Fri, 30 Nov 2012 15:23:16 -0800
8003240: x86: move MacroAssembler into separate file
Reviewed-by: kvn
1 /*
2 * Copyright (c) 1999, 2012, 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/java.hpp"
48 #include "runtime/javaCalls.hpp"
49 #include "runtime/mutexLocker.hpp"
50 #include "runtime/objectMonitor.hpp"
51 #include "runtime/osThread.hpp"
52 #include "runtime/perfMemory.hpp"
53 #include "runtime/sharedRuntime.hpp"
54 #include "runtime/statSampler.hpp"
55 #include "runtime/stubRoutines.hpp"
56 #include "runtime/threadCritical.hpp"
57 #include "runtime/timer.hpp"
58 #include "services/attachListener.hpp"
59 #include "services/runtimeService.hpp"
60 #include "thread_linux.inline.hpp"
61 #include "utilities/decoder.hpp"
62 #include "utilities/defaultStream.hpp"
63 #include "utilities/events.hpp"
64 #include "utilities/growableArray.hpp"
65 #include "utilities/vmError.hpp"
67 // put OS-includes here
68 # include <sys/types.h>
69 # include <sys/mman.h>
70 # include <sys/stat.h>
71 # include <sys/select.h>
72 # include <pthread.h>
73 # include <signal.h>
74 # include <errno.h>
75 # include <dlfcn.h>
76 # include <stdio.h>
77 # include <unistd.h>
78 # include <sys/resource.h>
79 # include <pthread.h>
80 # include <sys/stat.h>
81 # include <sys/time.h>
82 # include <sys/times.h>
83 # include <sys/utsname.h>
84 # include <sys/socket.h>
85 # include <sys/wait.h>
86 # include <pwd.h>
87 # include <poll.h>
88 # include <semaphore.h>
89 # include <fcntl.h>
90 # include <string.h>
91 # include <syscall.h>
92 # include <sys/sysinfo.h>
93 # include <gnu/libc-version.h>
94 # include <sys/ipc.h>
95 # include <sys/shm.h>
96 # include <link.h>
97 # include <stdint.h>
98 # include <inttypes.h>
99 # include <sys/ioctl.h>
101 #define MAX_PATH (2 * K)
103 // for timer info max values which include all bits
104 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
106 #define LARGEPAGES_BIT (1 << 6)
107 ////////////////////////////////////////////////////////////////////////////////
108 // global variables
109 julong os::Linux::_physical_memory = 0;
111 address os::Linux::_initial_thread_stack_bottom = NULL;
112 uintptr_t os::Linux::_initial_thread_stack_size = 0;
114 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
115 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
116 Mutex* os::Linux::_createThread_lock = NULL;
117 pthread_t os::Linux::_main_thread;
118 int os::Linux::_page_size = -1;
119 bool os::Linux::_is_floating_stack = false;
120 bool os::Linux::_is_NPTL = false;
121 bool os::Linux::_supports_fast_thread_cpu_time = false;
122 const char * os::Linux::_glibc_version = NULL;
123 const char * os::Linux::_libpthread_version = NULL;
125 static jlong initial_time_count=0;
127 static int clock_tics_per_sec = 100;
129 // For diagnostics to print a message once. see run_periodic_checks
130 static sigset_t check_signal_done;
131 static bool check_signals = true;;
133 static pid_t _initial_pid = 0;
135 /* Signal number used to suspend/resume a thread */
137 /* do not use any signal number less than SIGSEGV, see 4355769 */
138 static int SR_signum = SIGUSR2;
139 sigset_t SR_sigset;
141 /* Used to protect dlsym() calls */
142 static pthread_mutex_t dl_mutex;
144 #ifdef JAVASE_EMBEDDED
145 class MemNotifyThread: public Thread {
146 friend class VMStructs;
147 public:
148 virtual void run();
150 private:
151 static MemNotifyThread* _memnotify_thread;
152 int _fd;
154 public:
156 // Constructor
157 MemNotifyThread(int fd);
159 // Tester
160 bool is_memnotify_thread() const { return true; }
162 // Printing
163 char* name() const { return (char*)"Linux MemNotify Thread"; }
165 // Returns the single instance of the MemNotifyThread
166 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
168 // Create and start the single instance of MemNotifyThread
169 static void start();
170 };
171 #endif // JAVASE_EMBEDDED
173 // utility functions
175 static int SR_initialize();
176 static int SR_finalize();
178 julong os::available_memory() {
179 return Linux::available_memory();
180 }
182 julong os::Linux::available_memory() {
183 // values in struct sysinfo are "unsigned long"
184 struct sysinfo si;
185 sysinfo(&si);
187 return (julong)si.freeram * si.mem_unit;
188 }
190 julong os::physical_memory() {
191 return Linux::physical_memory();
192 }
194 julong os::allocatable_physical_memory(julong size) {
195 #ifdef _LP64
196 return size;
197 #else
198 julong result = MIN2(size, (julong)3800*M);
199 if (!is_allocatable(result)) {
200 // See comments under solaris for alignment considerations
201 julong reasonable_size = (julong)2*G - 2 * os::vm_page_size();
202 result = MIN2(size, reasonable_size);
203 }
204 return result;
205 #endif // _LP64
206 }
208 ////////////////////////////////////////////////////////////////////////////////
209 // environment support
211 bool os::getenv(const char* name, char* buf, int len) {
212 const char* val = ::getenv(name);
213 if (val != NULL && strlen(val) < (size_t)len) {
214 strcpy(buf, val);
215 return true;
216 }
217 if (len > 0) buf[0] = 0; // return a null string
218 return false;
219 }
222 // Return true if user is running as root.
224 bool os::have_special_privileges() {
225 static bool init = false;
226 static bool privileges = false;
227 if (!init) {
228 privileges = (getuid() != geteuid()) || (getgid() != getegid());
229 init = true;
230 }
231 return privileges;
232 }
235 #ifndef SYS_gettid
236 // i386: 224, ia64: 1105, amd64: 186, sparc 143
237 #ifdef __ia64__
238 #define SYS_gettid 1105
239 #elif __i386__
240 #define SYS_gettid 224
241 #elif __amd64__
242 #define SYS_gettid 186
243 #elif __sparc__
244 #define SYS_gettid 143
245 #else
246 #error define gettid for the arch
247 #endif
248 #endif
250 // Cpu architecture string
251 #if defined(ZERO)
252 static char cpu_arch[] = ZERO_LIBARCH;
253 #elif defined(IA64)
254 static char cpu_arch[] = "ia64";
255 #elif defined(IA32)
256 static char cpu_arch[] = "i386";
257 #elif defined(AMD64)
258 static char cpu_arch[] = "amd64";
259 #elif defined(ARM)
260 static char cpu_arch[] = "arm";
261 #elif defined(PPC)
262 static char cpu_arch[] = "ppc";
263 #elif defined(SPARC)
264 # ifdef _LP64
265 static char cpu_arch[] = "sparcv9";
266 # else
267 static char cpu_arch[] = "sparc";
268 # endif
269 #else
270 #error Add appropriate cpu_arch setting
271 #endif
274 // pid_t gettid()
275 //
276 // Returns the kernel thread id of the currently running thread. Kernel
277 // thread id is used to access /proc.
278 //
279 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
280 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
281 //
282 pid_t os::Linux::gettid() {
283 int rslt = syscall(SYS_gettid);
284 if (rslt == -1) {
285 // old kernel, no NPTL support
286 return getpid();
287 } else {
288 return (pid_t)rslt;
289 }
290 }
292 // Most versions of linux have a bug where the number of processors are
293 // determined by looking at the /proc file system. In a chroot environment,
294 // the system call returns 1. This causes the VM to act as if it is
295 // a single processor and elide locking (see is_MP() call).
296 static bool unsafe_chroot_detected = false;
297 static const char *unstable_chroot_error = "/proc file system not found.\n"
298 "Java may be unstable running multithreaded in a chroot "
299 "environment on Linux when /proc filesystem is not mounted.";
301 void os::Linux::initialize_system_info() {
302 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
303 if (processor_count() == 1) {
304 pid_t pid = os::Linux::gettid();
305 char fname[32];
306 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
307 FILE *fp = fopen(fname, "r");
308 if (fp == NULL) {
309 unsafe_chroot_detected = true;
310 } else {
311 fclose(fp);
312 }
313 }
314 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
315 assert(processor_count() > 0, "linux error");
316 }
318 void os::init_system_properties_values() {
319 // char arch[12];
320 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
322 // The next steps are taken in the product version:
323 //
324 // Obtain the JAVA_HOME value from the location of libjvm[_g].so.
325 // This library should be located at:
326 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm[_g].so.
327 //
328 // If "/jre/lib/" appears at the right place in the path, then we
329 // assume libjvm[_g].so is installed in a JDK and we use this path.
330 //
331 // Otherwise exit with message: "Could not create the Java virtual machine."
332 //
333 // The following extra steps are taken in the debugging version:
334 //
335 // If "/jre/lib/" does NOT appear at the right place in the path
336 // instead of exit check for $JAVA_HOME environment variable.
337 //
338 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
339 // then we append a fake suffix "hotspot/libjvm[_g].so" to this path so
340 // it looks like libjvm[_g].so is installed there
341 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm[_g].so.
342 //
343 // Otherwise exit.
344 //
345 // Important note: if the location of libjvm.so changes this
346 // code needs to be changed accordingly.
348 // The next few definitions allow the code to be verbatim:
349 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
350 #define getenv(n) ::getenv(n)
352 /*
353 * See ld(1):
354 * The linker uses the following search paths to locate required
355 * shared libraries:
356 * 1: ...
357 * ...
358 * 7: The default directories, normally /lib and /usr/lib.
359 */
360 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
361 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
362 #else
363 #define DEFAULT_LIBPATH "/lib:/usr/lib"
364 #endif
366 #define EXTENSIONS_DIR "/lib/ext"
367 #define ENDORSED_DIR "/lib/endorsed"
368 #define REG_DIR "/usr/java/packages"
370 {
371 /* sysclasspath, java_home, dll_dir */
372 {
373 char *home_path;
374 char *dll_path;
375 char *pslash;
376 char buf[MAXPATHLEN];
377 os::jvm_path(buf, sizeof(buf));
379 // Found the full path to libjvm.so.
380 // Now cut the path to <java_home>/jre if we can.
381 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
382 pslash = strrchr(buf, '/');
383 if (pslash != NULL)
384 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
385 dll_path = malloc(strlen(buf) + 1);
386 if (dll_path == NULL)
387 return;
388 strcpy(dll_path, buf);
389 Arguments::set_dll_dir(dll_path);
391 if (pslash != NULL) {
392 pslash = strrchr(buf, '/');
393 if (pslash != NULL) {
394 *pslash = '\0'; /* get rid of /<arch> */
395 pslash = strrchr(buf, '/');
396 if (pslash != NULL)
397 *pslash = '\0'; /* get rid of /lib */
398 }
399 }
401 home_path = malloc(strlen(buf) + 1);
402 if (home_path == NULL)
403 return;
404 strcpy(home_path, buf);
405 Arguments::set_java_home(home_path);
407 if (!set_boot_path('/', ':'))
408 return;
409 }
411 /*
412 * Where to look for native libraries
413 *
414 * Note: Due to a legacy implementation, most of the library path
415 * is set in the launcher. This was to accomodate linking restrictions
416 * on legacy Linux implementations (which are no longer supported).
417 * Eventually, all the library path setting will be done here.
418 *
419 * However, to prevent the proliferation of improperly built native
420 * libraries, the new path component /usr/java/packages is added here.
421 * Eventually, all the library path setting will be done here.
422 */
423 {
424 char *ld_library_path;
426 /*
427 * Construct the invariant part of ld_library_path. Note that the
428 * space for the colon and the trailing null are provided by the
429 * nulls included by the sizeof operator (so actually we allocate
430 * a byte more than necessary).
431 */
432 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
433 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
434 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
436 /*
437 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
438 * should always exist (until the legacy problem cited above is
439 * addressed).
440 */
441 char *v = getenv("LD_LIBRARY_PATH");
442 if (v != NULL) {
443 char *t = ld_library_path;
444 /* That's +1 for the colon and +1 for the trailing '\0' */
445 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
446 sprintf(ld_library_path, "%s:%s", v, t);
447 }
448 Arguments::set_library_path(ld_library_path);
449 }
451 /*
452 * Extensions directories.
453 *
454 * Note that the space for the colon and the trailing null are provided
455 * by the nulls included by the sizeof operator (so actually one byte more
456 * than necessary is allocated).
457 */
458 {
459 char *buf = malloc(strlen(Arguments::get_java_home()) +
460 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
461 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
462 Arguments::get_java_home());
463 Arguments::set_ext_dirs(buf);
464 }
466 /* Endorsed standards default directory. */
467 {
468 char * buf;
469 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
470 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
471 Arguments::set_endorsed_dirs(buf);
472 }
473 }
475 #undef malloc
476 #undef getenv
477 #undef EXTENSIONS_DIR
478 #undef ENDORSED_DIR
480 // Done
481 return;
482 }
484 ////////////////////////////////////////////////////////////////////////////////
485 // breakpoint support
487 void os::breakpoint() {
488 BREAKPOINT;
489 }
491 extern "C" void breakpoint() {
492 // use debugger to set breakpoint here
493 }
495 ////////////////////////////////////////////////////////////////////////////////
496 // signal support
498 debug_only(static bool signal_sets_initialized = false);
499 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
501 bool os::Linux::is_sig_ignored(int sig) {
502 struct sigaction oact;
503 sigaction(sig, (struct sigaction*)NULL, &oact);
504 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
505 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
506 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
507 return true;
508 else
509 return false;
510 }
512 void os::Linux::signal_sets_init() {
513 // Should also have an assertion stating we are still single-threaded.
514 assert(!signal_sets_initialized, "Already initialized");
515 // Fill in signals that are necessarily unblocked for all threads in
516 // the VM. Currently, we unblock the following signals:
517 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
518 // by -Xrs (=ReduceSignalUsage));
519 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
520 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
521 // the dispositions or masks wrt these signals.
522 // Programs embedding the VM that want to use the above signals for their
523 // own purposes must, at this time, use the "-Xrs" option to prevent
524 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
525 // (See bug 4345157, and other related bugs).
526 // In reality, though, unblocking these signals is really a nop, since
527 // these signals are not blocked by default.
528 sigemptyset(&unblocked_sigs);
529 sigemptyset(&allowdebug_blocked_sigs);
530 sigaddset(&unblocked_sigs, SIGILL);
531 sigaddset(&unblocked_sigs, SIGSEGV);
532 sigaddset(&unblocked_sigs, SIGBUS);
533 sigaddset(&unblocked_sigs, SIGFPE);
534 sigaddset(&unblocked_sigs, SR_signum);
536 if (!ReduceSignalUsage) {
537 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
538 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
539 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
540 }
541 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
542 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
543 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
544 }
545 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
546 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
547 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
548 }
549 }
550 // Fill in signals that are blocked by all but the VM thread.
551 sigemptyset(&vm_sigs);
552 if (!ReduceSignalUsage)
553 sigaddset(&vm_sigs, BREAK_SIGNAL);
554 debug_only(signal_sets_initialized = true);
556 }
558 // These are signals that are unblocked while a thread is running Java.
559 // (For some reason, they get blocked by default.)
560 sigset_t* os::Linux::unblocked_signals() {
561 assert(signal_sets_initialized, "Not initialized");
562 return &unblocked_sigs;
563 }
565 // These are the signals that are blocked while a (non-VM) thread is
566 // running Java. Only the VM thread handles these signals.
567 sigset_t* os::Linux::vm_signals() {
568 assert(signal_sets_initialized, "Not initialized");
569 return &vm_sigs;
570 }
572 // These are signals that are blocked during cond_wait to allow debugger in
573 sigset_t* os::Linux::allowdebug_blocked_signals() {
574 assert(signal_sets_initialized, "Not initialized");
575 return &allowdebug_blocked_sigs;
576 }
578 void os::Linux::hotspot_sigmask(Thread* thread) {
580 //Save caller's signal mask before setting VM signal mask
581 sigset_t caller_sigmask;
582 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
584 OSThread* osthread = thread->osthread();
585 osthread->set_caller_sigmask(caller_sigmask);
587 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
589 if (!ReduceSignalUsage) {
590 if (thread->is_VM_thread()) {
591 // Only the VM thread handles BREAK_SIGNAL ...
592 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
593 } else {
594 // ... all other threads block BREAK_SIGNAL
595 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
596 }
597 }
598 }
600 //////////////////////////////////////////////////////////////////////////////
601 // detecting pthread library
603 void os::Linux::libpthread_init() {
604 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
605 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
606 // generic name for earlier versions.
607 // Define macros here so we can build HotSpot on old systems.
608 # ifndef _CS_GNU_LIBC_VERSION
609 # define _CS_GNU_LIBC_VERSION 2
610 # endif
611 # ifndef _CS_GNU_LIBPTHREAD_VERSION
612 # define _CS_GNU_LIBPTHREAD_VERSION 3
613 # endif
615 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
616 if (n > 0) {
617 char *str = (char *)malloc(n, mtInternal);
618 confstr(_CS_GNU_LIBC_VERSION, str, n);
619 os::Linux::set_glibc_version(str);
620 } else {
621 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
622 static char _gnu_libc_version[32];
623 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
624 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
625 os::Linux::set_glibc_version(_gnu_libc_version);
626 }
628 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
629 if (n > 0) {
630 char *str = (char *)malloc(n, mtInternal);
631 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
632 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
633 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
634 // is the case. LinuxThreads has a hard limit on max number of threads.
635 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
636 // On the other hand, NPTL does not have such a limit, sysconf()
637 // will return -1 and errno is not changed. Check if it is really NPTL.
638 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
639 strstr(str, "NPTL") &&
640 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
641 free(str);
642 os::Linux::set_libpthread_version("linuxthreads");
643 } else {
644 os::Linux::set_libpthread_version(str);
645 }
646 } else {
647 // glibc before 2.3.2 only has LinuxThreads.
648 os::Linux::set_libpthread_version("linuxthreads");
649 }
651 if (strstr(libpthread_version(), "NPTL")) {
652 os::Linux::set_is_NPTL();
653 } else {
654 os::Linux::set_is_LinuxThreads();
655 }
657 // LinuxThreads have two flavors: floating-stack mode, which allows variable
658 // stack size; and fixed-stack mode. NPTL is always floating-stack.
659 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
660 os::Linux::set_is_floating_stack();
661 }
662 }
664 /////////////////////////////////////////////////////////////////////////////
665 // thread stack
667 // Force Linux kernel to expand current thread stack. If "bottom" is close
668 // to the stack guard, caller should block all signals.
669 //
670 // MAP_GROWSDOWN:
671 // A special mmap() flag that is used to implement thread stacks. It tells
672 // kernel that the memory region should extend downwards when needed. This
673 // allows early versions of LinuxThreads to only mmap the first few pages
674 // when creating a new thread. Linux kernel will automatically expand thread
675 // stack as needed (on page faults).
676 //
677 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
678 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
679 // region, it's hard to tell if the fault is due to a legitimate stack
680 // access or because of reading/writing non-exist memory (e.g. buffer
681 // overrun). As a rule, if the fault happens below current stack pointer,
682 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
683 // application (see Linux kernel fault.c).
684 //
685 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
686 // stack overflow detection.
687 //
688 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
689 // not use this flag. However, the stack of initial thread is not created
690 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
691 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
692 // and then attach the thread to JVM.
693 //
694 // To get around the problem and allow stack banging on Linux, we need to
695 // manually expand thread stack after receiving the SIGSEGV.
696 //
697 // There are two ways to expand thread stack to address "bottom", we used
698 // both of them in JVM before 1.5:
699 // 1. adjust stack pointer first so that it is below "bottom", and then
700 // touch "bottom"
701 // 2. mmap() the page in question
702 //
703 // Now alternate signal stack is gone, it's harder to use 2. For instance,
704 // if current sp is already near the lower end of page 101, and we need to
705 // call mmap() to map page 100, it is possible that part of the mmap() frame
706 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
707 // That will destroy the mmap() frame and cause VM to crash.
708 //
709 // The following code works by adjusting sp first, then accessing the "bottom"
710 // page to force a page fault. Linux kernel will then automatically expand the
711 // stack mapping.
712 //
713 // _expand_stack_to() assumes its frame size is less than page size, which
714 // should always be true if the function is not inlined.
716 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
717 #define NOINLINE
718 #else
719 #define NOINLINE __attribute__ ((noinline))
720 #endif
722 static void _expand_stack_to(address bottom) NOINLINE;
724 static void _expand_stack_to(address bottom) {
725 address sp;
726 size_t size;
727 volatile char *p;
729 // Adjust bottom to point to the largest address within the same page, it
730 // gives us a one-page buffer if alloca() allocates slightly more memory.
731 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
732 bottom += os::Linux::page_size() - 1;
734 // sp might be slightly above current stack pointer; if that's the case, we
735 // will alloca() a little more space than necessary, which is OK. Don't use
736 // os::current_stack_pointer(), as its result can be slightly below current
737 // stack pointer, causing us to not alloca enough to reach "bottom".
