buf64) >= 0) {
5318 mode = buf64.st_mode;
5319 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5320 /*
5321 * XXX: is the following call interruptible? If so, this might
5322 * need to go through the INTERRUPT_IO() wrapper as for other
5323 * blocking, interruptible calls in this file.
5324 */
5325 int n;
5326 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5327 *bytes = n;
5328 return 1;
5329 }
5330 }
5331 }
5332 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5333 return 0;
5334 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5335 return 0;
5336 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5337 return 0;
5338 }
5339 *bytes = end - cur;
5340 return 1;
5341 }
5343 int os::socket_available(int fd, jint *pbytes) {
5344 // Linux doc says EINTR not returned, unlike Solaris
5345 int ret = ::ioctl(fd, FIONREAD, pbytes);
5347 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5348 // is expected to return 0 on failure and 1 on success to the jdk.
5349 return (ret < 0) ? 0 : 1;
5350 }
5352 // Map a block of memory.
5353 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5354 char *addr, size_t bytes, bool read_only,
5355 bool allow_exec) {
5356 int prot;
5357 int flags = MAP_PRIVATE;
5359 if (read_only) {
5360 prot = PROT_READ;
5361 } else {
5362 prot = PROT_READ | PROT_WRITE;
5363 }
5365 if (allow_exec) {
5366 prot |= PROT_EXEC;
5367 }
5369 if (addr != NULL) {
5370 flags |= MAP_FIXED;
5371 }
5373 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5374 fd, file_offset);
5375 if (mapped_address == MAP_FAILED) {
5376 return NULL;
5377 }
5378 return mapped_address;
5379 }
5382 // Remap a block of memory.
5383 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5384 char *addr, size_t bytes, bool read_only,
5385 bool allow_exec) {
5386 // same as map_memory() on this OS
5387 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5388 allow_exec);
5389 }
5392 // Unmap a block of memory.
5393 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5394 return munmap(addr, bytes) == 0;
5395 }
5397 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5399 static clockid_t thread_cpu_clockid(Thread* thread) {
5400 pthread_t tid = thread->osthread()->pthread_id();
5401 clockid_t clockid;
5403 // Get thread clockid
5404 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5405 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5406 return clockid;
5407 }
5409 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5410 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5411 // of a thread.
5412 //
5413 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5414 // the fast estimate available on the platform.
5416 jlong os::current_thread_cpu_time() {
5417 if (os::Linux::supports_fast_thread_cpu_time()) {
5418 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5419 } else {
5420 // return user + sys since the cost is the same
5421 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5422 }
5423 }
5425 jlong os::thread_cpu_time(Thread* thread) {
5426 // consistent with what current_thread_cpu_time() returns
5427 if (os::Linux::supports_fast_thread_cpu_time()) {
5428 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5429 } else {
5430 return slow_thread_cpu_time(thread, true /* user + sys */);
5431 }
5432 }
5434 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5435 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5436 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5437 } else {
5438 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5439 }
5440 }
5442 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5443 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5444 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5445 } else {
5446 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5447 }
5448 }
5450 //
5451 // -1 on error.
5452 //
5454 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5455 static bool proc_task_unchecked = true;
5456 static const char *proc_stat_path = "/proc/%d/stat";
5457 pid_t tid = thread->osthread()->thread_id();
5458 char *s;
5459 char stat[2048];
5460 int statlen;
5461 char proc_name[64];
5462 int count;
5463 long sys_time, user_time;
5464 char cdummy;
5465 int idummy;
5466 long ldummy;
5467 FILE *fp;
5469 // The /proc/<tid>/stat aggregates per-process usage on
5470 // new Linux kernels 2.6+ where NPTL is supported.
5471 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5472 // See bug 6328462.
5473 // There possibly can be cases where there is no directory
5474 // /proc/self/task, so we check its availability.