738 sp = (address)&sp;
740 if (sp > bottom) {
741 size = sp - bottom;
742 p = (volatile char *)alloca(size);
743 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
744 p[0] = '\0';
745 }
746 }
748 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
749 assert(t!=NULL, "just checking");
750 assert(t->osthread()->expanding_stack(), "expand should be set");
751 assert(t->stack_base() != NULL, "stack_base was not initialized");
753 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
754 sigset_t mask_all, old_sigset;
755 sigfillset(&mask_all);
756 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
757 _expand_stack_to(addr);
758 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
759 return true;
760 }
761 return false;
762 }
764 //////////////////////////////////////////////////////////////////////////////
765 // create new thread
767 static address highest_vm_reserved_address();
769 // check if it's safe to start a new thread
770 static bool _thread_safety_check(Thread* thread) {
771 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
772 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
773 // Heap is mmap'ed at lower end of memory space. Thread stacks are
774 // allocated (MAP_FIXED) from high address space. Every thread stack
775 // occupies a fixed size slot (usually 2Mbytes, but user can change
776 // it to other values if they rebuild LinuxThreads).
777 //
778 // Problem with MAP_FIXED is that mmap() can still succeed even part of
779 // the memory region has already been mmap'ed. That means if we have too
780 // many threads and/or very large heap, eventually thread stack will
781 // collide with heap.
782 //
783 // Here we try to prevent heap/stack collision by comparing current
784 // stack bottom with the highest address that has been mmap'ed by JVM
785 // plus a safety margin for memory maps created by native code.
786 //
787 // This feature can be disabled by setting ThreadSafetyMargin to 0
788 //
789 if (ThreadSafetyMargin > 0) {
790 address stack_bottom = os::current_stack_base() - os::current_stack_size();
792 // not safe if our stack extends below the safety margin
793 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
794 } else {
795 return true;
796 }
797 } else {
798 // Floating stack LinuxThreads or NPTL:
799 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
800 // there's not enough space left, pthread_create() will fail. If we come
801 // here, that means enough space has been reserved for stack.
802 return true;
803 }
804 }
806 // Thread start routine for all newly created threads
807 static void *java_start(Thread *thread) {
808 // Try to randomize the cache line index of hot stack frames.
809 // This helps when threads of the same stack traces evict each other's
810 // cache lines. The threads can be either from the same JVM instance, or
811 // from different JVM instances. The benefit is especially true for
812 // processors with hyperthreading technology.
813 static int counter = 0;
814 int pid = os::current_process_id();
815 alloca(((pid ^ counter++) & 7) * 128);
817 ThreadLocalStorage::set_thread(thread);
819 OSThread* osthread = thread->osthread();
820 Monitor* sync = osthread->startThread_lock();
822 // non floating stack LinuxThreads needs extra check, see above
823 if (!_thread_safety_check(thread)) {
824 // notify parent thread
825 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
826 osthread->set_state(ZOMBIE);
827 sync->notify_all();
828 return NULL;
829 }
831 // thread_id is kernel thread id (similar to Solaris LWP id)
832 osthread->set_thread_id(os::Linux::gettid());
834 if (UseNUMA) {
835 int lgrp_id = os::numa_get_group_id();
836 if (lgrp_id != -1) {
837 thread->set_lgrp_id(lgrp_id);
838 }
839 }
840 // initialize signal mask for this thread
841 os::Linux::hotspot_sigmask(thread);
843 // initialize floating point control register
844 os::Linux::init_thread_fpu_state();
846 // handshaking with parent thread
847 {
848 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
850 // notify parent thread
851 osthread->set_state(INITIALIZED);
852 sync->notify_all();
854 // wait until os::start_thread()
855 while (osthread->get_state() == INITIALIZED) {
856 sync->wait(Mutex::_no_safepoint_check_flag);
857 }
858 }
860 // call one more level start routine
861 thread->run();
863 return 0;
864 }
866 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
867 assert(thread->osthread() == NULL, "caller responsible");
869 // Allocate the OSThread object
870 OSThread* osthread = new OSThread(NULL, NULL);
871 if (osthread == NULL) {
872 return false;
873 }
875 // set the correct thread state
876 osthread->set_thread_type(thr_type);
878 // Initial state is ALLOCATED but not INITIALIZED
879 osthread->set_state(ALLOCATED);
881 thread->set_osthread(osthread);
883 // init thread attributes
884 pthread_attr_t attr;
885 pthread_attr_init(&attr);
886 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
888 // stack size
889 if (os::Linux::supports_variable_stack_size()) {
890 // calculate stack size if it's not specified by caller
891 if (stack_size == 0) {
892 stack_size = os::Linux::default_stack_size(thr_type);
894 switch (thr_type) {
895 case os::java_thread:
896 // Java threads use ThreadStackSize which default value can be
897 // changed with the flag -Xss
898 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
899 stack_size = JavaThread::stack_size_at_create();
900 break;
901 case os::compiler_thread:
902 if (CompilerThreadStackSize > 0) {
903 stack_size = (size_t)(CompilerThreadStackSize * K);
904 break;
905 } // else fall through:
906 // use VMThreadStackSize if CompilerThreadStackSize is not defined
907 case os::vm_thread:
908 case os::pgc_thread:
909 case os::cgc_thread:
910 case os::watcher_thread:
911 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
912 break;
913 }
914 }
916 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
917 pthread_attr_setstacksize(&attr, stack_size);
918 } else {
919 // let pthread_create() pick the default value.
920 }
922 // glibc guard page
923 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
925 ThreadState state;
927 {
928 // Serialize thread creation if we are running with fixed stack LinuxThreads
929 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
930 if (lock) {
931 os::Linux::createThread_lock()->lock_without_safepoint_check();
932 }
934 pthread_t tid;
935 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
937 pthread_attr_destroy(&attr);
939 if (ret != 0) {
940 if (PrintMiscellaneous && (Verbose || WizardMode)) {
941 perror("pthread_create()");
942 }
943 // Need to clean up stuff we've allocated so far
944 thread->set_osthread(NULL);
945 delete osthread;
946 if (lock) os::Linux::createThread_lock()->unlock();
947 return false;
948 }
950 // Store pthread info into the OSThread
951 osthread->set_pthread_id(tid);
953 // Wait until child thread is either initialized or aborted
954 {
955 Monitor* sync_with_child = osthread->startThread_lock();
956 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
957 while ((state = osthread->get_state()) == ALLOCATED) {
958 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
959 }
960 }
962 if (lock) {
963 os::Linux::createThread_lock()->unlock();
964 }
965 }
967 // Aborted due to thread limit being reached
968 if (state == ZOMBIE) {
969 thread->set_osthread(NULL);
970 delete osthread;
971 return false;
972 }
974 // The thread is returned suspended (in state INITIALIZED),
975 // and is started higher up in the call chain
976 assert(state == INITIALIZED, "race condition");
977 return true;
978 }
980 /////////////////////////////////////////////////////////////////////////////
981 // attach existing thread
983 // bootstrap the main thread
984 bool os::create_main_thread(JavaThread* thread) {
985 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
986 return create_attached_thread(thread);
987 }
989 bool os::create_attached_thread(JavaThread* thread) {
990 #ifdef ASSERT
991 thread->verify_not_published();
992 #endif
994 // Allocate the OSThread object
995 OSThread* osthread = new OSThread(NULL, NULL);
997 if (osthread == NULL) {
998 return false;
999 }
1001 // Store pthread info into the OSThread
1002 osthread->set_thread_id(os::Linux::gettid());
1003 osthread->set_pthread_id(::pthread_self());
1005 // initialize floating point control register
1006 os::Linux::init_thread_fpu_state();
1008 // Initial thread state is RUNNABLE
1009 osthread->set_state(RUNNABLE);
1011 thread->set_osthread(osthread);
1013 if (UseNUMA) {
1014 int lgrp_id = os::numa_get_group_id();
1015 if (lgrp_id != -1) {
1016 thread->set_lgrp_id(lgrp_id);
1017 }
1018 }
1020 if (os::Linux::is_initial_thread()) {
1021 // If current thread is initial thread, its stack is mapped on demand,
1022 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1023 // the entire stack region to avoid SEGV in stack banging.
1024 // It is also useful to get around the heap-stack-gap problem on SuSE
1025 // kernel (see 4821821 for details). We first expand stack to the top
1026 // of yellow zone, then enable stack yellow zone (order is significant,
1027 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1028 // is no gap between the last two virtual memory regions.
1030 JavaThread *jt = (JavaThread *)thread;
1031 address addr = jt->stack_yellow_zone_base();
1032 assert(addr != NULL, "initialization problem?");
1033 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1035 osthread->set_expanding_stack();
1036 os::Linux::manually_expand_stack(jt, addr);
1037 osthread->clear_expanding_stack();
1038 }
1040 // initialize signal mask for this thread
1041 // and save the caller's signal mask
1042 os::Linux::hotspot_sigmask(thread);
1044 return true;
1045 }
1047 void os::pd_start_thread(Thread* thread) {
1048 OSThread * osthread = thread->osthread();
1049 assert(osthread->get_state() != INITIALIZED, "just checking");
1050 Monitor* sync_with_child = osthread->startThread_lock();
1051 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1052 sync_with_child->notify();
1053 }
1055 // Free Linux resources related to the OSThread
1056 void os::free_thread(OSThread* osthread) {
1057 assert(osthread != NULL, "osthread not set");
1059 if (Thread::current()->osthread() == osthread) {
1060 // Restore caller's signal mask
1061 sigset_t sigmask = osthread->caller_sigmask();
1062 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1063 }
1065 delete osthread;
1066 }
1068 //////////////////////////////////////////////////////////////////////////////
1069 // thread local storage
1071 int os::allocate_thread_local_storage() {
1072 pthread_key_t key;
1073 int rslt = pthread_key_create(&key, NULL);
1074 assert(rslt == 0, "cannot allocate thread local storage");
1075 return (int)key;
1076 }
1078 // Note: This is currently not used by VM, as we don't destroy TLS key
1079 // on VM exit.
1080 void os::free_thread_local_storage(int index) {
1081 int rslt = pthread_key_delete((pthread_key_t)index);
1082 assert(rslt == 0, "invalid index");
1083 }
1085 void os::thread_local_storage_at_put(int index, void* value) {
1086 int rslt = pthread_setspecific((pthread_key_t)index, value);
1087 assert(rslt == 0, "pthread_setspecific failed");
1088 }
1090 extern "C" Thread* get_thread() {
1091 return ThreadLocalStorage::thread();
1092 }
1094 //////////////////////////////////////////////////////////////////////////////
1095 // initial thread
1097 // Check if current thread is the initial thread, similar to Solaris thr_main.
1098 bool os::Linux::is_initial_thread(void) {
1099 char dummy;
1100 // If called before init complete, thread stack bottom will be null.
1101 // Can be called if fatal error occurs before initialization.
1102 if (initial_thread_stack_bottom() == NULL) return false;
1103 assert(initial_thread_stack_bottom() != NULL &&
1104 initial_thread_stack_size() != 0,
1105 "os::init did not locate initial thread's stack region");
1106 if ((address)&dummy >= initial_thread_stack_bottom() &&
1107 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1108 return true;
1109 else return false;
1110 }
1112 // Find the virtual memory area that contains addr
1113 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1114 FILE *fp = fopen("/proc/self/maps", "r");
1115 if (fp) {
1116 address low, high;
1117 while (!feof(fp)) {
1118 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1119 if (low <= addr && addr < high) {
1120 if (vma_low) *vma_low = low;
1121 if (vma_high) *vma_high = high;
1122 fclose (fp);
1123 return true;
1124 }
1125 }
1126 for (;;) {
1127 int ch = fgetc(fp);
1128 if (ch == EOF || ch == (int)'\n') break;
1129 }
1130 }
1131 fclose(fp);
1132 }
1133 return false;
1134 }
1136 // Locate initial thread stack. This special handling of initial thread stack
1137 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1138 // bogus value for initial thread.
1139 void os::Linux::capture_initial_stack(size_t max_size) {
1140 // stack size is the easy part, get it from RLIMIT_STACK
1141 size_t stack_size;
1142 struct rlimit rlim;
1143 getrlimit(RLIMIT_STACK, &rlim);
1144 stack_size = rlim.rlim_cur;
1146 // 6308388: a bug in ld.so will relocate its own .data section to the
1147 // lower end of primordial stack; reduce ulimit -s value a little bit
1148 // so we won't install guard page on ld.so's data section.
1149 stack_size -= 2 * page_size();
1151 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1152 // 7.1, in both cases we will get 2G in return value.
1153 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1154 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1155 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1156 // in case other parts in glibc still assumes 2M max stack size.
1157 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1158 #ifndef IA64
1159 if (stack_size > 2 * K * K) stack_size = 2 * K * K;
1160 #else
1161 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1162 if (stack_size > 4 * K * K) stack_size = 4 * K * K;
1163 #endif
1165 // Try to figure out where the stack base (top) is. This is harder.
1166 //
1167 // When an application is started, glibc saves the initial stack pointer in
1168 // a global variable "__libc_stack_end", which is then used by system
1169 // libraries. __libc_stack_end should be pretty close to stack top. The
1170 // variable is available since the very early days. However, because it is
1171 // a private interface, it could disappear in the future.
1172 //
1173 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1174 // to __libc_stack_end, it is very close to stack top, but isn't the real
1175 // stack top. Note that /proc may not exist if VM is running as a chroot
1176 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1177 // /proc/<pid>/stat could change in the future (though unlikely).
1178 //
1179 // We try __libc_stack_end first. If that doesn't work, look for
1180 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1181 // as a hint, which should work well in most cases.
1183 uintptr_t stack_start;
1185 // try __libc_stack_end first
1186 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1187 if (p && *p) {
1188 stack_start = *p;
1189 } else {
1190 // see if we can get the start_stack field from /proc/self/stat
1191 FILE *fp;
1192 int pid;
1193 char state;
1194 int ppid;
1195 int pgrp;
1196 int session;
1197 int nr;
1198 int tpgrp;
1199 unsigned long flags;
1200 unsigned long minflt;
1201 unsigned long cminflt;
1202 unsigned long majflt;
1203 unsigned long cmajflt;
1204 unsigned long utime;
1205 unsigned long stime;
1206 long cutime;
1207 long cstime;
1208 long prio;
1209 long nice;
1210 long junk;
1211 long it_real;
1212 uintptr_t start;
1213 uintptr_t vsize;
1214 intptr_t rss;
1215 uintptr_t rsslim;
1216 uintptr_t scodes;
1217 uintptr_t ecode;
1218 int i;
1220 // Figure what the primordial thread stack base is. Code is inspired
1221 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1222 // followed by command name surrounded by parentheses, state, etc.
1223 char stat[2048];
1224 int statlen;
1226 fp = fopen("/proc/self/stat", "r");
1227 if (fp) {
1228 statlen = fread(stat, 1, 2047, fp);
1229 stat[statlen] = '\0';
1230 fclose(fp);
1232 // Skip pid and the command string. Note that we could be dealing with
1233 // weird command names, e.g. user could decide to rename java launcher
1234 // to "java 1.4.2 :)", then the stat file would look like
1235 // 1234 (java 1.4.2 :)) R ... ...
1236 // We don't really need to know the command string, just find the last
1237 // occurrence of ")" and then start parsing from there. See bug 4726580.
1238 char * s = strrchr(stat, ')');
1240 i = 0;
1241 if (s) {
1242 // Skip blank chars
1243 do s++; while (isspace(*s));
1245 #define _UFM UINTX_FORMAT
1246 #define _DFM INTX_FORMAT
1248 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1249 /* 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 */
1250 i = sscanf(s, "%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu %ld %ld %ld %ld %ld %ld " _UFM _UFM _DFM _UFM _UFM _UFM _UFM,
1251 &state, /* 3 %c */
1252 &ppid, /* 4 %d */
1253 &pgrp, /* 5 %d */
1254 &session, /* 6 %d */
1255 &nr, /* 7 %d */
1256 &tpgrp, /* 8 %d */
1257 &flags, /* 9 %lu */
1258 &minflt, /* 10 %lu */
1259 &cminflt, /* 11 %lu */
1260 &majflt, /* 12 %lu */
1261 &cmajflt, /* 13 %lu */
1262 &utime, /* 14 %lu */
1263 &stime, /* 15 %lu */
1264 &cutime, /* 16 %ld */
1265 &cstime, /* 17 %ld */
1266 &prio, /* 18 %ld */
1267 &nice, /* 19 %ld */
1268 &junk, /* 20 %ld */
1269 &it_real, /* 21 %ld */
1270 &start, /* 22 UINTX_FORMAT */
1271 &vsize, /* 23 UINTX_FORMAT */
1272 &rss, /* 24 INTX_FORMAT */
1273 &rsslim, /* 25 UINTX_FORMAT */
1274 &scodes, /* 26 UINTX_FORMAT */
1275 &ecode, /* 27 UINTX_FORMAT */
1276 &stack_start); /* 28 UINTX_FORMAT */
1277 }
1279 #undef _UFM
1280 #undef _DFM
1282 if (i != 28 - 2) {
1283 assert(false, "Bad conversion from /proc/self/stat");
1284 // product mode - assume we are the initial thread, good luck in the
1285 // embedded case.
1286 warning("Can't detect initial thread stack location - bad conversion");
1287 stack_start = (uintptr_t) &rlim;
1288 }
1289 } else {
1290 // For some reason we can't open /proc/self/stat (for example, running on
1291 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1292 // most cases, so don't abort:
1293 warning("Can't detect initial thread stack location - no /proc/self/stat");
1294 stack_start = (uintptr_t) &rlim;
1295 }
1296 }
1298 // Now we have a pointer (stack_start) very close to the stack top, the
1299 // next thing to do is to figure out the exact location of stack top. We
1300 // can find out the virtual memory area that contains stack_start by
1301 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1302 // and its upper limit is the real stack top. (again, this would fail if
1303 // running inside chroot, because /proc may not exist.)
1305 uintptr_t stack_top;
1306 address low, high;
1307 if (find_vma((address)stack_start, &low, &high)) {
1308 // success, "high" is the true stack top. (ignore "low", because initial
1309 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1310 stack_top = (uintptr_t)high;
1311 } else {
1312 // failed, likely because /proc/self/maps does not exist
1313 warning("Can't detect initial thread stack location - find_vma failed");
1314 // best effort: stack_start is normally within a few pages below the real
1315 // stack top, use it as stack top, and reduce stack size so we won't put
1316 // guard page outside stack.
1317 stack_top = stack_start;
1318 stack_size -= 16 * page_size();
1319 }
1321 // stack_top could be partially down the page so align it
1322 stack_top = align_size_up(stack_top, page_size());
1324 if (max_size && stack_size > max_size) {
1325 _initial_thread_stack_size = max_size;
1326 } else {
1327 _initial_thread_stack_size = stack_size;
1328 }
1330 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1331 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1332 }
1334 ////////////////////////////////////////////////////////////////////////////////
1335 // time support
1337 // Time since start-up in seconds to a fine granularity.
1338 // Used by VMSelfDestructTimer and the MemProfiler.
1339 double os::elapsedTime() {
1341 return (double)(os::elapsed_counter()) * 0.000001;
1342 }
1344 jlong os::elapsed_counter() {
1345 timeval time;
1346 int status = gettimeofday(&time, NULL);
1347 return jlong(time.tv_sec) * 1000 * 1000 + jlong(time.tv_usec) - initial_time_count;
1348 }
1350 jlong os::elapsed_frequency() {
1351 return (1000 * 1000);
1352 }
1354 // For now, we say that linux does not support vtime. I have no idea
1355 // whether it can actually be made to (DLD, 9/13/05).
1357 bool os::supports_vtime() { return false; }
1358 bool os::enable_vtime() { return false; }
1359 bool os::vtime_enabled() { return false; }
1360 double os::elapsedVTime() {
1361 // better than nothing, but not much
1362 return elapsedTime();
1363 }
1365 jlong os::javaTimeMillis() {
1366 timeval time;
1367 int status = gettimeofday(&time, NULL);
1368 assert(status != -1, "linux error");
1369 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1370 }
1372 #ifndef CLOCK_MONOTONIC
1373 #define CLOCK_MONOTONIC (1)
1374 #endif
1376 void os::Linux::clock_init() {
1377 // we do dlopen's in this particular order due to bug in linux
1378 // dynamical loader (see 6348968) leading to crash on exit
1379 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1380 if (handle == NULL) {
1381 handle = dlopen("librt.so", RTLD_LAZY);
1382 }
1384 if (handle) {
1385 int (*clock_getres_func)(clockid_t, struct timespec*) =
1386 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1387 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1388 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1389 if (clock_getres_func && clock_gettime_func) {
1390 // See if monotonic clock is supported by the kernel. Note that some
1391 // early implementations simply return kernel jiffies (updated every
1392 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1393 // for nano time (though the monotonic property is still nice to have).
1394 // It's fixed in newer kernels, however clock_getres() still returns
1395 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1396 // resolution for now. Hopefully as people move to new kernels, this
1397 // won't be a problem.
1398 struct timespec res;
1399 struct timespec tp;
1400 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1401 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1402 // yes, monotonic clock is supported
1403 _clock_gettime = clock_gettime_func;
1404 } else {
1405 // close librt if there is no monotonic clock
1406 dlclose(handle);
1407 }
1408 }
1409 }
1410 }
1412 #ifndef SYS_clock_getres
1414 #if defined(IA32) || defined(AMD64)
1415 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1416 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1417 #else
1418 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1419 #define sys_clock_getres(x,y) -1
1420 #endif
1422 #else
1423 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1424 #endif
1426 void os::Linux::fast_thread_clock_init() {
1427 if (!UseLinuxPosixThreadCPUClocks) {
1428 return;
1429 }
1430 clockid_t clockid;
1431 struct timespec tp;
1432 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1433 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1435 // Switch to using fast clocks for thread cpu time if
1436 // the sys_clock_getres() returns 0 error code.
1437 // Note, that some kernels may support the current thread
1438 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1439 // returned by the pthread_getcpuclockid().
1440 // If the fast Posix clocks are supported then the sys_clock_getres()
1441 // must return at least tp.tv_sec == 0 which means a resolution
1442 // better than 1 sec. This is extra check for reliability.
1444 if(pthread_getcpuclockid_func &&
1445 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1446 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1448 _supports_fast_thread_cpu_time = true;
1449 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1450 }
1451 }
1453 jlong os::javaTimeNanos() {
1454 if (Linux::supports_monotonic_clock()) {
1455 struct timespec tp;
1456 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1457 assert(status == 0, "gettime error");
1458 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1459 return result;
1460 } else {
1461 timeval time;
1462 int status = gettimeofday(&time, NULL);
1463 assert(status != -1, "linux error");
1464 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1465 return 1000 * usecs;
1466 }
1467 }
1469 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1470 if (Linux::supports_monotonic_clock()) {
1471 info_ptr->max_value = ALL_64_BITS;
1473 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1474 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1475 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1476 } else {
1477 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1478 info_ptr->max_value = ALL_64_BITS;
1480 // gettimeofday is a real time clock so it skips
1481 info_ptr->may_skip_backward = true;
1482 info_ptr->may_skip_forward = true;
1483 }
1485 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1486 }
1488 // Return the real, user, and system times in seconds from an
1489 // arbitrary fixed point in the past.
1490 bool os::getTimesSecs(double* process_real_time,
1491 double* process_user_time,
1492 double* process_system_time) {
1493 struct tms ticks;
1494 clock_t real_ticks = times(&ticks);
1496 if (real_ticks == (clock_t) (-1)) {
1497 return false;
1498 } else {
1499 double ticks_per_second = (double) clock_tics_per_sec;
1500 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1501 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1502 *process_real_time = ((double) real_ticks) / ticks_per_second;
1504 return true;
1505 }
1506 }
1509 char * os::local_time_string(char *buf, size_t buflen) {
1510 struct tm t;
1511 time_t long_time;
1512 time(&long_time);
1513 localtime_r(&long_time, &t);
1514 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1515 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1516 t.tm_hour, t.tm_min, t.tm_sec);
1517 return buf;
1518 }
1520 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1521 return localtime_r(clock, res);
1522 }
1524 ////////////////////////////////////////////////////////////////////////////////
1525 // runtime exit support
1527 // Note: os::shutdown() might be called very early during initialization, or
1528 // called from signal handler. Before adding something to os::shutdown(), make
1529 // sure it is async-safe and can handle partially initialized VM.
1530 void os::shutdown() {
1532 // allow PerfMemory to attempt cleanup of any persistent resources
1533 perfMemory_exit();
1535 // needs to remove object in file system
1536 AttachListener::abort();
1538 // flush buffered output, finish log files
1539 ostream_abort();
1541 // Check for abort hook
1542 abort_hook_t abort_hook = Arguments::abort_hook();
1543 if (abort_hook != NULL) {
1544 abort_hook();
1545 }
1547 }
1549 // Note: os::abort() might be called very early during initialization, or
1550 // called from signal handler. Before adding something to os::abort(), make
1551 // sure it is async-safe and can handle partially initialized VM.
1552 void os::abort(bool dump_core) {
1553 os::shutdown();
1554 if (dump_core) {
1555 #ifndef PRODUCT
1556 fdStream out(defaultStream::output_fd());
1557 out.print_raw("Current thread is ");
1558 char buf[16];
1559 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1560 out.print_raw_cr(buf);
1561 out.print_raw_cr("Dumping core ...");
1562 #endif
1563 ::abort(); // dump core
1564 }
1566 ::exit(1);
1567 }
1569 // Die immediately, no exit hook, no abort hook, no cleanup.
1570 void os::die() {
1571 // _exit() on LinuxThreads only kills current thread
1572 ::abort();
1573 }
1575 // unused on linux for now.
1576 void os::set_error_file(const char *logfile) {}
1579 // This method is a copy of JDK's sysGetLastErrorString
1580 // from src/solaris/hpi/src/system_md.c
1582 size_t os::lasterror(char *buf, size_t len) {
1584 if (errno == 0) return 0;
1586 const char *s = ::strerror(errno);
1587 size_t n = ::strlen(s);
1588 if (n >= len) {
1589 n = len - 1;
1590 }
1591 ::strncpy(buf, s, n);
1592 buf[n] = '\0';
1593 return n;
1594 }
1596 intx os::current_thread_id() { return (intx)pthread_self(); }
1597 int os::current_process_id() {
1599 // Under the old linux thread library, linux gives each thread
1600 // its own process id. Because of this each thread will return
1601 // a different pid if this method were to return the result
1602 // of getpid(2). Linux provides no api that returns the pid
1603 // of the launcher thread for the vm. This implementation
1604 // returns a unique pid, the pid of the launcher thread
1605 // that starts the vm 'process'.
1607 // Under the NPTL, getpid() returns the same pid as the
1608 // launcher thread rather than a unique pid per thread.
1609 // Use gettid() if you want the old pre NPTL behaviour.
1611 // if you are looking for the result of a call to getpid() that
1612 // returns a unique pid for the calling thread, then look at the
1613 // OSThread::thread_id() method in osThread_linux.hpp file
1615 return (int)(_initial_pid ? _initial_pid : getpid());
1616 }
1618 // DLL functions
1620 const char* os::dll_file_extension() { return ".so"; }
1622 // This must be hard coded because it's the system's temporary
1623 // directory not the java application's temp directory, ala java.io.tmpdir.
1624 const char* os::get_temp_directory() { return "/tmp"; }
1626 static bool file_exists(const char* filename) {
1627 struct stat statbuf;
1628 if (filename == NULL || strlen(filename) == 0) {
1629 return false;
1630 }
1631 return os::stat(filename, &statbuf) == 0;
1632 }
1634 bool os::dll_build_name(char* buffer, size_t buflen,
1635 const char* pname, const char* fname) {
1636 bool retval = false;
1637 // Copied from libhpi
1638 const size_t pnamelen = pname ? strlen(pname) : 0;
1640 // Return error on buffer overflow.
1641 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1642 return retval;
1643 }
1645 if (pnamelen == 0) {
1646 snprintf(buffer, buflen, "lib%s.so", fname);
1647 retval = true;
1648 } else if (strchr(pname, *os::path_separator()) != NULL) {
1649 int n;
1650 char** pelements = split_path(pname, &n);
1651 for (int i = 0 ; i < n ; i++) {
1652 // Really shouldn't be NULL, but check can't hurt
1653 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1654 continue; // skip the empty path values
1655 }
1656 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1657 if (file_exists(buffer)) {
1658 retval = true;
1659 break;
1660 }
1661 }
1662 // release the storage
1663 for (int i = 0 ; i < n ; i++) {
1664 if (pelements[i] != NULL) {
1665 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1666 }
1667 }
1668 if (pelements != NULL) {
1669 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1670 }
1671 } else {
1672 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1673 retval = true;
1674 }
1675 return retval;
1676 }
1678 const char* os::get_current_directory(char *buf, int buflen) {
1679 return getcwd(buf, buflen);
1680 }
1682 // check if addr is inside libjvm[_g].so
1683 bool os::address_is_in_vm(address addr) {
1684 static address libjvm_base_addr;
1685 Dl_info dlinfo;
1687 if (libjvm_base_addr == NULL) {
1688 dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo);
1689 libjvm_base_addr = (address)dlinfo.dli_fbase;
1690 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1691 }
1693 if (dladdr((void *)addr, &dlinfo)) {
1694 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1695 }
1697 return false;
1698 }
1700 bool os::dll_address_to_function_name(address addr, char *buf,
1701 int buflen, int *offset) {
1702 Dl_info dlinfo;
1704 if (dladdr((void*)addr, &dlinfo) && dlinfo.dli_sname != NULL) {
1705 if (buf != NULL) {
1706 if(!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1707 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1708 }
1709 }
1710 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1711 return true;
1712 } else if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != 0) {
1713 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1714 buf, buflen, offset, dlinfo.dli_fname)) {
1715 return true;
1716 }
1717 }
1719 if (buf != NULL) buf[0] = '\0';
1720 if (offset != NULL) *offset = -1;
1721 return false;
1722 }
1724 struct _address_to_library_name {
1725 address addr; // input : memory address
1726 size_t buflen; // size of fname
1727 char* fname; // output: library name
1728 address base; // library base addr
1729 };
1731 static int address_to_library_name_callback(struct dl_phdr_info *info,
1732 size_t size, void *data) {
1733 int i;
1734 bool found = false;
1735 address libbase = NULL;
1736 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1738 // iterate through all loadable segments
1739 for (i = 0; i < info->dlpi_phnum; i++) {
1740 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1741 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1742 // base address of a library is the lowest address of its loaded
1743 // segments.
1744 if (libbase == NULL || libbase > segbase) {
1745 libbase = segbase;
1746 }
1747 // see if 'addr' is within current segment
1748 if (segbase <= d->addr &&
1749 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1750 found = true;
1751 }
1752 }
1753 }
1755 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1756 // so dll_address_to_library_name() can fall through to use dladdr() which
1757 // can figure out executable name from argv[0].
1758 if (found && info->dlpi_name && info->dlpi_name[0]) {
1759 d->base = libbase;
1760 if (d->fname) {
1761 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1762 }
1763 return 1;
1764 }
1765 return 0;
1766 }
1768 bool os::dll_address_to_library_name(address addr, char* buf,
1769 int buflen, int* offset) {
1770 Dl_info dlinfo;
1771 struct _address_to_library_name data;
1773 // There is a bug in old glibc dladdr() implementation that it could resolve
1774 // to wrong library name if the .so file has a base address != NULL. Here
1775 // we iterate through the program headers of all loaded libraries to find
1776 // out which library 'addr' really belongs to. This workaround can be
1777 // removed once the minimum requirement for glibc is moved to 2.3.x.
1778 data.addr = addr;
1779 data.fname = buf;
1780 data.buflen = buflen;
1781 data.base = NULL;
1782 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1784 if (rslt) {
1785 // buf already contains library name
1786 if (offset) *offset = addr - data.base;
1787 return true;
1788 } else if (dladdr((void*)addr, &dlinfo)){
1789 if (buf) jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1790 if (offset) *offset = addr - (address)dlinfo.dli_fbase;
1791 return true;
1792 } else {
1793 if (buf) buf[0] = '\0';
1794 if (offset) *offset = -1;
1795 return false;
1796 }
1797 }
1799 // Loads .dll/.so and
1800 // in case of error it checks if .dll/.so was built for the
1801 // same architecture as Hotspot is running on
1803 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1804 {
1805 void * result= ::dlopen(filename, RTLD_LAZY);
1806 if (result != NULL) {
1807 // Successful loading
1808 return result;
1809 }
1811 Elf32_Ehdr elf_head;
1813 // Read system error message into ebuf
1814 // It may or may not be overwritten below
1815 ::strncpy(ebuf, ::dlerror(), ebuflen-1);
1816 ebuf[ebuflen-1]='\0';
1817 int diag_msg_max_length=ebuflen-strlen(ebuf);
1818 char* diag_msg_buf=ebuf+strlen(ebuf);
1820 if (diag_msg_max_length==0) {
1821 // No more space in ebuf for additional diagnostics message
1822 return NULL;
1823 }
1826 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1828 if (file_descriptor < 0) {
1829 // Can't open library, report dlerror() message
1830 return NULL;
1831 }
1833 bool failed_to_read_elf_head=
1834 (sizeof(elf_head)!=
1835 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1837 ::close(file_descriptor);
1838 if (failed_to_read_elf_head) {
1839 // file i/o error - report dlerror() msg
1840 return NULL;
1841 }
1843 typedef struct {
1844 Elf32_Half code; // Actual value as defined in elf.h
1845 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1846 char elf_class; // 32 or 64 bit
1847 char endianess; // MSB or LSB
1848 char* name; // String representation
1849 } arch_t;
1851 #ifndef EM_486
1852 #define EM_486 6 /* Intel 80486 */
1853 #endif
1855 static const arch_t arch_array[]={
1856 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1857 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1858 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1859 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1860 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1861 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1862 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1863 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1864 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1865 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1866 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1867 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1868 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1869 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1870 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1871 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1872 };
1874 #if (defined IA32)
1875 static Elf32_Half running_arch_code=EM_386;
1876 #elif (defined AMD64)
1877 static Elf32_Half running_arch_code=EM_X86_64;
1878 #elif (defined IA64)
1879 static Elf32_Half running_arch_code=EM_IA_64;
1880 #elif (defined __sparc) && (defined _LP64)
1881 static Elf32_Half running_arch_code=EM_SPARCV9;
1882 #elif (defined __sparc) && (!defined _LP64)
1883 static Elf32_Half running_arch_code=EM_SPARC;
1884 #elif (defined __powerpc64__)
1885 static Elf32_Half running_arch_code=EM_PPC64;
1886 #elif (defined __powerpc__)
1887 static Elf32_Half running_arch_code=EM_PPC;
1888 #elif (defined ARM)
1889 static Elf32_Half running_arch_code=EM_ARM;
1890 #elif (defined S390)
1891 static Elf32_Half running_arch_code=EM_S390;
1892 #elif (defined ALPHA)
1893 static Elf32_Half running_arch_code=EM_ALPHA;
1894 #elif (defined MIPSEL)
1895 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1896 #elif (defined PARISC)
1897 static Elf32_Half running_arch_code=EM_PARISC;
1898 #elif (defined MIPS)
1899 static Elf32_Half running_arch_code=EM_MIPS;
1900 #elif (defined M68K)
1901 static Elf32_Half running_arch_code=EM_68K;
1902 #else
1903 #error Method os::dll_load requires that one of following is defined:\
1904 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1905 #endif
1907 // Identify compatability class for VM's architecture and library's architecture
1908 // Obtain string descriptions for architectures
1910 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1911 int running_arch_index=-1;
1913 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1914 if (running_arch_code == arch_array[i].code) {
1915 running_arch_index = i;
1916 }
1917 if (lib_arch.code == arch_array[i].code) {
1918 lib_arch.compat_class = arch_array[i].compat_class;
1919 lib_arch.name = arch_array[i].name;
1920 }
1921 }
1923 assert(running_arch_index != -1,
1924 "Didn't find running architecture code (running_arch_code) in arch_array");
1925 if (running_arch_index == -1) {
1926 // Even though running architecture detection failed
1927 // we may still continue with reporting dlerror() message
1928 return NULL;
1929 }
1931 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1932 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1933 return NULL;
1934 }
1936 #ifndef S390
1937 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1938 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1939 return NULL;
1940 }
1941 #endif // !S390
1943 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1944 if ( lib_arch.name!=NULL ) {
1945 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1946 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1947 lib_arch.name, arch_array[running_arch_index].name);
1948 } else {
1949 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1950 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1951 lib_arch.code,
1952 arch_array[running_arch_index].name);
1953 }
1954 }
1956 return NULL;
1957 }
1959 /*
1960 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
1961 * chances are you might want to run the generated bits against glibc-2.0
1962 * libdl.so, so always use locking for any version of glibc.
1963 */
1964 void* os::dll_lookup(void* handle, const char* name) {
1965 pthread_mutex_lock(&dl_mutex);
1966 void* res = dlsym(handle, name);
1967 pthread_mutex_unlock(&dl_mutex);
1968 return res;
1969 }
1972 static bool _print_ascii_file(const char* filename, outputStream* st) {
1973 int fd = ::open(filename, O_RDONLY);
1974 if (fd == -1) {
1975 return false;
1976 }
1978 char buf[32];
1979 int bytes;
1980 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
1981 st->print_raw(buf, bytes);
1982 }
1984 ::close(fd);
1986 return true;
1987 }
1989 void os::print_dll_info(outputStream *st) {
1990 st->print_cr("Dynamic libraries:");
1992 char fname[32];
1993 pid_t pid = os::Linux::gettid();
1995 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
1997 if (!_print_ascii_file(fname, st)) {
1998 st->print("Can not get library information for pid = %d\n", pid);
1999 }
2000 }
2002 void os::print_os_info_brief(outputStream* st) {
2003 os::Linux::print_distro_info(st);
2005 os::Posix::print_uname_info(st);
2007 os::Linux::print_libversion_info(st);
2009 }
2011 void os::print_os_info(outputStream* st) {
2012 st->print("OS:");
2014 os::Linux::print_distro_info(st);
2016 os::Posix::print_uname_info(st);
2018 // Print warning if unsafe chroot environment detected
2019 if (unsafe_chroot_detected) {
2020 st->print("WARNING!! ");
2021 st->print_cr(unstable_chroot_error);
2022 }
2024 os::Linux::print_libversion_info(st);
2026 os::Posix::print_rlimit_info(st);
2028 os::Posix::print_load_average(st);
2030 os::Linux::print_full_memory_info(st);
2031 }
2033 // Try to identify popular distros.
2034 // Most Linux distributions have /etc/XXX-release file, which contains
2035 // the OS version string. Some have more than one /etc/XXX-release file
2036 // (e.g. Mandrake has both /etc/mandrake-release and /etc/redhat-release.),
2037 // so the order is important.
2038 void os::Linux::print_distro_info(outputStream* st) {
2039 if (!_print_ascii_file("/etc/mandrake-release", st) &&
2040 !_print_ascii_file("/etc/sun-release", st) &&
2041 !_print_ascii_file("/etc/redhat-release", st) &&
2042 !_print_ascii_file("/etc/SuSE-release", st) &&
2043 !_print_ascii_file("/etc/turbolinux-release", st) &&
2044 !_print_ascii_file("/etc/gentoo-release", st) &&
2045 !_print_ascii_file("/etc/debian_version", st) &&
2046 !_print_ascii_file("/etc/ltib-release", st) &&
2047 !_print_ascii_file("/etc/angstrom-version", st)) {
2048 st->print("Linux");
2049 }
2050 st->cr();
2051 }
2053 void os::Linux::print_libversion_info(outputStream* st) {
2054 // libc, pthread
2055 st->print("libc:");
2056 st->print(os::Linux::glibc_version()); st->print(" ");
2057 st->print(os::Linux::libpthread_version()); st->print(" ");
2058 if (os::Linux::is_LinuxThreads()) {
2059 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2060 }
2061 st->cr();
2062 }
2064 void os::Linux::print_full_memory_info(outputStream* st) {
2065 st->print("\n/proc/meminfo:\n");
2066 _print_ascii_file("/proc/meminfo", st);
2067 st->cr();
2068 }
2070 void os::print_memory_info(outputStream* st) {
2072 st->print("Memory:");
2073 st->print(" %dk page", os::vm_page_size()>>10);
2075 // values in struct sysinfo are "unsigned long"
2076 struct sysinfo si;
2077 sysinfo(&si);
2079 st->print(", physical " UINT64_FORMAT "k",
2080 os::physical_memory() >> 10);
2081 st->print("(" UINT64_FORMAT "k free)",
2082 os::available_memory() >> 10);
2083 st->print(", swap " UINT64_FORMAT "k",
2084 ((jlong)si.totalswap * si.mem_unit) >> 10);
2085 st->print("(" UINT64_FORMAT "k free)",
2086 ((jlong)si.freeswap * si.mem_unit) >> 10);
2087 st->cr();
2088 }
2090 void os::pd_print_cpu_info(outputStream* st) {
2091 st->print("\n/proc/cpuinfo:\n");
2092 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2093 st->print(" <Not Available>");
2094 }
2095 st->cr();
2096 }
2098 // Taken from /usr/include/bits/siginfo.h Supposed to be architecture specific
2099 // but they're the same for all the linux arch that we support
2100 // and they're the same for solaris but there's no common place to put this.