5475 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5476 // This is executed only once
5477 proc_task_unchecked = false;
5478 fp = fopen("/proc/self/task", "r");
5479 if (fp != NULL) {
5480 proc_stat_path = "/proc/self/task/%d/stat";
5481 fclose(fp);
5482 }
5483 }
5485 sprintf(proc_name, proc_stat_path, tid);
5486 fp = fopen(proc_name, "r");
5487 if ( fp == NULL ) return -1;
5488 statlen = fread(stat, 1, 2047, fp);
5489 stat[statlen] = '\0';
5490 fclose(fp);
5492 // Skip pid and the command string. Note that we could be dealing with
5493 // weird command names, e.g. user could decide to rename java launcher
5494 // to "java 1.4.2 :)", then the stat file would look like
5495 // 1234 (java 1.4.2 :)) R ... ...
5496 // We don't really need to know the command string, just find the last
5497 // occurrence of ")" and then start parsing from there. See bug 4726580.
5498 s = strrchr(stat, ')');
5499 if (s == NULL ) return -1;
5501 // Skip blank chars
5502 do s++; while (isspace(*s));
5504 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5505 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5506 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5507 &user_time, &sys_time);
5508 if ( count != 13 ) return -1;
5509 if (user_sys_cpu_time) {
5510 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5511 } else {
5512 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5513 }
5514 }
5516 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5517 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5518 info_ptr->may_skip_backward = false; // elapsed time not wall time
5519 info_ptr->may_skip_forward = false; // elapsed time not wall time
5520 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5521 }
5523 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5524 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5525 info_ptr->may_skip_backward = false; // elapsed time not wall time
5526 info_ptr->may_skip_forward = false; // elapsed time not wall time
5527 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5528 }
5530 bool os::is_thread_cpu_time_supported() {
5531 return true;
5532 }
5534 // System loadavg support. Returns -1 if load average cannot be obtained.
5535 // Linux doesn't yet have a (official) notion of processor sets,
5536 // so just return the system wide load average.
5537 int os::loadavg(double loadavg[], int nelem) {
5538 return ::getloadavg(loadavg, nelem);
5539 }
5541 void os::pause() {
5542 char filename[MAX_PATH];
5543 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5544 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5545 } else {
5546 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5547 }
5549 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5550 if (fd != -1) {
5551 struct stat buf;
5552 ::close(fd);
5553 while (::stat(filename, &buf) == 0) {
5554 (void)::poll(NULL, 0, 100);
5555 }
5556 } else {
5557 jio_fprintf(stderr,
5558 "Could not open pause file '%s', continuing immediately.\n", filename);
5559 }
5560 }
5563 // Refer to the comments in os_solaris.cpp park-unpark.
5564 //
5565 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5566 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5567 // For specifics regarding the bug see GLIBC BUGID 261237 :
5568 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5569 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5570 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5571 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5572 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5573 // and monitorenter when we're using 1-0 locking. All those operations may result in
5574 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5575 // of libpthread avoids the problem, but isn't practical.
5576 //
5577 // Possible remedies:
5578 //
5579 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5580 // This is palliative and probabilistic, however. If the thread is preempted
5581 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5582 // than the minimum period may have passed, and the abstime may be stale (in the
5583 // past) resultin in a hang. Using this technique reduces the odds of a hang
5584 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5585 //
5586 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5587 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5588 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5589 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5590 // thread.
5591 //
5592 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5593 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5594 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5595 // This also works well. In fact it avoids kernel-level scalability impediments
5596 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5597 // timers in a graceful fashion.
5598 //
5599 // 4. When the abstime value is in the past it appears that control returns
5600 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5601 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5602 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5603 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5604 // It may be possible to avoid reinitialization by checking the return
5605 // value from pthread_cond_timedwait(). In addition to reinitializing the
5606 // condvar we must establish the invariant that cond_signal() is only called
5607 // within critical sections protected by the adjunct mutex. This prevents
5608 // cond_signal() from "seeing" a condvar that's in the midst of being
5609 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5610 // desirable signal-after-unlock optimization that avoids futile context switching.