2101 const char *ill_names[] = { "ILL0", "ILL_ILLOPC", "ILL_ILLOPN", "ILL_ILLADR",
2102 "ILL_ILLTRP", "ILL_PRVOPC", "ILL_PRVREG",
2103 "ILL_COPROC", "ILL_BADSTK" };
2105 const char *fpe_names[] = { "FPE0", "FPE_INTDIV", "FPE_INTOVF", "FPE_FLTDIV",
2106 "FPE_FLTOVF", "FPE_FLTUND", "FPE_FLTRES",
2107 "FPE_FLTINV", "FPE_FLTSUB", "FPE_FLTDEN" };
2109 const char *segv_names[] = { "SEGV0", "SEGV_MAPERR", "SEGV_ACCERR" };
2111 const char *bus_names[] = { "BUS0", "BUS_ADRALN", "BUS_ADRERR", "BUS_OBJERR" };
2113 void os::print_siginfo(outputStream* st, void* siginfo) {
2114 st->print("siginfo:");
2116 const int buflen = 100;
2117 char buf[buflen];
2118 siginfo_t *si = (siginfo_t*)siginfo;
2119 st->print("si_signo=%s: ", os::exception_name(si->si_signo, buf, buflen));
2120 if (si->si_errno != 0 && strerror_r(si->si_errno, buf, buflen) == 0) {
2121 st->print("si_errno=%s", buf);
2122 } else {
2123 st->print("si_errno=%d", si->si_errno);
2124 }
2125 const int c = si->si_code;
2126 assert(c > 0, "unexpected si_code");
2127 switch (si->si_signo) {
2128 case SIGILL:
2129 st->print(", si_code=%d (%s)", c, c > 8 ? "" : ill_names[c]);
2130 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2131 break;
2132 case SIGFPE:
2133 st->print(", si_code=%d (%s)", c, c > 9 ? "" : fpe_names[c]);
2134 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2135 break;
2136 case SIGSEGV:
2137 st->print(", si_code=%d (%s)", c, c > 2 ? "" : segv_names[c]);
2138 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2139 break;
2140 case SIGBUS:
2141 st->print(", si_code=%d (%s)", c, c > 3 ? "" : bus_names[c]);
2142 st->print(", si_addr=" PTR_FORMAT, si->si_addr);
2143 break;
2144 default:
2145 st->print(", si_code=%d", si->si_code);
2146 // no si_addr
2147 }
2149 if ((si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2150 UseSharedSpaces) {
2151 FileMapInfo* mapinfo = FileMapInfo::current_info();
2152 if (mapinfo->is_in_shared_space(si->si_addr)) {
2153 st->print("\n\nError accessing class data sharing archive." \
2154 " Mapped file inaccessible during execution, " \
2155 " possible disk/network problem.");
2156 }
2157 }
2158 st->cr();
2159 }
2162 static void print_signal_handler(outputStream* st, int sig,
2163 char* buf, size_t buflen);
2165 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2166 st->print_cr("Signal Handlers:");
2167 print_signal_handler(st, SIGSEGV, buf, buflen);
2168 print_signal_handler(st, SIGBUS , buf, buflen);
2169 print_signal_handler(st, SIGFPE , buf, buflen);
2170 print_signal_handler(st, SIGPIPE, buf, buflen);
2171 print_signal_handler(st, SIGXFSZ, buf, buflen);
2172 print_signal_handler(st, SIGILL , buf, buflen);
2173 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2174 print_signal_handler(st, SR_signum, buf, buflen);
2175 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2176 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2177 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2178 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2179 }
2181 static char saved_jvm_path[MAXPATHLEN] = {0};
2183 // Find the full path to the current module, libjvm.so or libjvm_g.so
2184 void os::jvm_path(char *buf, jint buflen) {
2185 // Error checking.
2186 if (buflen < MAXPATHLEN) {
2187 assert(false, "must use a large-enough buffer");
2188 buf[0] = '\0';
2189 return;
2190 }
2191 // Lazy resolve the path to current module.
2192 if (saved_jvm_path[0] != 0) {
2193 strcpy(buf, saved_jvm_path);
2194 return;
2195 }
2197 char dli_fname[MAXPATHLEN];
2198 bool ret = dll_address_to_library_name(
2199 CAST_FROM_FN_PTR(address, os::jvm_path),
2200 dli_fname, sizeof(dli_fname), NULL);
2201 assert(ret != 0, "cannot locate libjvm");
2202 char *rp = realpath(dli_fname, buf);
2203 if (rp == NULL)
2204 return;
2206 if (Arguments::created_by_gamma_launcher()) {
2207 // Support for the gamma launcher. Typical value for buf is
2208 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2209 // the right place in the string, then assume we are installed in a JDK and
2210 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2211 // up the path so it looks like libjvm.so is installed there (append a
2212 // fake suffix hotspot/libjvm.so).
2213 const char *p = buf + strlen(buf) - 1;
2214 for (int count = 0; p > buf && count < 5; ++count) {
2215 for (--p; p > buf && *p != '/'; --p)
2216 /* empty */ ;
2217 }
2219 if (strncmp(p, "/jre/lib/", 9) != 0) {
2220 // Look for JAVA_HOME in the environment.
2221 char* java_home_var = ::getenv("JAVA_HOME");
2222 if (java_home_var != NULL && java_home_var[0] != 0) {
2223 char* jrelib_p;
2224 int len;
2226 // Check the current module name "libjvm.so" or "libjvm_g.so".
2227 p = strrchr(buf, '/');
2228 assert(strstr(p, "/libjvm") == p, "invalid library name");
2229 p = strstr(p, "_g") ? "_g" : "";
2231 rp = realpath(java_home_var, buf);
2232 if (rp == NULL)
2233 return;
2235 // determine if this is a legacy image or modules image
2236 // modules image doesn't have "jre" subdirectory
2237 len = strlen(buf);
2238 jrelib_p = buf + len;
2239 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2240 if (0 != access(buf, F_OK)) {
2241 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2242 }
2244 if (0 == access(buf, F_OK)) {
2245 // Use current module name "libjvm[_g].so" instead of
2246 // "libjvm"debug_only("_g")".so" since for fastdebug version
2247 // we should have "libjvm.so" but debug_only("_g") adds "_g"!
2248 len = strlen(buf);
2249 snprintf(buf + len, buflen-len, "/hotspot/libjvm%s.so", p);
2250 } else {
2251 // Go back to path of .so
2252 rp = realpath(dli_fname, buf);
2253 if (rp == NULL)
2254 return;
2255 }
2256 }
2257 }
2258 }
2260 strcpy(saved_jvm_path, buf);
2261 }
2263 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2264 // no prefix required, not even "_"
2265 }
2267 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2268 // no suffix required
2269 }
2271 ////////////////////////////////////////////////////////////////////////////////
2272 // sun.misc.Signal support
2274 static volatile jint sigint_count = 0;
2276 static void
2277 UserHandler(int sig, void *siginfo, void *context) {
2278 // 4511530 - sem_post is serialized and handled by the manager thread. When
2279 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2280 // don't want to flood the manager thread with sem_post requests.
2281 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2282 return;
2284 // Ctrl-C is pressed during error reporting, likely because the error
2285 // handler fails to abort. Let VM die immediately.
2286 if (sig == SIGINT && is_error_reported()) {
2287 os::die();
2288 }
2290 os::signal_notify(sig);
2291 }
2293 void* os::user_handler() {
2294 return CAST_FROM_FN_PTR(void*, UserHandler);
2295 }
2297 extern "C" {
2298 typedef void (*sa_handler_t)(int);
2299 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2300 }
2302 void* os::signal(int signal_number, void* handler) {
2303 struct sigaction sigAct, oldSigAct;
2305 sigfillset(&(sigAct.sa_mask));
2306 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2307 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2309 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2310 // -1 means registration failed
2311 return (void *)-1;
2312 }
2314 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2315 }
2317 void os::signal_raise(int signal_number) {
2318 ::raise(signal_number);
2319 }
2321 /*
2322 * The following code is moved from os.cpp for making this
2323 * code platform specific, which it is by its very nature.
2324 */
2326 // Will be modified when max signal is changed to be dynamic
2327 int os::sigexitnum_pd() {
2328 return NSIG;
2329 }
2331 // a counter for each possible signal value
2332 static volatile jint pending_signals[NSIG+1] = { 0 };
2334 // Linux(POSIX) specific hand shaking semaphore.
2335 static sem_t sig_sem;
2337 void os::signal_init_pd() {
2338 // Initialize signal structures
2339 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2341 // Initialize signal semaphore
2342 ::sem_init(&sig_sem, 0, 0);
2343 }
2345 void os::signal_notify(int sig) {
2346 Atomic::inc(&pending_signals[sig]);
2347 ::sem_post(&sig_sem);
2348 }
2350 static int check_pending_signals(bool wait) {
2351 Atomic::store(0, &sigint_count);
2352 for (;;) {
2353 for (int i = 0; i < NSIG + 1; i++) {
2354 jint n = pending_signals[i];
2355 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2356 return i;
2357 }
2358 }
2359 if (!wait) {
2360 return -1;
2361 }
2362 JavaThread *thread = JavaThread::current();
2363 ThreadBlockInVM tbivm(thread);
2365 bool threadIsSuspended;
2366 do {
2367 thread->set_suspend_equivalent();
2368 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2369 ::sem_wait(&sig_sem);
2371 // were we externally suspended while we were waiting?
2372 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2373 if (threadIsSuspended) {
2374 //
2375 // The semaphore has been incremented, but while we were waiting
2376 // another thread suspended us. We don't want to continue running
2377 // while suspended because that would surprise the thread that
2378 // suspended us.
2379 //
2380 ::sem_post(&sig_sem);
2382 thread->java_suspend_self();
2383 }
2384 } while (threadIsSuspended);
2385 }
2386 }
2388 int os::signal_lookup() {
2389 return check_pending_signals(false);
2390 }
2392 int os::signal_wait() {
2393 return check_pending_signals(true);
2394 }
2396 ////////////////////////////////////////////////////////////////////////////////
2397 // Virtual Memory
2399 int os::vm_page_size() {
2400 // Seems redundant as all get out
2401 assert(os::Linux::page_size() != -1, "must call os::init");
2402 return os::Linux::page_size();
2403 }
2405 // Solaris allocates memory by pages.
2406 int os::vm_allocation_granularity() {
2407 assert(os::Linux::page_size() != -1, "must call os::init");
2408 return os::Linux::page_size();
2409 }
2411 // Rationale behind this function:
2412 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2413 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2414 // samples for JITted code. Here we create private executable mapping over the code cache
2415 // and then we can use standard (well, almost, as mapping can change) way to provide
2416 // info for the reporting script by storing timestamp and location of symbol
2417 void linux_wrap_code(char* base, size_t size) {
2418 static volatile jint cnt = 0;
2420 if (!UseOprofile) {
2421 return;
2422 }
2424 char buf[PATH_MAX+1];
2425 int num = Atomic::add(1, &cnt);
2427 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2428 os::get_temp_directory(), os::current_process_id(), num);
2429 unlink(buf);
2431 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2433 if (fd != -1) {
2434 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2435 if (rv != (off_t)-1) {
2436 if (::write(fd, "", 1) == 1) {
2437 mmap(base, size,
2438 PROT_READ|PROT_WRITE|PROT_EXEC,
2439 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2440 }
2441 }
2442 ::close(fd);
2443 unlink(buf);
2444 }
2445 }
2447 // NOTE: Linux kernel does not really reserve the pages for us.
2448 // All it does is to check if there are enough free pages
2449 // left at the time of mmap(). This could be a potential
2450 // problem.
2451 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2452 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2453 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2454 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2455 if (res != (uintptr_t) MAP_FAILED) {
2456 if (UseNUMAInterleaving) {
2457 numa_make_global(addr, size);
2458 }
2459 return true;
2460 }
2461 return false;
2462 }
2464 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2465 #ifndef MAP_HUGETLB
2466 #define MAP_HUGETLB 0x40000
2467 #endif
2469 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2470 #ifndef MADV_HUGEPAGE
2471 #define MADV_HUGEPAGE 14
2472 #endif
2474 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2475 bool exec) {
2476 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2477 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2478 uintptr_t res =
2479 (uintptr_t) ::mmap(addr, size, prot,
2480 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_HUGETLB,
2481 -1, 0);
2482 if (res != (uintptr_t) MAP_FAILED) {
2483 if (UseNUMAInterleaving) {
2484 numa_make_global(addr, size);
2485 }
2486 return true;
2487 }
2488 // Fall through and try to use small pages
2489 }
2491 if (commit_memory(addr, size, exec)) {
2492 realign_memory(addr, size, alignment_hint);
2493 return true;
2494 }
2495 return false;
2496 }
2498 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2499 if (UseHugeTLBFS && alignment_hint > (size_t)vm_page_size()) {
2500 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2501 // be supported or the memory may already be backed by huge pages.
2502 ::madvise(addr, bytes, MADV_HUGEPAGE);
2503 }
2504 }
2506 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2507 // This method works by doing an mmap over an existing mmaping and effectively discarding
2508 // the existing pages. However it won't work for SHM-based large pages that cannot be
2509 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2510 // small pages on top of the SHM segment. This method always works for small pages, so we
2511 // allow that in any case.
2512 if (alignment_hint <= (size_t)os::vm_page_size() || !UseSHM) {
2513 commit_memory(addr, bytes, alignment_hint, false);
2514 }
2515 }
2517 void os::numa_make_global(char *addr, size_t bytes) {
2518 Linux::numa_interleave_memory(addr, bytes);
2519 }
2521 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2522 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2523 }
2525 bool os::numa_topology_changed() { return false; }
2527 size_t os::numa_get_groups_num() {
2528 int max_node = Linux::numa_max_node();
2529 return max_node > 0 ? max_node + 1 : 1;
2530 }
2532 int os::numa_get_group_id() {
2533 int cpu_id = Linux::sched_getcpu();
2534 if (cpu_id != -1) {
2535 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2536 if (lgrp_id != -1) {
2537 return lgrp_id;
2538 }
2539 }
2540 return 0;
2541 }
2543 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2544 for (size_t i = 0; i < size; i++) {
2545 ids[i] = i;
2546 }
2547 return size;
2548 }
2550 bool os::get_page_info(char *start, page_info* info) {
2551 return false;
2552 }
2554 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2555 return end;
2556 }
2559 int os::Linux::sched_getcpu_syscall(void) {
2560 unsigned int cpu;
2561 int retval = -1;
2563 #if defined(IA32)
2564 # ifndef SYS_getcpu
2565 # define SYS_getcpu 318
2566 # endif
2567 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2568 #elif defined(AMD64)
2569 // Unfortunately we have to bring all these macros here from vsyscall.h
2570 // to be able to compile on old linuxes.
2571 # define __NR_vgetcpu 2
2572 # define VSYSCALL_START (-10UL << 20)
2573 # define VSYSCALL_SIZE 1024
2574 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2575 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2576 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2577 retval = vgetcpu(&cpu, NULL, NULL);
2578 #endif
2580 return (retval == -1) ? retval : cpu;
2581 }
2583 // Something to do with the numa-aware allocator needs these symbols
2584 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2585 extern "C" JNIEXPORT void numa_error(char *where) { }
2586 extern "C" JNIEXPORT int fork1() { return fork(); }
2589 // If we are running with libnuma version > 2, then we should
2590 // be trying to use symbols with versions 1.1
2591 // If we are running with earlier version, which did not have symbol versions,
2592 // we should use the base version.
2593 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2594 void *f = dlvsym(handle, name, "libnuma_1.1");
2595 if (f == NULL) {
2596 f = dlsym(handle, name);
2597 }
2598 return f;
2599 }
2601 bool os::Linux::libnuma_init() {
2602 // sched_getcpu() should be in libc.
2603 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2604 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2606 // If it's not, try a direct syscall.
2607 if (sched_getcpu() == -1)
2608 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2610 if (sched_getcpu() != -1) { // Does it work?
2611 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2612 if (handle != NULL) {
2613 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2614 libnuma_dlsym(handle, "numa_node_to_cpus")));
2615 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2616 libnuma_dlsym(handle, "numa_max_node")));
2617 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2618 libnuma_dlsym(handle, "numa_available")));
2619 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2620 libnuma_dlsym(handle, "numa_tonode_memory")));
2621 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2622 libnuma_dlsym(handle, "numa_interleave_memory")));
2625 if (numa_available() != -1) {
2626 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2627 // Create a cpu -> node mapping
2628 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2629 rebuild_cpu_to_node_map();
2630 return true;
2631 }
2632 }
2633 }
2634 return false;
2635 }
2637 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2638 // The table is later used in get_node_by_cpu().
2639 void os::Linux::rebuild_cpu_to_node_map() {
2640 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2641 // in libnuma (possible values are starting from 16,
2642 // and continuing up with every other power of 2, but less
2643 // than the maximum number of CPUs supported by kernel), and
2644 // is a subject to change (in libnuma version 2 the requirements
2645 // are more reasonable) we'll just hardcode the number they use
2646 // in the library.
2647 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2649 size_t cpu_num = os::active_processor_count();
2650 size_t cpu_map_size = NCPUS / BitsPerCLong;
2651 size_t cpu_map_valid_size =
2652 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2654 cpu_to_node()->clear();
2655 cpu_to_node()->at_grow(cpu_num - 1);
2656 size_t node_num = numa_get_groups_num();
2658 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2659 for (size_t i = 0; i < node_num; i++) {
2660 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2661 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2662 if (cpu_map[j] != 0) {
2663 for (size_t k = 0; k < BitsPerCLong; k++) {
2664 if (cpu_map[j] & (1UL << k)) {
2665 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2666 }
2667 }
2668 }
2669 }
2670 }
2671 }
2672 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2673 }
2675 int os::Linux::get_node_by_cpu(int cpu_id) {
2676 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2677 return cpu_to_node()->at(cpu_id);
2678 }
2679 return -1;
2680 }
2682 GrowableArray<int>* os::Linux::_cpu_to_node;
2683 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2684 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2685 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2686 os::Linux::numa_available_func_t os::Linux::_numa_available;
2687 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2688 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2689 unsigned long* os::Linux::_numa_all_nodes;
2691 bool os::pd_uncommit_memory(char* addr, size_t size) {
2692 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2693 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2694 return res != (uintptr_t) MAP_FAILED;
2695 }
2697 // Linux uses a growable mapping for the stack, and if the mapping for
2698 // the stack guard pages is not removed when we detach a thread the
2699 // stack cannot grow beyond the pages where the stack guard was
2700 // mapped. If at some point later in the process the stack expands to
2701 // that point, the Linux kernel cannot expand the stack any further
2702 // because the guard pages are in the way, and a segfault occurs.
2703 //
2704 // However, it's essential not to split the stack region by unmapping
2705 // a region (leaving a hole) that's already part of the stack mapping,
2706 // so if the stack mapping has already grown beyond the guard pages at
2707 // the time we create them, we have to truncate the stack mapping.
2708 // So, we need to know the extent of the stack mapping when
2709 // create_stack_guard_pages() is called.
2711 // Find the bounds of the stack mapping. Return true for success.
2712 //
2713 // We only need this for stacks that are growable: at the time of
2714 // writing thread stacks don't use growable mappings (i.e. those
2715 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2716 // only applies to the main thread.
2718 static
2719 bool get_stack_bounds(uintptr_t *bottom, uintptr_t *top) {
2721 char buf[128];
2722 int fd, sz;
2724 if ((fd = ::open("/proc/self/maps", O_RDONLY)) < 0) {
2725 return false;
2726 }
2728 const char kw[] = "[stack]";
2729 const int kwlen = sizeof(kw)-1;
2731 // Address part of /proc/self/maps couldn't be more than 128 bytes
2732 while ((sz = os::get_line_chars(fd, buf, sizeof(buf))) > 0) {
2733 if (sz > kwlen && ::memcmp(buf+sz-kwlen, kw, kwlen) == 0) {
2734 // Extract addresses
2735 if (sscanf(buf, "%" SCNxPTR "-%" SCNxPTR, bottom, top) == 2) {
2736 uintptr_t sp = (uintptr_t) __builtin_frame_address(0);
2737 if (sp >= *bottom && sp <= *top) {
2738 ::close(fd);
2739 return true;
2740 }
2741 }
2742 }
2743 }
2745 ::close(fd);
2746 return false;
2747 }
2750 // If the (growable) stack mapping already extends beyond the point
2751 // where we're going to put our guard pages, truncate the mapping at
2752 // that point by munmap()ping it. This ensures that when we later
2753 // munmap() the guard pages we don't leave a hole in the stack
2754 // mapping. This only affects the main/initial thread, but guard
2755 // against future OS changes
2756 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2757 uintptr_t stack_extent, stack_base;
2758 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2759 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2760 assert(os::Linux::is_initial_thread(),
2761 "growable stack in non-initial thread");
2762 if (stack_extent < (uintptr_t)addr)
2763 ::munmap((void*)stack_extent, (uintptr_t)addr - stack_extent);
2764 }
2766 return os::commit_memory(addr, size);
2767 }
2769 // If this is a growable mapping, remove the guard pages entirely by
2770 // munmap()ping them. If not, just call uncommit_memory(). This only
2771 // affects the main/initial thread, but guard against future OS changes
2772 bool os::remove_stack_guard_pages(char* addr, size_t size) {
2773 uintptr_t stack_extent, stack_base;
2774 bool chk_bounds = NOT_DEBUG(os::Linux::is_initial_thread()) DEBUG_ONLY(true);
2775 if (chk_bounds && get_stack_bounds(&stack_extent, &stack_base)) {
2776 assert(os::Linux::is_initial_thread(),
2777 "growable stack in non-initial thread");
2779 return ::munmap(addr, size) == 0;
2780 }
2782 return os::uncommit_memory(addr, size);
2783 }
2785 static address _highest_vm_reserved_address = NULL;
2787 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
2788 // at 'requested_addr'. If there are existing memory mappings at the same
2789 // location, however, they will be overwritten. If 'fixed' is false,
2790 // 'requested_addr' is only treated as a hint, the return value may or
2791 // may not start from the requested address. Unlike Linux mmap(), this
2792 // function returns NULL to indicate failure.