5611 //
5612 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5613 // structure when a condvar is used or initialized. cond_destroy() would
5614 // release the helper structure. Our reinitialize-after-timedwait fix
5615 // put excessive stress on malloc/free and locks protecting the c-heap.
5616 //
5617 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5618 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5619 // and only enabling the work-around for vulnerable environments.
5621 // utility to compute the abstime argument to timedwait:
5622 // millis is the relative timeout time
5623 // abstime will be the absolute timeout time
5624 // TODO: replace compute_abstime() with unpackTime()
5626 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5627 if (millis < 0) millis = 0;
5629 jlong seconds = millis / 1000;
5630 millis %= 1000;
5631 if (seconds > 50000000) { // see man cond_timedwait(3T)
5632 seconds = 50000000;
5633 }
5635 if (os::Linux::supports_monotonic_clock()) {
5636 struct timespec now;
5637 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5638 assert_status(status == 0, status, "clock_gettime");
5639 abstime->tv_sec = now.tv_sec + seconds;
5640 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5641 if (nanos >= NANOSECS_PER_SEC) {
5642 abstime->tv_sec += 1;
5643 nanos -= NANOSECS_PER_SEC;
5644 }
5645 abstime->tv_nsec = nanos;
5646 } else {
5647 struct timeval now;
5648 int status = gettimeofday(&now, NULL);
5649 assert(status == 0, "gettimeofday");
5650 abstime->tv_sec = now.tv_sec + seconds;
5651 long usec = now.tv_usec + millis * 1000;
5652 if (usec >= 1000000) {
5653 abstime->tv_sec += 1;
5654 usec -= 1000000;
5655 }
5656 abstime->tv_nsec = usec * 1000;
5657 }
5658 return abstime;
5659 }
5662 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5663 // Conceptually TryPark() should be equivalent to park(0).
5665 int os::PlatformEvent::TryPark() {
5666 for (;;) {
5667 const int v = _Event ;
5668 guarantee ((v == 0) || (v == 1), "invariant") ;
5669 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5670 }
5671 }
5673 void os::PlatformEvent::park() { // AKA "down()"
5674 // Invariant: Only the thread associated with the Event/PlatformEvent
5675 // may call park().
5676 // TODO: assert that _Assoc != NULL or _Assoc == Self
5677 int v ;
5678 for (;;) {
5679 v = _Event ;
5680 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5681 }
5682 guarantee (v >= 0, "invariant") ;
5683 if (v == 0) {
5684 // Do this the hard way by blocking ...
5685 int status = pthread_mutex_lock(_mutex);
5686 assert_status(status == 0, status, "mutex_lock");
5687 guarantee (_nParked == 0, "invariant") ;
5688 ++ _nParked ;
5689 while (_Event < 0) {
5690 status = pthread_cond_wait(_cond, _mutex);
5691 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5692 // Treat this the same as if the wait was interrupted
5693 if (status == ETIME) { status = EINTR; }
5694 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5695 }
5696 -- _nParked ;
5698 _Event = 0 ;
5699 status = pthread_mutex_unlock(_mutex);
5700 assert_status(status == 0, status, "mutex_unlock");
5701 // Paranoia to ensure our locked and lock-free paths interact
5702 // correctly with each other.
5703 OrderAccess::fence();
5704 }
5705 guarantee (_Event >= 0, "invariant") ;
5706 }
5708 int os::PlatformEvent::park(jlong millis) {
5709 guarantee (_nParked == 0, "invariant") ;
5711 int v ;
5712 for (;;) {
5713 v = _Event ;
5714 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5715 }
5716 guarantee (v >= 0, "invariant") ;
5717 if (v != 0) return OS_OK ;
5719 // We do this the hard way, by blocking the thread.
5720 // Consider enforcing a minimum timeout value.
5721 struct timespec abst;
5722 compute_abstime(&abst, millis);
5724 int ret = OS_TIMEOUT;
5725 int status = pthread_mutex_lock(_mutex);
5726 assert_status(status == 0, status, "mutex_lock");
5727 guarantee (_nParked == 0, "invariant") ;
5728 ++_nParked ;
5730 // Object.wait(timo) will return because of
5731 // (a) notification
5732 // (b) timeout
5733 // (c) thread.interrupt
5734 //
5735 // Thread.interrupt and object.notify{All} both call Event::set.
5736 // That is, we treat thread.interrupt as a special case of notification.
5737 // The underlying Solaris implementation, cond_timedwait, admits
5738 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5739 // JVM from making those visible to Java code. As such, we must
5740 // filter out spurious wakeups. We assume all ETIME returns are valid.
5741 //
5742 // TODO: properly differentiate simultaneous notify+interrupt.
5743 // In that case, we should propagate the notify to another waiter.
5745 while (_Event < 0) {
5746 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5747 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5748 pthread_cond_destroy (_cond);
5749 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5750 }
5751 assert_status(status == 0 || status == EINTR ||
5752 status == ETIME || status == ETIMEDOUT,
5753 status, "cond_timedwait");
5754 if (!FilterSpuriousWakeups) break ; // previous semantics
5755 if (status == ETIME || status == ETIMEDOUT) break ;
5756 // We consume and ignore EINTR and spurious wakeups.
5757 }
5758 --_nParked ;
5759 if (_Event >= 0) {
5760 ret = OS_OK;
5761 }
5762 _Event = 0 ;
5763 status = pthread_mutex_unlock(_mutex);
5764 assert_status(status == 0, status, "mutex_unlock");
5765 assert (_nParked == 0, "invariant") ;
5766 // Paranoia to ensure our locked and lock-free paths interact
5767 // correctly with each other.
5768 OrderAccess::fence();
5769 return ret;
5770 }
5772 void os::PlatformEvent::unpark() {
5773 // Transitions for _Event:
5774 // 0 :=> 1
5775 // 1 :=> 1
5776 // -1 :=> either 0 or 1; must signal target thread
5777 // That is, we can safely transition _Event from -1 to either
5778 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5779 // unpark() calls.
5780 // See also: "Semaphores in Plan 9" by Mullender & Cox
5781 //
5782 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5783 // that it will take two back-to-back park() calls for the owning
5784 // thread to block. This has the benefit of forcing a spurious return
5785 // from the first park() call after an unpark() call which will help
5786 // shake out uses of park() and unpark() without condition variables.
5788 if (Atomic::xchg(1, &_Event) >= 0) return;
5790 // Wait for the thread associated with the event to vacate
5791 int status = pthread_mutex_lock(_mutex);
5792 assert_status(status == 0, status, "mutex_lock");
5793 int AnyWaiters = _nParked;
5794 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5795 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5796 AnyWaiters = 0;
5797 pthread_cond_signal(_cond);
5798 }
5799 status = pthread_mutex_unlock(_mutex);
5800 assert_status(status == 0, status, "mutex_unlock");
5801 if (AnyWaiters != 0) {
5802 status = pthread_cond_signal(_cond);
5803 assert_status(status == 0, status, "cond_signal");
5804 }
5806 // Note that we signal() _after dropping the lock for "immortal" Events.
5807 // This is safe and avoids a common class of futile wakeups. In rare
5808 // circumstances this can cause a thread to return prematurely from
5809 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5810 // simply re-test the condition and re-park itself.
5811 }
5814 // JSR166
5815 // -------------------------------------------------------
5817 /*
5818 * The solaris and linux implementations of park/unpark are fairly
5819 * conservative for now, but can be improved. They currently use a
5820 * mutex/condvar pair, plus a a count.
5821 * Park decrements count if > 0, else does a condvar wait. Unpark
5822 * sets count to 1 and signals condvar. Only one thread ever waits
5823 * on the condvar. Contention seen when trying to park implies that someone
5824 * is unparking you, so don't wait. And spurious returns are fine, so there
5825 * is no need to track notifications.
5826 */
5828 /*
5829 * This code is common to linux and solaris and will be moved to a
5830 * common place in dolphin.