2793 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
2794 char * addr;
2795 int flags;
2797 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
2798 if (fixed) {
2799 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
2800 flags |= MAP_FIXED;
2801 }
2803 // Map uncommitted pages PROT_READ and PROT_WRITE, change access
2804 // to PROT_EXEC if executable when we commit the page.
2805 addr = (char*)::mmap(requested_addr, bytes, PROT_READ|PROT_WRITE,
2806 flags, -1, 0);
2808 if (addr != MAP_FAILED) {
2809 // anon_mmap() should only get called during VM initialization,
2810 // don't need lock (actually we can skip locking even it can be called
2811 // from multiple threads, because _highest_vm_reserved_address is just a
2812 // hint about the upper limit of non-stack memory regions.)
2813 if ((address)addr + bytes > _highest_vm_reserved_address) {
2814 _highest_vm_reserved_address = (address)addr + bytes;
2815 }
2816 }
2818 return addr == MAP_FAILED ? NULL : addr;
2819 }
2821 // Don't update _highest_vm_reserved_address, because there might be memory
2822 // regions above addr + size. If so, releasing a memory region only creates
2823 // a hole in the address space, it doesn't help prevent heap-stack collision.
2824 //
2825 static int anon_munmap(char * addr, size_t size) {
2826 return ::munmap(addr, size) == 0;
2827 }
2829 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
2830 size_t alignment_hint) {
2831 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
2832 }
2834 bool os::pd_release_memory(char* addr, size_t size) {
2835 return anon_munmap(addr, size);
2836 }
2838 static address highest_vm_reserved_address() {
2839 return _highest_vm_reserved_address;
2840 }
2842 static bool linux_mprotect(char* addr, size_t size, int prot) {
2843 // Linux wants the mprotect address argument to be page aligned.
2844 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
2846 // According to SUSv3, mprotect() should only be used with mappings
2847 // established by mmap(), and mmap() always maps whole pages. Unaligned
2848 // 'addr' likely indicates problem in the VM (e.g. trying to change
2849 // protection of malloc'ed or statically allocated memory). Check the
2850 // caller if you hit this assert.
2851 assert(addr == bottom, "sanity check");
2853 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
2854 return ::mprotect(bottom, size, prot) == 0;
2855 }
2857 // Set protections specified
2858 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
2859 bool is_committed) {
2860 unsigned int p = 0;
2861 switch (prot) {
2862 case MEM_PROT_NONE: p = PROT_NONE; break;
2863 case MEM_PROT_READ: p = PROT_READ; break;
2864 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
2865 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
2866 default:
2867 ShouldNotReachHere();
2868 }
2869 // is_committed is unused.
2870 return linux_mprotect(addr, bytes, p);
2871 }
2873 bool os::guard_memory(char* addr, size_t size) {
2874 return linux_mprotect(addr, size, PROT_NONE);
2875 }
2877 bool os::unguard_memory(char* addr, size_t size) {
2878 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
2879 }
2881 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
2882 bool result = false;
2883 void *p = mmap (NULL, page_size, PROT_READ|PROT_WRITE,
2884 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
2885 -1, 0);
2887 if (p != (void *) -1) {
2888 // We don't know if this really is a huge page or not.
2889 FILE *fp = fopen("/proc/self/maps", "r");
2890 if (fp) {
2891 while (!feof(fp)) {
2892 char chars[257];
2893 long x = 0;
2894 if (fgets(chars, sizeof(chars), fp)) {
2895 if (sscanf(chars, "%lx-%*x", &x) == 1
2896 && x == (long)p) {
2897 if (strstr (chars, "hugepage")) {
2898 result = true;
2899 break;
2900 }
2901 }
2902 }
2903 }
2904 fclose(fp);
2905 }
2906 munmap (p, page_size);
2907 if (result)
2908 return true;
2909 }
2911 if (warn) {
2912 warning("HugeTLBFS is not supported by the operating system.");
2913 }
2915 return result;
2916 }
2918 /*
2919 * Set the coredump_filter bits to include largepages in core dump (bit 6)
2920 *
2921 * From the coredump_filter documentation:
2922 *
2923 * - (bit 0) anonymous private memory
2924 * - (bit 1) anonymous shared memory
2925 * - (bit 2) file-backed private memory
2926 * - (bit 3) file-backed shared memory
2927 * - (bit 4) ELF header pages in file-backed private memory areas (it is
2928 * effective only if the bit 2 is cleared)
2929 * - (bit 5) hugetlb private memory
2930 * - (bit 6) hugetlb shared memory
2931 */
2932 static void set_coredump_filter(void) {
2933 FILE *f;
2934 long cdm;
2936 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
2937 return;
2938 }
2940 if (fscanf(f, "%lx", &cdm) != 1) {
2941 fclose(f);
2942 return;
2943 }
2945 rewind(f);
2947 if ((cdm & LARGEPAGES_BIT) == 0) {
2948 cdm |= LARGEPAGES_BIT;
2949 fprintf(f, "%#lx", cdm);
2950 }
2952 fclose(f);
2953 }
2955 // Large page support
2957 static size_t _large_page_size = 0;
2959 void os::large_page_init() {
2960 if (!UseLargePages) {
2961 UseHugeTLBFS = false;
2962 UseSHM = false;
2963 return;
2964 }
2966 if (FLAG_IS_DEFAULT(UseHugeTLBFS) && FLAG_IS_DEFAULT(UseSHM)) {
2967 // If UseLargePages is specified on the command line try both methods,
2968 // if it's default, then try only HugeTLBFS.
2969 if (FLAG_IS_DEFAULT(UseLargePages)) {
2970 UseHugeTLBFS = true;
2971 } else {
2972 UseHugeTLBFS = UseSHM = true;
2973 }
2974 }
2976 if (LargePageSizeInBytes) {
2977 _large_page_size = LargePageSizeInBytes;
2978 } else {
2979 // large_page_size on Linux is used to round up heap size. x86 uses either
2980 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
2981 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
2982 // page as large as 256M.
2983 //
2984 // Here we try to figure out page size by parsing /proc/meminfo and looking
2985 // for a line with the following format:
2986 // Hugepagesize: 2048 kB
2987 //
2988 // If we can't determine the value (e.g. /proc is not mounted, or the text
2989 // format has been changed), we'll use the largest page size supported by
2990 // the processor.
2992 #ifndef ZERO
2993 _large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
2994 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
2995 #endif // ZERO
2997 FILE *fp = fopen("/proc/meminfo", "r");
2998 if (fp) {
2999 while (!feof(fp)) {
3000 int x = 0;
3001 char buf[16];
3002 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3003 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3004 _large_page_size = x * K;
3005 break;
3006 }
3007 } else {
3008 // skip to next line
3009 for (;;) {
3010 int ch = fgetc(fp);
3011 if (ch == EOF || ch == (int)'\n') break;
3012 }
3013 }
3014 }
3015 fclose(fp);
3016 }
3017 }
3019 // print a warning if any large page related flag is specified on command line
3020 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3022 const size_t default_page_size = (size_t)Linux::page_size();
3023 if (_large_page_size > default_page_size) {
3024 _page_sizes[0] = _large_page_size;
3025 _page_sizes[1] = default_page_size;
3026 _page_sizes[2] = 0;
3027 }
3028 UseHugeTLBFS = UseHugeTLBFS &&
3029 Linux::hugetlbfs_sanity_check(warn_on_failure, _large_page_size);
3031 if (UseHugeTLBFS)
3032 UseSHM = false;
3034 UseLargePages = UseHugeTLBFS || UseSHM;
3036 set_coredump_filter();
3037 }
3039 #ifndef SHM_HUGETLB
3040 #define SHM_HUGETLB 04000
3041 #endif
3043 char* os::reserve_memory_special(size_t bytes, char* req_addr, bool exec) {
3044 // "exec" is passed in but not used. Creating the shared image for
3045 // the code cache doesn't have an SHM_X executable permission to check.
3046 assert(UseLargePages && UseSHM, "only for SHM large pages");
3048 key_t key = IPC_PRIVATE;
3049 char *addr;
3051 bool warn_on_failure = UseLargePages &&
3052 (!FLAG_IS_DEFAULT(UseLargePages) ||
3053 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3054 );
3055 char msg[128];
3057 // Create a large shared memory region to attach to based on size.
3058 // Currently, size is the total size of the heap
3059 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3060 if (shmid == -1) {
3061 // Possible reasons for shmget failure:
3062 // 1. shmmax is too small for Java heap.
3063 // > check shmmax value: cat /proc/sys/kernel/shmmax
3064 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3065 // 2. not enough large page memory.
3066 // > check available large pages: cat /proc/meminfo
3067 // > increase amount of large pages:
3068 // echo new_value > /proc/sys/vm/nr_hugepages
3069 // Note 1: different Linux may use different name for this property,
3070 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3071 // Note 2: it's possible there's enough physical memory available but
3072 // they are so fragmented after a long run that they can't
3073 // coalesce into large pages. Try to reserve large pages when
3074 // the system is still "fresh".
3075 if (warn_on_failure) {
3076 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3077 warning(msg);
3078 }
3079 return NULL;
3080 }
3082 // attach to the region
3083 addr = (char*)shmat(shmid, req_addr, 0);
3084 int err = errno;
3086 // Remove shmid. If shmat() is successful, the actual shared memory segment
3087 // will be deleted when it's detached by shmdt() or when the process
3088 // terminates. If shmat() is not successful this will remove the shared
3089 // segment immediately.
3090 shmctl(shmid, IPC_RMID, NULL);
3092 if ((intptr_t)addr == -1) {
3093 if (warn_on_failure) {
3094 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3095 warning(msg);
3096 }
3097 return NULL;
3098 }
3100 if ((addr != NULL) && UseNUMAInterleaving) {
3101 numa_make_global(addr, bytes);
3102 }
3104 return addr;
3105 }
3107 bool os::release_memory_special(char* base, size_t bytes) {
3108 // detaching the SHM segment will also delete it, see reserve_memory_special()
3109 int rslt = shmdt(base);
3110 return rslt == 0;
3111 }
3113 size_t os::large_page_size() {
3114 return _large_page_size;
3115 }
3117 // HugeTLBFS allows application to commit large page memory on demand;
3118 // with SysV SHM the entire memory region must be allocated as shared
3119 // memory.
3120 bool os::can_commit_large_page_memory() {
3121 return UseHugeTLBFS;
3122 }
3124 bool os::can_execute_large_page_memory() {
3125 return UseHugeTLBFS;
3126 }
3128 // Reserve memory at an arbitrary address, only if that area is
3129 // available (and not reserved for something else).
3131 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3132 const int max_tries = 10;
3133 char* base[max_tries];
3134 size_t size[max_tries];
3135 const size_t gap = 0x000000;
3137 // Assert only that the size is a multiple of the page size, since
3138 // that's all that mmap requires, and since that's all we really know
3139 // about at this low abstraction level. If we need higher alignment,
3140 // we can either pass an alignment to this method or verify alignment
3141 // in one of the methods further up the call chain. See bug 5044738.
3142 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3144 // Repeatedly allocate blocks until the block is allocated at the
3145 // right spot. Give up after max_tries. Note that reserve_memory() will
3146 // automatically update _highest_vm_reserved_address if the call is
3147 // successful. The variable tracks the highest memory address every reserved
3148 // by JVM. It is used to detect heap-stack collision if running with
3149 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3150 // space than needed, it could confuse the collision detecting code. To
3151 // solve the problem, save current _highest_vm_reserved_address and
3152 // calculate the correct value before return.
3153 address old_highest = _highest_vm_reserved_address;
3155 // Linux mmap allows caller to pass an address as hint; give it a try first,
3156 // if kernel honors the hint then we can return immediately.
3157 char * addr = anon_mmap(requested_addr, bytes, false);
3158 if (addr == requested_addr) {
3159 return requested_addr;
3160 }
3162 if (addr != NULL) {
3163 // mmap() is successful but it fails to reserve at the requested address
3164 anon_munmap(addr, bytes);
3165 }
3167 int i;
3168 for (i = 0; i < max_tries; ++i) {
3169 base[i] = reserve_memory(bytes);
3171 if (base[i] != NULL) {
3172 // Is this the block we wanted?
3173 if (base[i] == requested_addr) {
3174 size[i] = bytes;
3175 break;
3176 }
3178 // Does this overlap the block we wanted? Give back the overlapped
3179 // parts and try again.
3181 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3182 if (top_overlap >= 0 && top_overlap < bytes) {
3183 unmap_memory(base[i], top_overlap);
3184 base[i] += top_overlap;
3185 size[i] = bytes - top_overlap;
3186 } else {
3187 size_t bottom_overlap = base[i] + bytes - requested_addr;
3188 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3189 unmap_memory(requested_addr, bottom_overlap);
3190 size[i] = bytes - bottom_overlap;
3191 } else {
3192 size[i] = bytes;
3193 }
3194 }
3195 }
3196 }
3198 // Give back the unused reserved pieces.
3200 for (int j = 0; j < i; ++j) {
3201 if (base[j] != NULL) {
3202 unmap_memory(base[j], size[j]);
3203 }
3204 }
3206 if (i < max_tries) {
3207 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3208 return requested_addr;
3209 } else {
3210 _highest_vm_reserved_address = old_highest;
3211 return NULL;
3212 }
3213 }
3215 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3216 return ::read(fd, buf, nBytes);
3217 }
3219 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3220 // Solaris uses poll(), linux uses park().
3221 // Poll() is likely a better choice, assuming that Thread.interrupt()
3222 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3223 // SIGSEGV, see 4355769.
3225 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3226 assert(thread == Thread::current(), "thread consistency check");
3228 ParkEvent * const slp = thread->_SleepEvent ;
3229 slp->reset() ;
3230 OrderAccess::fence() ;
3232 if (interruptible) {
3233 jlong prevtime = javaTimeNanos();
3235 for (;;) {
3236 if (os::is_interrupted(thread, true)) {
3237 return OS_INTRPT;
3238 }
3240 jlong newtime = javaTimeNanos();
3242 if (newtime - prevtime < 0) {
3243 // time moving backwards, should only happen if no monotonic clock
3244 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3245 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3246 } else {
3247 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3248 }
3250 if(millis <= 0) {
3251 return OS_OK;
3252 }
3254 prevtime = newtime;
3256 {
3257 assert(thread->is_Java_thread(), "sanity check");
3258 JavaThread *jt = (JavaThread *) thread;
3259 ThreadBlockInVM tbivm(jt);
3260 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3262 jt->set_suspend_equivalent();
3263 // cleared by handle_special_suspend_equivalent_condition() or
3264 // java_suspend_self() via check_and_wait_while_suspended()
3266 slp->park(millis);
3268 // were we externally suspended while we were waiting?
3269 jt->check_and_wait_while_suspended();
3270 }
3271 }
3272 } else {
3273 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3274 jlong prevtime = javaTimeNanos();
3276 for (;;) {
3277 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3278 // the 1st iteration ...
3279 jlong newtime = javaTimeNanos();
3281 if (newtime - prevtime < 0) {
3282 // time moving backwards, should only happen if no monotonic clock
3283 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3284 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3285 } else {
3286 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3287 }
3289 if(millis <= 0) break ;
3291 prevtime = newtime;
3292 slp->park(millis);
3293 }
3294 return OS_OK ;
3295 }
3296 }
3298 int os::naked_sleep() {
3299 // %% make the sleep time an integer flag. for now use 1 millisec.
3300 return os::sleep(Thread::current(), 1, false);
3301 }
3303 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3304 void os::infinite_sleep() {
3305 while (true) { // sleep forever ...
3306 ::sleep(100); // ... 100 seconds at a time
3307 }
3308 }
3310 // Used to convert frequent JVM_Yield() to nops
3311 bool os::dont_yield() {
3312 return DontYieldALot;
3313 }
3315 void os::yield() {
3316 sched_yield();
3317 }
3319 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3321 void os::yield_all(int attempts) {
3322 // Yields to all threads, including threads with lower priorities
3323 // Threads on Linux are all with same priority. The Solaris style
3324 // os::yield_all() with nanosleep(1ms) is not necessary.
3325 sched_yield();
3326 }
3328 // Called from the tight loops to possibly influence time-sharing heuristics
3329 void os::loop_breaker(int attempts) {
3330 os::yield_all(attempts);
3331 }
3333 ////////////////////////////////////////////////////////////////////////////////
3334 // thread priority support
3336 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3337 // only supports dynamic priority, static priority must be zero. For real-time
3338 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3339 // However, for large multi-threaded applications, SCHED_RR is not only slower
3340 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3341 // of 5 runs - Sep 2005).
3342 //
3343 // The following code actually changes the niceness of kernel-thread/LWP. It
3344 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3345 // not the entire user process, and user level threads are 1:1 mapped to kernel
3346 // threads. It has always been the case, but could change in the future. For
3347 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3348 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3350 int os::java_to_os_priority[CriticalPriority + 1] = {
3351 19, // 0 Entry should never be used
3353 4, // 1 MinPriority
3354 3, // 2
3355 2, // 3
3357 1, // 4
3358 0, // 5 NormPriority
3359 -1, // 6
3361 -2, // 7
3362 -3, // 8
3363 -4, // 9 NearMaxPriority
3365 -5, // 10 MaxPriority
3367 -5 // 11 CriticalPriority
3368 };
3370 static int prio_init() {
3371 if (ThreadPriorityPolicy == 1) {
3372 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3373 // if effective uid is not root. Perhaps, a more elegant way of doing
3374 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3375 if (geteuid() != 0) {
3376 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3377 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3378 }
3379 ThreadPriorityPolicy = 0;
3380 }
3381 }
3382 if (UseCriticalJavaThreadPriority) {
3383 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3384 }
3385 return 0;
3386 }
3388 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3389 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3391 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3392 return (ret == 0) ? OS_OK : OS_ERR;
3393 }
3395 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3396 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3397 *priority_ptr = java_to_os_priority[NormPriority];
3398 return OS_OK;
3399 }
3401 errno = 0;
3402 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3403 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3404 }
3406 // Hint to the underlying OS that a task switch would not be good.
3407 // Void return because it's a hint and can fail.
3408 void os::hint_no_preempt() {}
3410 ////////////////////////////////////////////////////////////////////////////////
3411 // suspend/resume support
3413 // the low-level signal-based suspend/resume support is a remnant from the
3414 // old VM-suspension that used to be for java-suspension, safepoints etc,
3415 // within hotspot. Now there is a single use-case for this:
3416 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3417 // that runs in the watcher thread.
3418 // The remaining code is greatly simplified from the more general suspension
3419 // code that used to be used.
3420 //
3421 // The protocol is quite simple:
3422 // - suspend:
3423 // - sends a signal to the target thread
3424 // - polls the suspend state of the osthread using a yield loop
3425 // - target thread signal handler (SR_handler) sets suspend state
3426 // and blocks in sigsuspend until continued
3427 // - resume:
3428 // - sets target osthread state to continue
3429 // - sends signal to end the sigsuspend loop in the SR_handler
3430 //
3431 // Note that the SR_lock plays no role in this suspend/resume protocol.
3432 //
3434 static void resume_clear_context(OSThread *osthread) {
3435 osthread->set_ucontext(NULL);
3436 osthread->set_siginfo(NULL);
3438 // notify the suspend action is completed, we have now resumed
3439 osthread->sr.clear_suspended();
3440 }
3442 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3443 osthread->set_ucontext(context);
3444 osthread->set_siginfo(siginfo);
3445 }
3447 //
3448 // Handler function invoked when a thread's execution is suspended or
3449 // resumed. We have to be careful that only async-safe functions are
3450 // called here (Note: most pthread functions are not async safe and
3451 // should be avoided.)
3452 //
3453 // Note: sigwait() is a more natural fit than sigsuspend() from an
3454 // interface point of view, but sigwait() prevents the signal hander
3455 // from being run. libpthread would get very confused by not having
3456 // its signal handlers run and prevents sigwait()'s use with the
3457 // mutex granting granting signal.
3458 //
3459 // Currently only ever called on the VMThread
3460 //
3461 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3462 // Save and restore errno to avoid confusing native code with EINTR
3463 // after sigsuspend.
3464 int old_errno = errno;
3466 Thread* thread = Thread::current();
3467 OSThread* osthread = thread->osthread();
3468 assert(thread->is_VM_thread(), "Must be VMThread");
3469 // read current suspend action
3470 int action = osthread->sr.suspend_action();
3471 if (action == SR_SUSPEND) {
3472 suspend_save_context(osthread, siginfo, context);
3474 // Notify the suspend action is about to be completed. do_suspend()
3475 // waits until SR_SUSPENDED is set and then returns. We will wait
3476 // here for a resume signal and that completes the suspend-other
3477 // action. do_suspend/do_resume is always called as a pair from
3478 // the same thread - so there are no races
3480 // notify the caller
3481 osthread->sr.set_suspended();
3483 sigset_t suspend_set; // signals for sigsuspend()
3485 // get current set of blocked signals and unblock resume signal
3486 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3487 sigdelset(&suspend_set, SR_signum);
3489 // wait here until we are resumed
3490 do {
3491 sigsuspend(&suspend_set);
3492 // ignore all returns until we get a resume signal
3493 } while (osthread->sr.suspend_action() != SR_CONTINUE);
3495 resume_clear_context(osthread);
3497 } else {
3498 assert(action == SR_CONTINUE, "unexpected sr action");
3499 // nothing special to do - just leave the handler
3500 }
3502 errno = old_errno;
3503 }
3506 static int SR_initialize() {
3507 struct sigaction act;
3508 char *s;
3509 /* Get signal number to use for suspend/resume */
3510 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3511 int sig = ::strtol(s, 0, 10);
3512 if (sig > 0 || sig < _NSIG) {
3513 SR_signum = sig;
3514 }
3515 }
3517 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3518 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3520 sigemptyset(&SR_sigset);
3521 sigaddset(&SR_sigset, SR_signum);
3523 /* Set up signal handler for suspend/resume */
3524 act.sa_flags = SA_RESTART|SA_SIGINFO;
3525 act.sa_handler = (void (*)(int)) SR_handler;
3527 // SR_signum is blocked by default.
3528 // 4528190 - We also need to block pthread restart signal (32 on all
3529 // supported Linux platforms). Note that LinuxThreads need to block
3530 // this signal for all threads to work properly. So we don't have
3531 // to use hard-coded signal number when setting up the mask.
3532 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3534 if (sigaction(SR_signum, &act, 0) == -1) {
3535 return -1;
3536 }
3538 // Save signal flag
3539 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3540 return 0;
3541 }
3543 static int SR_finalize() {
3544 return 0;
3545 }
3548 // returns true on success and false on error - really an error is fatal
3549 // but this seems the normal response to library errors
3550 static bool do_suspend(OSThread* osthread) {
3551 // mark as suspended and send signal
3552 osthread->sr.set_suspend_action(SR_SUSPEND);
3553 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3554 assert_status(status == 0, status, "pthread_kill");
3556 // check status and wait until notified of suspension
3557 if (status == 0) {
3558 for (int i = 0; !osthread->sr.is_suspended(); i++) {
3559 os::yield_all(i);
3560 }
3561 osthread->sr.set_suspend_action(SR_NONE);
3562 return true;
3563 }
3564 else {
3565 osthread->sr.set_suspend_action(SR_NONE);
3566 return false;
3567 }
3568 }
3570 static void do_resume(OSThread* osthread) {
3571 assert(osthread->sr.is_suspended(), "thread should be suspended");
3572 osthread->sr.set_suspend_action(SR_CONTINUE);
3574 int status = pthread_kill(osthread->pthread_id(), SR_signum);
3575 assert_status(status == 0, status, "pthread_kill");
3576 // check status and wait unit notified of resumption
3577 if (status == 0) {
3578 for (int i = 0; osthread->sr.is_suspended(); i++) {
3579 os::yield_all(i);
3580 }
3581 }
3582 osthread->sr.set_suspend_action(SR_NONE);
3583 }
3585 ////////////////////////////////////////////////////////////////////////////////
3586 // interrupt support
3588 void os::interrupt(Thread* thread) {
3589 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3590 "possibility of dangling Thread pointer");
3592 OSThread* osthread = thread->osthread();
3594 if (!osthread->interrupted()) {
3595 osthread->set_interrupted(true);
3596 // More than one thread can get here with the same value of osthread,
3597 // resulting in multiple notifications. We do, however, want the store
3598 // to interrupted() to be visible to other threads before we execute unpark().
3599 OrderAccess::fence();
3600 ParkEvent * const slp = thread->_SleepEvent ;
3601 if (slp != NULL) slp->unpark() ;
3602 }
3604 // For JSR166. Unpark even if interrupt status already was set
3605 if (thread->is_Java_thread())
3606 ((JavaThread*)thread)->parker()->unpark();
3608 ParkEvent * ev = thread->_ParkEvent ;
3609 if (ev != NULL) ev->unpark() ;
3611 }
3613 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
3614 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
3615 "possibility of dangling Thread pointer");
3617 OSThread* osthread = thread->osthread();
3619 bool interrupted = osthread->interrupted();
3621 if (interrupted && clear_interrupted) {
3622 osthread->set_interrupted(false);
3623 // consider thread->_SleepEvent->reset() ... optional optimization
3624 }
3626 return interrupted;
3627 }
3629 ///////////////////////////////////////////////////////////////////////////////////
3630 // signal handling (except suspend/resume)
3632 // This routine may be used by user applications as a "hook" to catch signals.
3633 // The user-defined signal handler must pass unrecognized signals to this
3634 // routine, and if it returns true (non-zero), then the signal handler must
3635 // return immediately. If the flag "abort_if_unrecognized" is true, then this
3636 // routine will never retun false (zero), but instead will execute a VM panic
3637 // routine kill the process.
3638 //
3639 // If this routine returns false, it is OK to call it again. This allows
3640 // the user-defined signal handler to perform checks either before or after
3641 // the VM performs its own checks. Naturally, the user code would be making
3642 // a serious error if it tried to handle an exception (such as a null check
3643 // or breakpoint) that the VM was generating for its own correct operation.
3644 //
3645 // This routine may recognize any of the following kinds of signals:
3646 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
3647 // It should be consulted by handlers for any of those signals.
3648 //
3649 // The caller of this routine must pass in the three arguments supplied
3650 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
3651 // field of the structure passed to sigaction(). This routine assumes that
3652 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
3653 //
3654 // Note that the VM will print warnings if it detects conflicting signal
3655 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
3656 //
3657 extern "C" JNIEXPORT int
3658 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
3659 void* ucontext, int abort_if_unrecognized);
3661 void signalHandler(int sig, siginfo_t* info, void* uc) {
3662 assert(info != NULL && uc != NULL, "it must be old kernel");
3663 JVM_handle_linux_signal(sig, info, uc, true);
3664 }
3667 // This boolean allows users to forward their own non-matching signals
3668 // to JVM_handle_linux_signal, harmlessly.
3669 bool os::Linux::signal_handlers_are_installed = false;
3671 // For signal-chaining
3672 struct sigaction os::Linux::sigact[MAXSIGNUM];
3673 unsigned int os::Linux::sigs = 0;
3674 bool os::Linux::libjsig_is_loaded = false;
3675 typedef struct sigaction *(*get_signal_t)(int);
3676 get_signal_t os::Linux::get_signal_action = NULL;
3678 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
3679 struct sigaction *actp = NULL;
3681 if (libjsig_is_loaded) {
3682 // Retrieve the old signal handler from libjsig
3683 actp = (*get_signal_action)(sig);
3684 }
3685 if (actp == NULL) {
3686 // Retrieve the preinstalled signal handler from jvm
3687 actp = get_preinstalled_handler(sig);
3688 }
3690 return actp;
3691 }
3693 static bool call_chained_handler(struct sigaction *actp, int sig,
3694 siginfo_t *siginfo, void *context) {
3695 // Call the old signal handler
3696 if (actp->sa_handler == SIG_DFL) {
3697 // It's more reasonable to let jvm treat it as an unexpected exception
3698 // instead of taking the default action.
3699 return false;
3700 } else if (actp->sa_handler != SIG_IGN) {
3701 if ((actp->sa_flags & SA_NODEFER) == 0) {
3702 // automaticlly block the signal
3703 sigaddset(&(actp->sa_mask), sig);
3704 }
3706 sa_handler_t hand;
3707 sa_sigaction_t sa;
3708 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
3709 // retrieve the chained handler
3710 if (siginfo_flag_set) {
3711 sa = actp->sa_sigaction;
3712 } else {
3713 hand = actp->sa_handler;
3714 }
3716 if ((actp->sa_flags & SA_RESETHAND) != 0) {
3717 actp->sa_handler = SIG_DFL;
3718 }
3720 // try to honor the signal mask
3721 sigset_t oset;
3722 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
3724 // call into the chained handler
3725 if (siginfo_flag_set) {
3726 (*sa)(sig, siginfo, context);
3727 } else {
3728 (*hand)(sig);
3729 }
3731 // restore the signal mask
3732 pthread_sigmask(SIG_SETMASK, &oset, 0);
3733 }
3734 // Tell jvm's signal handler the signal is taken care of.
3735 return true;
3736 }
3738 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
3739 bool chained = false;
3740 // signal-chaining
3741 if (UseSignalChaining) {
3742 struct sigaction *actp = get_chained_signal_action(sig);
3743 if (actp != NULL) {
3744 chained = call_chained_handler(actp, sig, siginfo, context);
3745 }
3746 }
3747 return chained;
3748 }
3750 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
3751 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
3752 return &sigact[sig];
3753 }
3754 return NULL;
3755 }
3757 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
3758 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3759 sigact[sig] = oldAct;
3760 sigs |= (unsigned int)1 << sig;
3761 }
3763 // for diagnostic
3764 int os::Linux::sigflags[MAXSIGNUM];
3766 int os::Linux::get_our_sigflags(int sig) {
3767 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3768 return sigflags[sig];
3769 }
3771 void os::Linux::set_our_sigflags(int sig, int flags) {
3772 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3773 sigflags[sig] = flags;
3774 }
3776 void os::Linux::set_signal_handler(int sig, bool set_installed) {
3777 // Check for overwrite.
3778 struct sigaction oldAct;
3779 sigaction(sig, (struct sigaction*)NULL, &oldAct);
3781 void* oldhand = oldAct.sa_sigaction
3782 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3783 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3784 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
3785 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
3786 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
3787 if (AllowUserSignalHandlers || !set_installed) {
3788 // Do not overwrite; user takes responsibility to forward to us.
3789 return;
3790 } else if (UseSignalChaining) {
3791 // save the old handler in jvm
3792 save_preinstalled_handler(sig, oldAct);
3793 // libjsig also interposes the sigaction() call below and saves the
3794 // old sigaction on it own.
3795 } else {
3796 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
3797 "%#lx for signal %d.", (long)oldhand, sig));
3798 }
3799 }
3801 struct sigaction sigAct;
3802 sigfillset(&(sigAct.sa_mask));
3803 sigAct.sa_handler = SIG_DFL;
3804 if (!set_installed) {
3805 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3806 } else {
3807 sigAct.sa_sigaction = signalHandler;
3808 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
3809 }
3810 // Save flags, which are set by ours
3811 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
3812 sigflags[sig] = sigAct.sa_flags;
3814 int ret = sigaction(sig, &sigAct, &oldAct);
3815 assert(ret == 0, "check");
3817 void* oldhand2 = oldAct.sa_sigaction
3818 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
3819 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
3820 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
3821 }
3823 // install signal handlers for signals that HotSpot needs to
3824 // handle in order to support Java-level exception handling.
3826 void os::Linux::install_signal_handlers() {
3827 if (!signal_handlers_are_installed) {
3828 signal_handlers_are_installed = true;
3830 // signal-chaining
3831 typedef void (*signal_setting_t)();
3832 signal_setting_t begin_signal_setting = NULL;
3833 signal_setting_t end_signal_setting = NULL;
3834 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3835 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
3836 if (begin_signal_setting != NULL) {
3837 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
3838 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
3839 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
3840 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
3841 libjsig_is_loaded = true;
3842 assert(UseSignalChaining, "should enable signal-chaining");
3843 }
3844 if (libjsig_is_loaded) {
3845 // Tell libjsig jvm is setting signal handlers
3846 (*begin_signal_setting)();
3847 }
3849 set_signal_handler(SIGSEGV, true);
3850 set_signal_handler(SIGPIPE, true);
3851 set_signal_handler(SIGBUS, true);
3852 set_signal_handler(SIGILL, true);
3853 set_signal_handler(SIGFPE, true);
3854 set_signal_handler(SIGXFSZ, true);
3856 if (libjsig_is_loaded) {
3857 // Tell libjsig jvm finishes setting signal handlers
3858 (*end_signal_setting)();
3859 }
3861 // We don't activate signal checker if libjsig is in place, we trust ourselves
3862 // and if UserSignalHandler is installed all bets are off.
3863 // Log that signal checking is off only if -verbose:jni is specified.
3864 if (CheckJNICalls) {
3865 if (libjsig_is_loaded) {
3866 if (PrintJNIResolving) {
3867 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
3868 }
3869 check_signals = false;
3870 }
3871 if (AllowUserSignalHandlers) {
3872 if (PrintJNIResolving) {
3873 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
3874 }
3875 check_signals = false;
3876 }
3877 }
3878 }
3879 }
3881 // This is the fastest way to get thread cpu time on Linux.
3882 // Returns cpu time (user+sys) for any thread, not only for current.
3883 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
3884 // It might work on 2.6.10+ with a special kernel/glibc patch.
3885 // For reference, please, see IEEE Std 1003.1-2004:
3886 // http://www.unix.org/single_unix_specification
3888 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
3889 struct timespec tp;
3890 int rc = os::Linux::clock_gettime(clockid, &tp);
3891 assert(rc == 0, "clock_gettime is expected to return 0 code");
3893 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
3894 }
3896 /////
3897 // glibc on Linux platform uses non-documented flag
3898 // to indicate, that some special sort of signal
3899 // trampoline is used.
3900 // We will never set this flag, and we should
3901 // ignore this flag in our diagnostic
3902 #ifdef SIGNIFICANT_SIGNAL_MASK
3903 #undef SIGNIFICANT_SIGNAL_MASK
3904 #endif
3905 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
3907 static const char* get_signal_handler_name(address handler,
3908 char* buf, int buflen) {
3909 int offset;
3910 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
3911 if (found) {
3912 // skip directory names
3913 const char *p1, *p2;
3914 p1 = buf;
3915 size_t len = strlen(os::file_separator());
3916 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
3917 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
3918 } else {
3919 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
3920 }
3921 return buf;
3922 }
3924 static void print_signal_handler(outputStream* st, int sig,
3925 char* buf, size_t buflen) {
3926 struct sigaction sa;
3928 sigaction(sig, NULL, &sa);
3930 // See comment for SIGNIFICANT_SIGNAL_MASK define
3931 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
3933 st->print("%s: ", os::exception_name(sig, buf, buflen));
3935 address handler = (sa.sa_flags & SA_SIGINFO)
3936 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
3937 : CAST_FROM_FN_PTR(address, sa.sa_handler);
3939 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
3940 st->print("SIG_DFL");
3941 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
3942 st->print("SIG_IGN");
3943 } else {
3944 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
3945 }
3947 st->print(", sa_mask[0]=" PTR32_FORMAT, *(uint32_t*)&sa.sa_mask);
3949 address rh = VMError::get_resetted_sighandler(sig);
3950 // May be, handler was resetted by VMError?
3951 if(rh != NULL) {
3952 handler = rh;
3953 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
3954 }
3956 st->print(", sa_flags=" PTR32_FORMAT, sa.sa_flags);
3958 // Check: is it our handler?
3959 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
3960 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
3961 // It is our signal handler
3962 // check for flags, reset system-used one!
3963 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
3964 st->print(
3965 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
3966 os::Linux::get_our_sigflags(sig));
3967 }
3968 }
3969 st->cr();
3970 }
3973 #define DO_SIGNAL_CHECK(sig) \
3974 if (!sigismember(&check_signal_done, sig)) \
3975 os::Linux::check_signal_handler(sig)
3977 // This method is a periodic task to check for misbehaving JNI applications
3978 // under CheckJNI, we can add any periodic checks here
3980 void os::run_periodic_checks() {
3982 if (check_signals == false) return;
3984 // SEGV and BUS if overridden could potentially prevent
3985 // generation of hs*.log in the event of a crash, debugging
3986 // such a case can be very challenging, so we absolutely
3987 // check the following for a good measure:
3988 DO_SIGNAL_CHECK(SIGSEGV);
3989 DO_SIGNAL_CHECK(SIGILL);
3990 DO_SIGNAL_CHECK(SIGFPE);
3991 DO_SIGNAL_CHECK(SIGBUS);
3992 DO_SIGNAL_CHECK(SIGPIPE);
3993 DO_SIGNAL_CHECK(SIGXFSZ);
3996 // ReduceSignalUsage allows the user to override these handlers
3997 // see comments at the very top and jvm_solaris.h
3998 if (!ReduceSignalUsage) {
3999 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4000 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4001 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4002 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4003 }
4005 DO_SIGNAL_CHECK(SR_signum);
4006 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4007 }
4009 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4011 static os_sigaction_t os_sigaction = NULL;
4013 void os::Linux::check_signal_handler(int sig) {
4014 char buf[O_BUFLEN];
4015 address jvmHandler = NULL;
4018 struct sigaction act;
4019 if (os_sigaction == NULL) {
4020 // only trust the default sigaction, in case it has been interposed
4021 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4022 if (os_sigaction == NULL) return;
4023 }
4025 os_sigaction(sig, (struct sigaction*)NULL, &act);
4028 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4030 address thisHandler = (act.sa_flags & SA_SIGINFO)
4031 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4032 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4035 switch(sig) {
4036 case SIGSEGV:
4037 case SIGBUS:
4038 case SIGFPE:
4039 case SIGPIPE:
4040 case SIGILL:
4041 case SIGXFSZ:
4042 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4043 break;
4045 case SHUTDOWN1_SIGNAL:
4046 case SHUTDOWN2_SIGNAL:
4047 case SHUTDOWN3_SIGNAL:
4048 case BREAK_SIGNAL:
4049 jvmHandler = (address)user_handler();
4050 break;
4052 case INTERRUPT_SIGNAL:
4053 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4054 break;
4056 default:
4057 if (sig == SR_signum) {
4058 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4059 } else {
4060 return;
4061 }
4062 break;
4063 }
4065 if (thisHandler != jvmHandler) {
4066 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4067 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4068 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4069 // No need to check this sig any longer
4070 sigaddset(&check_signal_done, sig);
4071 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4072 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4073 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4074 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4075 // No need to check this sig any longer
4076 sigaddset(&check_signal_done, sig);
4077 }
4079 // Dump all the signal
4080 if (sigismember(&check_signal_done, sig)) {
4081 print_signal_handlers(tty, buf, O_BUFLEN);
4082 }
4083 }
4085 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4087 extern bool signal_name(int signo, char* buf, size_t len);
4089 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4090 if (0 < exception_code && exception_code <= SIGRTMAX) {
4091 // signal
4092 if (!signal_name(exception_code, buf, size)) {
4093 jio_snprintf(buf, size, "SIG%d", exception_code);
4094 }
4095 return buf;
4096 } else {
4097 return NULL;
4098 }
4099 }
4101 // this is called _before_ the most of global arguments have been parsed
4102 void os::init(void) {
4103 char dummy; /* used to get a guess on initial stack address */
4104 // first_hrtime = gethrtime();
4106 // With LinuxThreads the JavaMain thread pid (primordial thread)
4107 // is different than the pid of the java launcher thread.
4108 // So, on Linux, the launcher thread pid is passed to the VM
4109 // via the sun.java.launcher.pid property.
4110 // Use this property instead of getpid() if it was correctly passed.
4111 // See bug 6351349.
4112 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4114 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4116 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4118 init_random(1234567);
4120 ThreadCritical::initialize();
4122 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4123 if (Linux::page_size() == -1) {
4124 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4125 strerror(errno)));
4126 }
4127 init_page_sizes((size_t) Linux::page_size());
4129 Linux::initialize_system_info();
4131 // main_thread points to the aboriginal thread
4132 Linux::_main_thread = pthread_self();
4134 Linux::clock_init();
4135 initial_time_count = os::elapsed_counter();
4136 pthread_mutex_init(&dl_mutex, NULL);
4137 }
4139 // To install functions for atexit system call
4140 extern "C" {
4141 static void perfMemory_exit_helper() {
4142 perfMemory_exit();
4143 }
4144 }
4146 // this is called _after_ the global arguments have been parsed
4147 jint os::init_2(void)
4148 {
4149 Linux::fast_thread_clock_init();
4151 // Allocate a single page and mark it as readable for safepoint polling
4152 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4153 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4155 os::set_polling_page( polling_page );
4157 #ifndef PRODUCT
4158 if(Verbose && PrintMiscellaneous)
4159 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4160 #endif
4162 if (!UseMembar) {
4163 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4164 guarantee( mem_serialize_page != NULL, "mmap Failed for memory serialize page");
4165 os::set_memory_serialize_page( mem_serialize_page );
4167 #ifndef PRODUCT
4168 if(Verbose && PrintMiscellaneous)
4169 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4170 #endif
4171 }
4173 os::large_page_init();
4175 // initialize suspend/resume support - must do this before signal_sets_init()
4176 if (SR_initialize() != 0) {
4177 perror("SR_initialize failed");
4178 return JNI_ERR;
4179 }
4181 Linux::signal_sets_init();
4182 Linux::install_signal_handlers();
4184 // Check minimum allowable stack size for thread creation and to initialize
4185 // the java system classes, including StackOverflowError - depends on page
4186 // size. Add a page for compiler2 recursion in main thread.