5831 *
5832 * The passed in time value is either a relative time in nanoseconds
5833 * or an absolute time in milliseconds. Either way it has to be unpacked
5834 * into suitable seconds and nanoseconds components and stored in the
5835 * given timespec structure.
5836 * Given time is a 64-bit value and the time_t used in the timespec is only
5837 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5838 * overflow if times way in the future are given. Further on Solaris versions
5839 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5840 * number of seconds, in abstime, is less than current_time + 100,000,000.
5841 * As it will be 28 years before "now + 100000000" will overflow we can
5842 * ignore overflow and just impose a hard-limit on seconds using the value
5843 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5844 * years from "now".
5845 */
5847 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5848 assert (time > 0, "convertTime");
5849 time_t max_secs = 0;
5851 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5852 struct timeval now;
5853 int status = gettimeofday(&now, NULL);
5854 assert(status == 0, "gettimeofday");
5856 max_secs = now.tv_sec + MAX_SECS;
5858 if (isAbsolute) {
5859 jlong secs = time / 1000;
5860 if (secs > max_secs) {
5861 absTime->tv_sec = max_secs;
5862 } else {
5863 absTime->tv_sec = secs;
5864 }
5865 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5866 } else {
5867 jlong secs = time / NANOSECS_PER_SEC;
5868 if (secs >= MAX_SECS) {
5869 absTime->tv_sec = max_secs;
5870 absTime->tv_nsec = 0;
5871 } else {
5872 absTime->tv_sec = now.tv_sec + secs;
5873 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5874 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5875 absTime->tv_nsec -= NANOSECS_PER_SEC;
5876 ++absTime->tv_sec; // note: this must be <= max_secs
5877 }
5878 }
5879 }
5880 } else {
5881 // must be relative using monotonic clock
5882 struct timespec now;
5883 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5884 assert_status(status == 0, status, "clock_gettime");
5885 max_secs = now.tv_sec + MAX_SECS;
5886 jlong secs = time / NANOSECS_PER_SEC;
5887 if (secs >= MAX_SECS) {
5888 absTime->tv_sec = max_secs;
5889 absTime->tv_nsec = 0;
5890 } else {
5891 absTime->tv_sec = now.tv_sec + secs;
5892 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5893 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5894 absTime->tv_nsec -= NANOSECS_PER_SEC;
5895 ++absTime->tv_sec; // note: this must be <= max_secs
5896 }
5897 }
5898 }
5899 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5900 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5901 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5902 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5903 }
5905 void Parker::park(bool isAbsolute, jlong time) {
5906 // Ideally we'd do something useful while spinning, such
5907 // as calling unpackTime().
5909 // Optional fast-path check:
5910 // Return immediately if a permit is available.
5911 // We depend on Atomic::xchg() having full barrier semantics
5912 // since we are doing a lock-free update to _counter.
5913 if (Atomic::xchg(0, &_counter) > 0) return;
5915 Thread* thread = Thread::current();
5916 assert(thread->is_Java_thread(), "Must be JavaThread");
5917 JavaThread *jt = (JavaThread *)thread;
5919 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5920 // Check interrupt before trying to wait
5921 if (Thread::is_interrupted(thread, false)) {
5922 return;
5923 }
5925 // Next, demultiplex/decode time arguments
5926 timespec absTime;
5927 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5928 return;
5929 }
5930 if (time > 0) {
5931 unpackTime(&absTime, isAbsolute, time);
5932 }
5935 // Enter safepoint region
5936 // Beware of deadlocks such as 6317397.
5937 // The per-thread Parker:: mutex is a classic leaf-lock.
5938 // In particular a thread must never block on the Threads_lock while
5939 // holding the Parker:: mutex. If safepoints are pending both the
5940 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5941 ThreadBlockInVM tbivm(jt);
5943 // Don't wait if cannot get lock since interference arises from
5944 // unblocking. Also. check interrupt before trying wait
5945 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5946 return;
5947 }
5949 int status ;
5950 if (_counter > 0) { // no wait needed
5951 _counter = 0;
5952 status = pthread_mutex_unlock(_mutex);
5953 assert (status == 0, "invariant") ;
5954 // Paranoia to ensure our locked and lock-free paths interact
5955 // correctly with each other and Java-level accesses.
5956 OrderAccess::fence();
5957 return;
5958 }
5960 #ifdef ASSERT
5961 // Don't catch signals while blocked; let the running threads have the signals.
5962 // (This allows a debugger to break into the running thread.)
5963 sigset_t oldsigs;
5964 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5965 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5966 #endif
5968 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5969 jt->set_suspend_equivalent();
5970 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5972 assert(_cur_index == -1, "invariant");
5973 if (time == 0) {
5974 _cur_index = REL_INDEX; // arbitrary choice when not timed
5975 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5976 } else {
5977 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5978 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5979 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5980 pthread_cond_destroy (&_cond[_cur_index]) ;
5981 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5982 }
5983 }
5984 _cur_index = -1;
5985 assert_status(status == 0 || status == EINTR ||
5986 status == ETIME || status == ETIMEDOUT,
5987 status, "cond_timedwait");
5989 #ifdef ASSERT
5990 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5991 #endif
5993 _counter = 0 ;
5994 status = pthread_mutex_unlock(_mutex) ;
5995 assert_status(status == 0, status, "invariant") ;
5996 // Paranoia to ensure our locked and lock-free paths interact
5997 // correctly with each other and Java-level accesses.
5998 OrderAccess::fence();
6000 // If externally suspended while waiting, re-suspend
6001 if (jt->handle_special_suspend_equivalent_condition()) {
6002 jt->java_suspend_self();
6003 }
6004 }
6006 void Parker::unpark() {
6007 int s, status ;
6008 status = pthread_mutex_lock(_mutex);
6009 assert (status == 0, "invariant") ;
6010 s = _counter;
6011 _counter = 1;
6012 if (s < 1) {
6013 // thread might be parked
6014 if (_cur_index != -1) {
6015 // thread is definitely parked
6016 if (WorkAroundNPTLTimedWaitHang) {
6017 status = pthread_cond_signal (&_cond[_cur_index]);
6018 assert (status == 0, "invariant");
6019 status = pthread_mutex_unlock(_mutex);
6020 assert (status == 0, "invariant");
6021 } else {
6022 status = pthread_mutex_unlock(_mutex);
6023 assert (status == 0, "invariant");
6024 status = pthread_cond_signal (&_cond[_cur_index]);
6025 assert (status == 0, "invariant");
6026 }
6027 } else {
6028 pthread_mutex_unlock(_mutex);
6029 assert (status == 0, "invariant") ;
6030 }
6031 } else {
6032 pthread_mutex_unlock(_mutex);
6033 assert (status == 0, "invariant") ;
6034 }
6035 }
6038 extern char** environ;
6040 #ifndef __NR_fork
6041 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
6042 #endif
6044 #ifndef __NR_execve
6045 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
6046 #endif
6048 // Run the specified command in a separate process. Return its exit value,
6049 // or -1 on failure (e.g. can't fork a new process).
6050 // Unlike system(), this function can be called from signal handler. It
6051 // doesn't block SIGINT et al.
6052 int os::fork_and_exec(char* cmd) {
6053 const char * argv[4] = {"sh", "-c", cmd, NULL};
6055 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
6056 // pthread_atfork handlers and reset pthread library. All we need is a
6057 // separate process to execve. Make a direct syscall to fork process.
6058 // On IA64 there's no fork syscall, we have to use fork() and hope for
6059 // the best...
6060 pid_t pid = NOT_IA64(syscall(__NR_fork);)
6061 IA64_ONLY(fork();)
6063 if (pid < 0) {
6064 // fork failed
6065 return -1;
6067 } else if (pid == 0) {
6068 // child process
6070 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6071 // first to kill every thread on the thread list. Because this list is
6072 // not reset by fork() (see notes above), execve() will instead kill
6073 // every thread in the parent process. We know this is the only thread
6074 // in the new process, so make a system call directly.