4187 // Add in 2*BytesPerWord times page size to account for VM stack during
4188 // class initialization depending on 32 or 64 bit VM.
4189 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4190 (size_t)(StackYellowPages+StackRedPages+StackShadowPages+
4191 2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::page_size());
4193 size_t threadStackSizeInBytes = ThreadStackSize * K;
4194 if (threadStackSizeInBytes != 0 &&
4195 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4196 tty->print_cr("\nThe stack size specified is too small, "
4197 "Specify at least %dk",
4198 os::Linux::min_stack_allowed/ K);
4199 return JNI_ERR;
4200 }
4202 // Make the stack size a multiple of the page size so that
4203 // the yellow/red zones can be guarded.
4204 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4205 vm_page_size()));
4207 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4209 Linux::libpthread_init();
4210 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4211 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4212 Linux::glibc_version(), Linux::libpthread_version(),
4213 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4214 }
4216 if (UseNUMA) {
4217 if (!Linux::libnuma_init()) {
4218 UseNUMA = false;
4219 } else {
4220 if ((Linux::numa_max_node() < 1)) {
4221 // There's only one node(they start from 0), disable NUMA.
4222 UseNUMA = false;
4223 }
4224 }
4225 // With SHM large pages we cannot uncommit a page, so there's not way
4226 // we can make the adaptive lgrp chunk resizing work. If the user specified
4227 // both UseNUMA and UseLargePages (or UseSHM) on the command line - warn and
4228 // disable adaptive resizing.
4229 if (UseNUMA && UseLargePages && UseSHM) {
4230 if (!FLAG_IS_DEFAULT(UseNUMA)) {
4231 if (FLAG_IS_DEFAULT(UseLargePages) && FLAG_IS_DEFAULT(UseSHM)) {
4232 UseLargePages = false;
4233 } else {
4234 warning("UseNUMA is not fully compatible with SHM large pages, disabling adaptive resizing");
4235 UseAdaptiveSizePolicy = false;
4236 UseAdaptiveNUMAChunkSizing = false;
4237 }
4238 } else {
4239 UseNUMA = false;
4240 }
4241 }
4242 if (!UseNUMA && ForceNUMA) {
4243 UseNUMA = true;
4244 }
4245 }
4247 if (MaxFDLimit) {
4248 // set the number of file descriptors to max. print out error
4249 // if getrlimit/setrlimit fails but continue regardless.
4250 struct rlimit nbr_files;
4251 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4252 if (status != 0) {
4253 if (PrintMiscellaneous && (Verbose || WizardMode))
4254 perror("os::init_2 getrlimit failed");
4255 } else {
4256 nbr_files.rlim_cur = nbr_files.rlim_max;
4257 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4258 if (status != 0) {
4259 if (PrintMiscellaneous && (Verbose || WizardMode))
4260 perror("os::init_2 setrlimit failed");
4261 }
4262 }
4263 }
4265 // Initialize lock used to serialize thread creation (see os::create_thread)
4266 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4268 // at-exit methods are called in the reverse order of their registration.
4269 // atexit functions are called on return from main or as a result of a
4270 // call to exit(3C). There can be only 32 of these functions registered
4271 // and atexit() does not set errno.
4273 if (PerfAllowAtExitRegistration) {
4274 // only register atexit functions if PerfAllowAtExitRegistration is set.
4275 // atexit functions can be delayed until process exit time, which
4276 // can be problematic for embedded VM situations. Embedded VMs should
4277 // call DestroyJavaVM() to assure that VM resources are released.
4279 // note: perfMemory_exit_helper atexit function may be removed in
4280 // the future if the appropriate cleanup code can be added to the
4281 // VM_Exit VMOperation's doit method.
4282 if (atexit(perfMemory_exit_helper) != 0) {
4283 warning("os::init2 atexit(perfMemory_exit_helper) failed");
4284 }
4285 }
4287 // initialize thread priority policy
4288 prio_init();
4290 return JNI_OK;
4291 }
4293 // this is called at the end of vm_initialization
4294 void os::init_3(void)
4295 {
4296 #ifdef JAVASE_EMBEDDED
4297 // Start the MemNotifyThread
4298 if (LowMemoryProtection) {
4299 MemNotifyThread::start();
4300 }
4301 return;
4302 #endif
4303 }
4305 // Mark the polling page as unreadable
4306 void os::make_polling_page_unreadable(void) {
4307 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4308 fatal("Could not disable polling page");
4309 };
4311 // Mark the polling page as readable
4312 void os::make_polling_page_readable(void) {
4313 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4314 fatal("Could not enable polling page");
4315 }
4316 };
4318 int os::active_processor_count() {
4319 // Linux doesn't yet have a (official) notion of processor sets,
4320 // so just return the number of online processors.
4321 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4322 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4323 return online_cpus;
4324 }
4326 void os::set_native_thread_name(const char *name) {
4327 // Not yet implemented.
4328 return;
4329 }
4331 bool os::distribute_processes(uint length, uint* distribution) {
4332 // Not yet implemented.
4333 return false;
4334 }
4336 bool os::bind_to_processor(uint processor_id) {
4337 // Not yet implemented.
4338 return false;
4339 }
4341 ///
4343 // Suspends the target using the signal mechanism and then grabs the PC before
4344 // resuming the target. Used by the flat-profiler only
4345 ExtendedPC os::get_thread_pc(Thread* thread) {
4346 // Make sure that it is called by the watcher for the VMThread
4347 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4348 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4350 ExtendedPC epc;
4352 OSThread* osthread = thread->osthread();
4353 if (do_suspend(osthread)) {
4354 if (osthread->ucontext() != NULL) {
4355 epc = os::Linux::ucontext_get_pc(osthread->ucontext());
4356 } else {
4357 // NULL context is unexpected, double-check this is the VMThread
4358 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4359 }
4360 do_resume(osthread);
4361 }
4362 // failure means pthread_kill failed for some reason - arguably this is
4363 // a fatal problem, but such problems are ignored elsewhere
4365 return epc;
4366 }
4368 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4369 {
4370 if (is_NPTL()) {
4371 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4372 } else {
4373 #ifndef IA64
4374 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4375 // word back to default 64bit precision if condvar is signaled. Java
4376 // wants 53bit precision. Save and restore current value.
4377 int fpu = get_fpu_control_word();
4378 #endif // IA64
4379 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4380 #ifndef IA64
4381 set_fpu_control_word(fpu);
4382 #endif // IA64
4383 return status;
4384 }
4385 }
4387 ////////////////////////////////////////////////////////////////////////////////
4388 // debug support
4390 static address same_page(address x, address y) {
4391 int page_bits = -os::vm_page_size();
4392 if ((intptr_t(x) & page_bits) == (intptr_t(y) & page_bits))
4393 return x;
4394 else if (x > y)
4395 return (address)(intptr_t(y) | ~page_bits) + 1;
4396 else
4397 return (address)(intptr_t(y) & page_bits);
4398 }
4400 bool os::find(address addr, outputStream* st) {
4401 Dl_info dlinfo;
4402 memset(&dlinfo, 0, sizeof(dlinfo));
4403 if (dladdr(addr, &dlinfo)) {
4404 st->print(PTR_FORMAT ": ", addr);
4405 if (dlinfo.dli_sname != NULL) {
4406 st->print("%s+%#x", dlinfo.dli_sname,
4407 addr - (intptr_t)dlinfo.dli_saddr);
4408 } else if (dlinfo.dli_fname) {
4409 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4410 } else {
4411 st->print("<absolute address>");
4412 }
4413 if (dlinfo.dli_fname) {
4414 st->print(" in %s", dlinfo.dli_fname);
4415 }
4416 if (dlinfo.dli_fbase) {
4417 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4418 }
4419 st->cr();
4421 if (Verbose) {
4422 // decode some bytes around the PC
4423 address begin = same_page(addr-40, addr);
4424 address end = same_page(addr+40, addr);
4425 address lowest = (address) dlinfo.dli_sname;
4426 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4427 if (begin < lowest) begin = lowest;
4428 Dl_info dlinfo2;
4429 if (dladdr(end, &dlinfo2) && dlinfo2.dli_saddr != dlinfo.dli_saddr
4430 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4431 end = (address) dlinfo2.dli_saddr;
4432 Disassembler::decode(begin, end, st);
4433 }
4434 return true;
4435 }
4436 return false;
4437 }
4439 ////////////////////////////////////////////////////////////////////////////////
4440 // misc
4442 // This does not do anything on Linux. This is basically a hook for being
4443 // able to use structured exception handling (thread-local exception filters)
4444 // on, e.g., Win32.
4445 void
4446 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4447 JavaCallArguments* args, Thread* thread) {
4448 f(value, method, args, thread);
4449 }
4451 void os::print_statistics() {
4452 }
4454 int os::message_box(const char* title, const char* message) {
4455 int i;
4456 fdStream err(defaultStream::error_fd());
4457 for (i = 0; i < 78; i++) err.print_raw("=");
4458 err.cr();
4459 err.print_raw_cr(title);
4460 for (i = 0; i < 78; i++) err.print_raw("-");
4461 err.cr();
4462 err.print_raw_cr(message);
4463 for (i = 0; i < 78; i++) err.print_raw("=");
4464 err.cr();
4466 char buf[16];
4467 // Prevent process from exiting upon "read error" without consuming all CPU
4468 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4470 return buf[0] == 'y' || buf[0] == 'Y';
4471 }
4473 int os::stat(const char *path, struct stat *sbuf) {
4474 char pathbuf[MAX_PATH];
4475 if (strlen(path) > MAX_PATH - 1) {
4476 errno = ENAMETOOLONG;
4477 return -1;
4478 }
4479 os::native_path(strcpy(pathbuf, path));
4480 return ::stat(pathbuf, sbuf);
4481 }
4483 bool os::check_heap(bool force) {
4484 return true;
4485 }
4487 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4488 return ::vsnprintf(buf, count, format, args);
4489 }
4491 // Is a (classpath) directory empty?
4492 bool os::dir_is_empty(const char* path) {
4493 DIR *dir = NULL;
4494 struct dirent *ptr;
4496 dir = opendir(path);
4497 if (dir == NULL) return true;
4499 /* Scan the directory */
4500 bool result = true;
4501 char buf[sizeof(struct dirent) + MAX_PATH];
4502 while (result && (ptr = ::readdir(dir)) != NULL) {
4503 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
4504 result = false;
4505 }
4506 }
4507 closedir(dir);
4508 return result;
4509 }
4511 // This code originates from JDK's sysOpen and open64_w
4512 // from src/solaris/hpi/src/system_md.c
4514 #ifndef O_DELETE
4515 #define O_DELETE 0x10000
4516 #endif
4518 // Open a file. Unlink the file immediately after open returns
4519 // if the specified oflag has the O_DELETE flag set.
4520 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
4522 int os::open(const char *path, int oflag, int mode) {
4524 if (strlen(path) > MAX_PATH - 1) {
4525 errno = ENAMETOOLONG;
4526 return -1;
4527 }
4528 int fd;
4529 int o_delete = (oflag & O_DELETE);
4530 oflag = oflag & ~O_DELETE;
4532 fd = ::open64(path, oflag, mode);
4533 if (fd == -1) return -1;
4535 //If the open succeeded, the file might still be a directory
4536 {
4537 struct stat64 buf64;
4538 int ret = ::fstat64(fd, &buf64);
4539 int st_mode = buf64.st_mode;
4541 if (ret != -1) {
4542 if ((st_mode & S_IFMT) == S_IFDIR) {
4543 errno = EISDIR;
4544 ::close(fd);
4545 return -1;
4546 }
4547 } else {
4548 ::close(fd);
4549 return -1;
4550 }
4551 }
4553 /*
4554 * All file descriptors that are opened in the JVM and not
4555 * specifically destined for a subprocess should have the
4556 * close-on-exec flag set. If we don't set it, then careless 3rd
4557 * party native code might fork and exec without closing all
4558 * appropriate file descriptors (e.g. as we do in closeDescriptors in
4559 * UNIXProcess.c), and this in turn might:
4560 *
4561 * - cause end-of-file to fail to be detected on some file
4562 * descriptors, resulting in mysterious hangs, or
4563 *
4564 * - might cause an fopen in the subprocess to fail on a system
4565 * suffering from bug 1085341.
4566 *
4567 * (Yes, the default setting of the close-on-exec flag is a Unix
4568 * design flaw)
4569 *
4570 * See:
4571 * 1085341: 32-bit stdio routines should support file descriptors >255
4572 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
4573 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
4574 */
4575 #ifdef FD_CLOEXEC
4576 {
4577 int flags = ::fcntl(fd, F_GETFD);
4578 if (flags != -1)
4579 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
4580 }
4581 #endif
4583 if (o_delete != 0) {
4584 ::unlink(path);
4585 }
4586 return fd;
4587 }
4590 // create binary file, rewriting existing file if required
4591 int os::create_binary_file(const char* path, bool rewrite_existing) {
4592 int oflags = O_WRONLY | O_CREAT;
4593 if (!rewrite_existing) {
4594 oflags |= O_EXCL;
4595 }
4596 return ::open64(path, oflags, S_IREAD | S_IWRITE);
4597 }
4599 // return current position of file pointer
4600 jlong os::current_file_offset(int fd) {
4601 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
4602 }
4604 // move file pointer to the specified offset
4605 jlong os::seek_to_file_offset(int fd, jlong offset) {
4606 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
4607 }
4609 // This code originates from JDK's sysAvailable
4610 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
4612 int os::available(int fd, jlong *bytes) {
4613 jlong cur, end;
4614 int mode;
4615 struct stat64 buf64;
4617 if (::fstat64(fd, &buf64) >= 0) {
4618 mode = buf64.st_mode;
4619 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
4620 /*
4621 * XXX: is the following call interruptible? If so, this might
4622 * need to go through the INTERRUPT_IO() wrapper as for other
4623 * blocking, interruptible calls in this file.
4624 */
4625 int n;
4626 if (::ioctl(fd, FIONREAD, &n) >= 0) {
4627 *bytes = n;
4628 return 1;
4629 }
4630 }
4631 }
4632 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
4633 return 0;
4634 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
4635 return 0;
4636 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
4637 return 0;
4638 }
4639 *bytes = end - cur;
4640 return 1;
4641 }
4643 int os::socket_available(int fd, jint *pbytes) {
4644 // Linux doc says EINTR not returned, unlike Solaris
4645 int ret = ::ioctl(fd, FIONREAD, pbytes);
4647 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
4648 // is expected to return 0 on failure and 1 on success to the jdk.
4649 return (ret < 0) ? 0 : 1;
4650 }
4652 // Map a block of memory.
4653 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
4654 char *addr, size_t bytes, bool read_only,
4655 bool allow_exec) {
4656 int prot;
4657 int flags = MAP_PRIVATE;
4659 if (read_only) {
4660 prot = PROT_READ;
4661 } else {
4662 prot = PROT_READ | PROT_WRITE;
4663 }
4665 if (allow_exec) {
4666 prot |= PROT_EXEC;
4667 }
4669 if (addr != NULL) {
4670 flags |= MAP_FIXED;
4671 }
4673 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
4674 fd, file_offset);
4675 if (mapped_address == MAP_FAILED) {
4676 return NULL;
4677 }
4678 return mapped_address;
4679 }
4682 // Remap a block of memory.
4683 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
4684 char *addr, size_t bytes, bool read_only,
4685 bool allow_exec) {
4686 // same as map_memory() on this OS
4687 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
4688 allow_exec);
4689 }
4692 // Unmap a block of memory.
4693 bool os::pd_unmap_memory(char* addr, size_t bytes) {
4694 return munmap(addr, bytes) == 0;
4695 }
4697 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
4699 static clockid_t thread_cpu_clockid(Thread* thread) {
4700 pthread_t tid = thread->osthread()->pthread_id();
4701 clockid_t clockid;
4703 // Get thread clockid
4704 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
4705 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
4706 return clockid;
4707 }
4709 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
4710 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
4711 // of a thread.
4712 //
4713 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
4714 // the fast estimate available on the platform.
4716 jlong os::current_thread_cpu_time() {
4717 if (os::Linux::supports_fast_thread_cpu_time()) {
4718 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4719 } else {
4720 // return user + sys since the cost is the same
4721 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
4722 }
4723 }
4725 jlong os::thread_cpu_time(Thread* thread) {
4726 // consistent with what current_thread_cpu_time() returns
4727 if (os::Linux::supports_fast_thread_cpu_time()) {
4728 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4729 } else {
4730 return slow_thread_cpu_time(thread, true /* user + sys */);
4731 }
4732 }
4734 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
4735 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4736 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
4737 } else {
4738 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
4739 }
4740 }
4742 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4743 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
4744 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
4745 } else {
4746 return slow_thread_cpu_time(thread, user_sys_cpu_time);
4747 }
4748 }
4750 //
4751 // -1 on error.
4752 //
4754 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
4755 static bool proc_pid_cpu_avail = true;
4756 static bool proc_task_unchecked = true;
4757 static const char *proc_stat_path = "/proc/%d/stat";
4758 pid_t tid = thread->osthread()->thread_id();
4759 int i;
4760 char *s;
4761 char stat[2048];
4762 int statlen;
4763 char proc_name[64];
4764 int count;
4765 long sys_time, user_time;
4766 char string[64];
4767 char cdummy;
4768 int idummy;
4769 long ldummy;
4770 FILE *fp;
4772 // We first try accessing /proc/<pid>/cpu since this is faster to
4773 // process. If this file is not present (linux kernels 2.5 and above)
4774 // then we open /proc/<pid>/stat.
4775 if ( proc_pid_cpu_avail ) {
4776 sprintf(proc_name, "/proc/%d/cpu", tid);
4777 fp = fopen(proc_name, "r");
4778 if ( fp != NULL ) {
4779 count = fscanf( fp, "%s %lu %lu\n", string, &user_time, &sys_time);
4780 fclose(fp);
4781 if ( count != 3 ) return -1;
4783 if (user_sys_cpu_time) {
4784 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4785 } else {
4786 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4787 }
4788 }
4789 else proc_pid_cpu_avail = false;
4790 }
4792 // The /proc/<tid>/stat aggregates per-process usage on
4793 // new Linux kernels 2.6+ where NPTL is supported.
4794 // The /proc/self/task/<tid>/stat still has the per-thread usage.
4795 // See bug 6328462.
4796 // There can be no directory /proc/self/task on kernels 2.4 with NPTL
4797 // and possibly in some other cases, so we check its availability.
4798 if (proc_task_unchecked && os::Linux::is_NPTL()) {
4799 // This is executed only once
4800 proc_task_unchecked = false;
4801 fp = fopen("/proc/self/task", "r");
4802 if (fp != NULL) {
4803 proc_stat_path = "/proc/self/task/%d/stat";
4804 fclose(fp);
4805 }
4806 }
4808 sprintf(proc_name, proc_stat_path, tid);
4809 fp = fopen(proc_name, "r");
4810 if ( fp == NULL ) return -1;
4811 statlen = fread(stat, 1, 2047, fp);
4812 stat[statlen] = '\0';
4813 fclose(fp);
4815 // Skip pid and the command string. Note that we could be dealing with
4816 // weird command names, e.g. user could decide to rename java launcher
4817 // to "java 1.4.2 :)", then the stat file would look like
4818 // 1234 (java 1.4.2 :)) R ... ...
4819 // We don't really need to know the command string, just find the last
4820 // occurrence of ")" and then start parsing from there. See bug 4726580.
4821 s = strrchr(stat, ')');
4822 i = 0;
4823 if (s == NULL ) return -1;
4825 // Skip blank chars
4826 do s++; while (isspace(*s));
4828 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
4829 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
4830 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
4831 &user_time, &sys_time);
4832 if ( count != 13 ) return -1;
4833 if (user_sys_cpu_time) {
4834 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
4835 } else {
4836 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
4837 }
4838 }
4840 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4841 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4842 info_ptr->may_skip_backward = false; // elapsed time not wall time
4843 info_ptr->may_skip_forward = false; // elapsed time not wall time
4844 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4845 }
4847 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
4848 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
4849 info_ptr->may_skip_backward = false; // elapsed time not wall time
4850 info_ptr->may_skip_forward = false; // elapsed time not wall time
4851 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
4852 }
4854 bool os::is_thread_cpu_time_supported() {
4855 return true;
4856 }
4858 // System loadavg support. Returns -1 if load average cannot be obtained.