6075 // IA64 should use normal execve() from glibc to match the glibc fork()
6076 // above.
6077 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6078 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6080 // execve failed
6081 _exit(-1);
6083 } else {
6084 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6085 // care about the actual exit code, for now.
6087 int status;
6089 // Wait for the child process to exit. This returns immediately if
6090 // the child has already exited. */
6091 while (waitpid(pid, &status, 0) < 0) {
6092 switch (errno) {
6093 case ECHILD: return 0;
6094 case EINTR: break;
6095 default: return -1;
6096 }
6097 }
6099 if (WIFEXITED(status)) {
6100 // The child exited normally; get its exit code.
6101 return WEXITSTATUS(status);
6102 } else if (WIFSIGNALED(status)) {
6103 // The child exited because of a signal
6104 // The best value to return is 0x80 + signal number,
6105 // because that is what all Unix shells do, and because
6106 // it allows callers to distinguish between process exit and
6107 // process death by signal.
6108 return 0x80 + WTERMSIG(status);
6109 } else {
6110 // Unknown exit code; pass it through
6111 return status;
6112 }
6113 }
6114 }
6116 // is_headless_jre()
6117 //
6118 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6119 // in order to report if we are running in a headless jre
6120 //
6121 // Since JDK8 xawt/libmawt.so was moved into the same directory
6122 // as libawt.so, and renamed libawt_xawt.so
6123 //
6124 bool os::is_headless_jre() {
6125 struct stat statbuf;
6126 char buf[MAXPATHLEN];
6127 char libmawtpath[MAXPATHLEN];
6128 const char *xawtstr = "/xawt/libmawt.so";
6129 const char *new_xawtstr = "/libawt_xawt.so";
6130 char *p;
6132 // Get path to libjvm.so
6133 os::jvm_path(buf, sizeof(buf));
6135 // Get rid of libjvm.so
6136 p = strrchr(buf, '/');
6137 if (p == NULL) return false;
6138 else *p = '\0';
6140 // Get rid of client or server
6141 p = strrchr(buf, '/');
6142 if (p == NULL) return false;
6143 else *p = '\0';
6145 // check xawt/libmawt.so
6146 strcpy(libmawtpath, buf);
6147 strcat(libmawtpath, xawtstr);
6148 if (::stat(libmawtpath, &statbuf) == 0) return false;
6150 // check libawt_xawt.so
6151 strcpy(libmawtpath, buf);
6152 strcat(libmawtpath, new_xawtstr);
6153 if (::stat(libmawtpath, &statbuf) == 0) return false;
6155 return true;
6156 }
6158 // Get the default path to the core file
6159 // Returns the length of the string
6160 int os::get_core_path(char* buffer, size_t bufferSize) {
6161 const char* p = get_current_directory(buffer, bufferSize);
6163 if (p == NULL) {
6164 assert(p != NULL, "failed to get current directory");
6165 return 0;
6166 }
6168 return strlen(buffer);
6169 }
6171 #ifdef JAVASE_EMBEDDED
6172 //
6173 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6174 //
6175 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6177 // ctor
6178 //
6179 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6180 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6181 _fd = fd;
6183 if (os::create_thread(this, os::os_thread)) {
6184 _memnotify_thread = this;
6185 os::set_priority(this, NearMaxPriority);
6186 os::start_thread(this);
6187 }
6188 }
6190 // Where all the work gets done
6191 //
6192 void MemNotifyThread::run() {
6193 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6195 // Set up the select arguments
6196 fd_set rfds;
6197 if (_fd != -1) {
6198 FD_ZERO(&rfds);
6199 FD_SET(_fd, &rfds);
6200 }
6202 // Now wait for the mem_notify device to wake up
6203 while (1) {
6204 // Wait for the mem_notify device to signal us..
6205 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6206 if (rc == -1) {
6207 perror("select!\n");
6208 break;
6209 } else if (rc) {
6210 //ssize_t free_before = os::available_memory();
6211 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6213 // The kernel is telling us there is not much memory left...
6214 // try to do something about that
6216 // If we are not already in a GC, try one.
6217 if (!Universe::heap()->is_gc_active()) {
6218 Universe::heap()->collect(GCCause::_allocation_failure);
6220 //ssize_t free_after = os::available_memory();
6221 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6222 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6223 }
6224 // We might want to do something like the following if we find the GC's are not helping...
6225 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6226 }
6227 }
6228 }
6230 //
6231 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6232 //
6233 void MemNotifyThread::start() {
6234 int fd;
6235 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6236 if (fd < 0) {
6237 return;
6238 }
6240 if (memnotify_thread() == NULL) {
6241 new MemNotifyThread(fd);
6242 }
6243 }
6245 #endif // JAVASE_EMBEDDED
6248 /////////////// Unit tests ///////////////
6250 #ifndef PRODUCT
6252 #define test_log(...) \
6253 do {\
6254 if (VerboseInternalVMTests) { \
6255 tty->print_cr(__VA_ARGS__); \
6256 tty->flush(); \
6257 }\
6258 } while (false)
6260 class TestReserveMemorySpecial : AllStatic {
6261 public:
6262 static void small_page_write(void* addr, size_t size) {
6263 size_t page_size = os::vm_page_size();
6265 char* end = (char*)addr + size;
6266 for (char* p = (char*)addr; p < end; p += page_size) {
6267 *p = 1;
6268 }
6269 }
6271 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6272 if (!UseHugeTLBFS) {
6273 return;
6274 }
6276 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6278 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6280 if (addr != NULL) {
6281 small_page_write(addr, size);
6283 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6284 }
6285 }
6287 static void test_reserve_memory_special_huge_tlbfs_only() {
6288 if (!UseHugeTLBFS) {
6289 return;
6290 }
6292 size_t lp = os::large_page_size();
6294 for (size_t size = lp; size <= lp * 10; size += lp) {
6295 test_reserve_memory_special_huge_tlbfs_only(size);
6296 }
6297 }
6299 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6300 if (!UseHugeTLBFS) {
6301 return;
6302 }
6304 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6305 size, alignment);
6307 assert(size >= os::large_page_size(), "Incorrect input to test");
6309 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6311 if (addr != NULL) {
6312 small_page_write(addr, size);
6314 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6315 }
6316 }
6318 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6319 size_t lp = os::large_page_size();
6320 size_t ag = os::vm_allocation_granularity();
6322 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6323 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6324 }
6325 }
6327 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6328 size_t lp = os::large_page_size();
6329 size_t ag = os::vm_allocation_granularity();
6331 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6332 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6333 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6334 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6335 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6336 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6337 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6338 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6339 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6340 }
6342 static void test_reserve_memory_special_huge_tlbfs() {
6343 if (!UseHugeTLBFS) {
6344 return;
6345 }
6347 test_reserve_memory_special_huge_tlbfs_only();
6348 test_reserve_memory_special_huge_tlbfs_mixed();
6349 }
6351 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6352 if (!UseSHM) {
6353 return;
6354 }
6356 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6358 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6360 if (addr != NULL) {
6361 assert(is_ptr_aligned(addr, alignment), "Check");
6362 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6364 small_page_write(addr, size);
6366 os::Linux::release_memory_special_shm(addr, size);
6367 }
6368 }
6370 static void test_reserve_memory_special_shm() {
6371 size_t lp = os::large_page_size();
6372 size_t ag = os::vm_allocation_granularity();
6374 for (size_t size = ag; size < lp * 3; size += ag) {
6375 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6376 test_reserve_memory_special_shm(size, alignment);
6377 }
6378 }
6379 }
6381 static void test() {
6382 test_reserve_memory_special_huge_tlbfs();
6383 test_reserve_memory_special_shm();
6384 }
6385 };
6387 void TestReserveMemorySpecial_test() {
6388 TestReserveMemorySpecial::test();
6389 }
6391 #endif