4859 // Linux doesn't yet have a (official) notion of processor sets,
4860 // so just return the system wide load average.
4861 int os::loadavg(double loadavg[], int nelem) {
4862 return ::getloadavg(loadavg, nelem);
4863 }
4865 void os::pause() {
4866 char filename[MAX_PATH];
4867 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
4868 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
4869 } else {
4870 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
4871 }
4873 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
4874 if (fd != -1) {
4875 struct stat buf;
4876 ::close(fd);
4877 while (::stat(filename, &buf) == 0) {
4878 (void)::poll(NULL, 0, 100);
4879 }
4880 } else {
4881 jio_fprintf(stderr,
4882 "Could not open pause file '%s', continuing immediately.\n", filename);
4883 }
4884 }
4887 // Refer to the comments in os_solaris.cpp park-unpark.
4888 //
4889 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
4890 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
4891 // For specifics regarding the bug see GLIBC BUGID 261237 :
4892 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
4893 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
4894 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
4895 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
4896 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
4897 // and monitorenter when we're using 1-0 locking. All those operations may result in
4898 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
4899 // of libpthread avoids the problem, but isn't practical.
4900 //
4901 // Possible remedies:
4902 //
4903 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
4904 // This is palliative and probabilistic, however. If the thread is preempted
4905 // between the call to compute_abstime() and pthread_cond_timedwait(), more
4906 // than the minimum period may have passed, and the abstime may be stale (in the
4907 // past) resultin in a hang. Using this technique reduces the odds of a hang
4908 // but the JVM is still vulnerable, particularly on heavily loaded systems.
4909 //
4910 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
4911 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
4912 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
4913 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
4914 // thread.
4915 //
4916 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
4917 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
4918 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
4919 // This also works well. In fact it avoids kernel-level scalability impediments
4920 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
4921 // timers in a graceful fashion.
4922 //
4923 // 4. When the abstime value is in the past it appears that control returns
4924 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
4925 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
4926 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
4927 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
4928 // It may be possible to avoid reinitialization by checking the return
4929 // value from pthread_cond_timedwait(). In addition to reinitializing the
4930 // condvar we must establish the invariant that cond_signal() is only called
4931 // within critical sections protected by the adjunct mutex. This prevents
4932 // cond_signal() from "seeing" a condvar that's in the midst of being
4933 // reinitialized or that is corrupt. Sadly, this invariant obviates the
4934 // desirable signal-after-unlock optimization that avoids futile context switching.
4935 //
4936 // I'm also concerned that some versions of NTPL might allocate an auxilliary
4937 // structure when a condvar is used or initialized. cond_destroy() would
4938 // release the helper structure. Our reinitialize-after-timedwait fix
4939 // put excessive stress on malloc/free and locks protecting the c-heap.
4940 //
4941 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
4942 // It may be possible to refine (4) by checking the kernel and NTPL verisons
4943 // and only enabling the work-around for vulnerable environments.
4945 // utility to compute the abstime argument to timedwait:
4946 // millis is the relative timeout time
4947 // abstime will be the absolute timeout time
4948 // TODO: replace compute_abstime() with unpackTime()
4950 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
4951 if (millis < 0) millis = 0;
4952 struct timeval now;
4953 int status = gettimeofday(&now, NULL);
4954 assert(status == 0, "gettimeofday");
4955 jlong seconds = millis / 1000;
4956 millis %= 1000;
4957 if (seconds > 50000000) { // see man cond_timedwait(3T)
4958 seconds = 50000000;
4959 }
4960 abstime->tv_sec = now.tv_sec + seconds;
4961 long usec = now.tv_usec + millis * 1000;
4962 if (usec >= 1000000) {
4963 abstime->tv_sec += 1;
4964 usec -= 1000000;
4965 }
4966 abstime->tv_nsec = usec * 1000;
4967 return abstime;
4968 }
4971 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
4972 // Conceptually TryPark() should be equivalent to park(0).
4974 int os::PlatformEvent::TryPark() {
4975 for (;;) {
4976 const int v = _Event ;
4977 guarantee ((v == 0) || (v == 1), "invariant") ;
4978 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
4979 }
4980 }
4982 void os::PlatformEvent::park() { // AKA "down()"
4983 // Invariant: Only the thread associated with the Event/PlatformEvent
4984 // may call park().
4985 // TODO: assert that _Assoc != NULL or _Assoc == Self
4986 int v ;
4987 for (;;) {
4988 v = _Event ;
4989 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
4990 }
4991 guarantee (v >= 0, "invariant") ;
4992 if (v == 0) {
4993 // Do this the hard way by blocking ...
4994 int status = pthread_mutex_lock(_mutex);
4995 assert_status(status == 0, status, "mutex_lock");
4996 guarantee (_nParked == 0, "invariant") ;
4997 ++ _nParked ;
4998 while (_Event < 0) {
4999 status = pthread_cond_wait(_cond, _mutex);
5000 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5001 // Treat this the same as if the wait was interrupted
5002 if (status == ETIME) { status = EINTR; }
5003 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5004 }
5005 -- _nParked ;
5007 // In theory we could move the ST of 0 into _Event past the unlock(),
5008 // but then we'd need a MEMBAR after the ST.
5009 _Event = 0 ;
5010 status = pthread_mutex_unlock(_mutex);
5011 assert_status(status == 0, status, "mutex_unlock");
5012 }
5013 guarantee (_Event >= 0, "invariant") ;
5014 }
5016 int os::PlatformEvent::park(jlong millis) {
5017 guarantee (_nParked == 0, "invariant") ;
5019 int v ;
5020 for (;;) {
5021 v = _Event ;
5022 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5023 }
5024 guarantee (v >= 0, "invariant") ;
5025 if (v != 0) return OS_OK ;
5027 // We do this the hard way, by blocking the thread.
5028 // Consider enforcing a minimum timeout value.
5029 struct timespec abst;
5030 compute_abstime(&abst, millis);
5032 int ret = OS_TIMEOUT;
5033 int status = pthread_mutex_lock(_mutex);
5034 assert_status(status == 0, status, "mutex_lock");
5035 guarantee (_nParked == 0, "invariant") ;
5036 ++_nParked ;
5038 // Object.wait(timo) will return because of
5039 // (a) notification
5040 // (b) timeout
5041 // (c) thread.interrupt
5042 //
5043 // Thread.interrupt and object.notify{All} both call Event::set.
5044 // That is, we treat thread.interrupt as a special case of notification.
5045 // The underlying Solaris implementation, cond_timedwait, admits
5046 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5047 // JVM from making those visible to Java code. As such, we must
5048 // filter out spurious wakeups. We assume all ETIME returns are valid.
5049 //
5050 // TODO: properly differentiate simultaneous notify+interrupt.
5051 // In that case, we should propagate the notify to another waiter.
5053 while (_Event < 0) {
5054 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5055 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5056 pthread_cond_destroy (_cond);
5057 pthread_cond_init (_cond, NULL) ;
5058 }
5059 assert_status(status == 0 || status == EINTR ||
5060 status == ETIME || status == ETIMEDOUT,
5061 status, "cond_timedwait");
5062 if (!FilterSpuriousWakeups) break ; // previous semantics
5063 if (status == ETIME || status == ETIMEDOUT) break ;
5064 // We consume and ignore EINTR and spurious wakeups.
5065 }
5066 --_nParked ;
5067 if (_Event >= 0) {
5068 ret = OS_OK;
5069 }
5070 _Event = 0 ;
5071 status = pthread_mutex_unlock(_mutex);
5072 assert_status(status == 0, status, "mutex_unlock");
5073 assert (_nParked == 0, "invariant") ;
5074 return ret;
5075 }
5077 void os::PlatformEvent::unpark() {
5078 int v, AnyWaiters ;
5079 for (;;) {
5080 v = _Event ;
5081 if (v > 0) {
5082 // The LD of _Event could have reordered or be satisfied
5083 // by a read-aside from this processor's write buffer.
5084 // To avoid problems execute a barrier and then
5085 // ratify the value.
5086 OrderAccess::fence() ;
5087 if (_Event == v) return ;
5088 continue ;
5089 }
5090 if (Atomic::cmpxchg (v+1, &_Event, v) == v) break ;
5091 }
5092 if (v < 0) {
5093 // Wait for the thread associated with the event to vacate
5094 int status = pthread_mutex_lock(_mutex);
5095 assert_status(status == 0, status, "mutex_lock");
5096 AnyWaiters = _nParked ;
5097 assert (AnyWaiters == 0 || AnyWaiters == 1, "invariant") ;
5098 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5099 AnyWaiters = 0 ;
5100 pthread_cond_signal (_cond);
5101 }
5102 status = pthread_mutex_unlock(_mutex);
5103 assert_status(status == 0, status, "mutex_unlock");
5104 if (AnyWaiters != 0) {
5105 status = pthread_cond_signal(_cond);
5106 assert_status(status == 0, status, "cond_signal");
5107 }
5108 }
5110 // Note that we signal() _after dropping the lock for "immortal" Events.
5111 // This is safe and avoids a common class of futile wakeups. In rare
5112 // circumstances this can cause a thread to return prematurely from
5113 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5114 // simply re-test the condition and re-park itself.
5115 }
5118 // JSR166
5119 // -------------------------------------------------------
5121 /*
5122 * The solaris and linux implementations of park/unpark are fairly
5123 * conservative for now, but can be improved. They currently use a
5124 * mutex/condvar pair, plus a a count.
5125 * Park decrements count if > 0, else does a condvar wait. Unpark
5126 * sets count to 1 and signals condvar. Only one thread ever waits
5127 * on the condvar. Contention seen when trying to park implies that someone
5128 * is unparking you, so don't wait. And spurious returns are fine, so there
5129 * is no need to track notifications.
5130 */
5132 #define MAX_SECS 100000000
5133 /*
5134 * This code is common to linux and solaris and will be moved to a
5135 * common place in dolphin.
5136 *
5137 * The passed in time value is either a relative time in nanoseconds
5138 * or an absolute time in milliseconds. Either way it has to be unpacked
5139 * into suitable seconds and nanoseconds components and stored in the
5140 * given timespec structure.
5141 * Given time is a 64-bit value and the time_t used in the timespec is only
5142 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5143 * overflow if times way in the future are given. Further on Solaris versions
5144 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5145 * number of seconds, in abstime, is less than current_time + 100,000,000.
5146 * As it will be 28 years before "now + 100000000" will overflow we can
5147 * ignore overflow and just impose a hard-limit on seconds using the value
5148 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5149 * years from "now".
5150 */
5152 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5153 assert (time > 0, "convertTime");
5155 struct timeval now;
5156 int status = gettimeofday(&now, NULL);
5157 assert(status == 0, "gettimeofday");
5159 time_t max_secs = now.tv_sec + MAX_SECS;
5161 if (isAbsolute) {
5162 jlong secs = time / 1000;
5163 if (secs > max_secs) {
5164 absTime->tv_sec = max_secs;
5165 }
5166 else {
5167 absTime->tv_sec = secs;
5168 }
5169 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5170 }
5171 else {
5172 jlong secs = time / NANOSECS_PER_SEC;
5173 if (secs >= MAX_SECS) {
5174 absTime->tv_sec = max_secs;
5175 absTime->tv_nsec = 0;
5176 }
5177 else {
5178 absTime->tv_sec = now.tv_sec + secs;
5179 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5180 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5181 absTime->tv_nsec -= NANOSECS_PER_SEC;
5182 ++absTime->tv_sec; // note: this must be <= max_secs
5183 }
5184 }
5185 }
5186 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5187 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5188 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5189 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5190 }
5192 void Parker::park(bool isAbsolute, jlong time) {
5193 // Optional fast-path check:
5194 // Return immediately if a permit is available.
5195 if (_counter > 0) {
5196 _counter = 0 ;
5197 OrderAccess::fence();
5198 return ;
5199 }
5201 Thread* thread = Thread::current();
5202 assert(thread->is_Java_thread(), "Must be JavaThread");
5203 JavaThread *jt = (JavaThread *)thread;
5205 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5206 // Check interrupt before trying to wait
5207 if (Thread::is_interrupted(thread, false)) {
5208 return;
5209 }
5211 // Next, demultiplex/decode time arguments
5212 timespec absTime;
5213 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5214 return;
5215 }
5216 if (time > 0) {
5217 unpackTime(&absTime, isAbsolute, time);
5218 }
5221 // Enter safepoint region
5222 // Beware of deadlocks such as 6317397.
5223 // The per-thread Parker:: mutex is a classic leaf-lock.
5224 // In particular a thread must never block on the Threads_lock while
5225 // holding the Parker:: mutex. If safepoints are pending both the
5226 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5227 ThreadBlockInVM tbivm(jt);
5229 // Don't wait if cannot get lock since interference arises from
5230 // unblocking. Also. check interrupt before trying wait
5231 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5232 return;
5233 }
5235 int status ;
5236 if (_counter > 0) { // no wait needed
5237 _counter = 0;
5238 status = pthread_mutex_unlock(_mutex);
5239 assert (status == 0, "invariant") ;
5240 OrderAccess::fence();
5241 return;
5242 }
5244 #ifdef ASSERT
5245 // Don't catch signals while blocked; let the running threads have the signals.
5246 // (This allows a debugger to break into the running thread.)
5247 sigset_t oldsigs;
5248 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5249 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5250 #endif
5252 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5253 jt->set_suspend_equivalent();
5254 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5256 if (time == 0) {
5257 status = pthread_cond_wait (_cond, _mutex) ;
5258 } else {
5259 status = os::Linux::safe_cond_timedwait (_cond, _mutex, &absTime) ;
5260 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5261 pthread_cond_destroy (_cond) ;
5262 pthread_cond_init (_cond, NULL);
5263 }
5264 }
5265 assert_status(status == 0 || status == EINTR ||
5266 status == ETIME || status == ETIMEDOUT,
5267 status, "cond_timedwait");
5269 #ifdef ASSERT
5270 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5271 #endif
5273 _counter = 0 ;
5274 status = pthread_mutex_unlock(_mutex) ;
5275 assert_status(status == 0, status, "invariant") ;
5276 // If externally suspended while waiting, re-suspend
5277 if (jt->handle_special_suspend_equivalent_condition()) {
5278 jt->java_suspend_self();
5279 }
5281 OrderAccess::fence();
5282 }
5284 void Parker::unpark() {
5285 int s, status ;
5286 status = pthread_mutex_lock(_mutex);
5287 assert (status == 0, "invariant") ;
5288 s = _counter;
5289 _counter = 1;
5290 if (s < 1) {
5291 if (WorkAroundNPTLTimedWaitHang) {
5292 status = pthread_cond_signal (_cond) ;
5293 assert (status == 0, "invariant") ;
5294 status = pthread_mutex_unlock(_mutex);
5295 assert (status == 0, "invariant") ;
5296 } else {
5297 status = pthread_mutex_unlock(_mutex);
5298 assert (status == 0, "invariant") ;
5299 status = pthread_cond_signal (_cond) ;
5300 assert (status == 0, "invariant") ;
5301 }
5302 } else {
5303 pthread_mutex_unlock(_mutex);
5304 assert (status == 0, "invariant") ;
5305 }
5306 }
5309 extern char** environ;
5311 #ifndef __NR_fork
5312 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5313 #endif
5315 #ifndef __NR_execve
5316 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5317 #endif
5319 // Run the specified command in a separate process. Return its exit value,
5320 // or -1 on failure (e.g. can't fork a new process).
5321 // Unlike system(), this function can be called from signal handler. It
5322 // doesn't block SIGINT et al.
5323 int os::fork_and_exec(char* cmd) {
5324 const char * argv[4] = {"sh", "-c", cmd, NULL};
5326 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5327 // pthread_atfork handlers and reset pthread library. All we need is a
5328 // separate process to execve. Make a direct syscall to fork process.
5329 // On IA64 there's no fork syscall, we have to use fork() and hope for
5330 // the best...
5331 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5332 IA64_ONLY(fork();)
5334 if (pid < 0) {
5335 // fork failed
5336 return -1;
5338 } else if (pid == 0) {
5339 // child process
5341 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5342 // first to kill every thread on the thread list. Because this list is
5343 // not reset by fork() (see notes above), execve() will instead kill
5344 // every thread in the parent process. We know this is the only thread
5345 // in the new process, so make a system call directly.
5346 // IA64 should use normal execve() from glibc to match the glibc fork()
5347 // above.
5348 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5349 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5351 // execve failed
5352 _exit(-1);
5354 } else {
5355 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5356 // care about the actual exit code, for now.
5358 int status;
5360 // Wait for the child process to exit. This returns immediately if
5361 // the child has already exited. */
5362 while (waitpid(pid, &status, 0) < 0) {
5363 switch (errno) {
5364 case ECHILD: return 0;
5365 case EINTR: break;
5366 default: return -1;
5367 }
5368 }
5370 if (WIFEXITED(status)) {
5371 // The child exited normally; get its exit code.
5372 return WEXITSTATUS(status);
5373 } else if (WIFSIGNALED(status)) {
5374 // The child exited because of a signal
5375 // The best value to return is 0x80 + signal number,
5376 // because that is what all Unix shells do, and because
5377 // it allows callers to distinguish between process exit and
5378 // process death by signal.
5379 return 0x80 + WTERMSIG(status);
5380 } else {
5381 // Unknown exit code; pass it through
5382 return status;
5383 }
5384 }
5385 }
5387 // is_headless_jre()
5388 //
5389 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5390 // in order to report if we are running in a headless jre
5391 //
5392 // Since JDK8 xawt/libmawt.so was moved into the same directory
5393 // as libawt.so, and renamed libawt_xawt.so
5394 //
5395 bool os::is_headless_jre() {
5396 struct stat statbuf;
5397 char buf[MAXPATHLEN];
5398 char libmawtpath[MAXPATHLEN];
5399 const char *xawtstr = "/xawt/libmawt.so";
5400 const char *new_xawtstr = "/libawt_xawt.so";
5401 char *p;
5403 // Get path to libjvm.so
5404 os::jvm_path(buf, sizeof(buf));
5406 // Get rid of libjvm.so
5407 p = strrchr(buf, '/');
5408 if (p == NULL) return false;
5409 else *p = '\0';
5411 // Get rid of client or server
5412 p = strrchr(buf, '/');
5413 if (p == NULL) return false;
5414 else *p = '\0';
5416 // check xawt/libmawt.so
5417 strcpy(libmawtpath, buf);
5418 strcat(libmawtpath, xawtstr);
5419 if (::stat(libmawtpath, &statbuf) == 0) return false;
5421 // check libawt_xawt.so
5422 strcpy(libmawtpath, buf);
5423 strcat(libmawtpath, new_xawtstr);
5424 if (::stat(libmawtpath, &statbuf) == 0) return false;
5426 return true;
5427 }
5429 // Get the default path to the core file
5430 // Returns the length of the string
5431 int os::get_core_path(char* buffer, size_t bufferSize) {
5432 const char* p = get_current_directory(buffer, bufferSize);
5434 if (p == NULL) {
5435 assert(p != NULL, "failed to get current directory");
5436 return 0;
5437 }
5439 return strlen(buffer);
5440 }
5442 #ifdef JAVASE_EMBEDDED
5443 //
5444 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5445 //
5446 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5448 // ctor
5449 //
5450 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5451 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5452 _fd = fd;
5454 if (os::create_thread(this, os::os_thread)) {
5455 _memnotify_thread = this;
5456 os::set_priority(this, NearMaxPriority);
5457 os::start_thread(this);
5458 }
5459 }
5461 // Where all the work gets done
5462 //
5463 void MemNotifyThread::run() {
5464 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5466 // Set up the select arguments
5467 fd_set rfds;
5468 if (_fd != -1) {
5469 FD_ZERO(&rfds);
5470 FD_SET(_fd, &rfds);
5471 }
5473 // Now wait for the mem_notify device to wake up
5474 while (1) {
5475 // Wait for the mem_notify device to signal us..
5476 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
5477 if (rc == -1) {
5478 perror("select!\n");
5479 break;
5480 } else if (rc) {
5481 //ssize_t free_before = os::available_memory();
5482 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
5484 // The kernel is telling us there is not much memory left...
5485 // try to do something about that
5487 // If we are not already in a GC, try one.
5488 if (!Universe::heap()->is_gc_active()) {
5489 Universe::heap()->collect(GCCause::_allocation_failure);
5491 //ssize_t free_after = os::available_memory();
5492 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
5493 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
5494 }
5495 // We might want to do something like the following if we find the GC's are not helping...
5496 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
5497 }
5498 }
5499 }
5501 //
5502 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
5503 //
5504 void MemNotifyThread::start() {
5505 int fd;
5506 fd = open ("/dev/mem_notify", O_RDONLY, 0);
5507 if (fd < 0) {
5508 return;
5509 }
5511 if (memnotify_thread() == NULL) {
5512 new MemNotifyThread(fd);
5513 }
5514 }
5515 #endif // JAVASE_EMBEDDED