Sat, 09 Nov 2019 20:29:45 +0800
Merge
1 /*
2 * Copyright (c) 1999, 2019, 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 "osContainer_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/interfaceSupport.hpp"
48 #include "runtime/init.hpp"
49 #include "runtime/java.hpp"
50 #include "runtime/javaCalls.hpp"
51 #include "runtime/mutexLocker.hpp"
52 #include "runtime/objectMonitor.hpp"
53 #include "runtime/orderAccess.inline.hpp"
54 #include "runtime/osThread.hpp"
55 #include "runtime/perfMemory.hpp"
56 #include "runtime/sharedRuntime.hpp"
57 #include "runtime/statSampler.hpp"
58 #include "runtime/stubRoutines.hpp"
59 #include "runtime/thread.inline.hpp"
60 #include "runtime/threadCritical.hpp"
61 #include "runtime/timer.hpp"
62 #include "services/attachListener.hpp"
63 #include "services/memTracker.hpp"
64 #include "services/runtimeService.hpp"
65 #include "utilities/decoder.hpp"
66 #include "utilities/defaultStream.hpp"
67 #include "utilities/events.hpp"
68 #include "utilities/elfFile.hpp"
69 #include "utilities/growableArray.hpp"
70 #include "utilities/vmError.hpp"
72 // put OS-includes here
73 # include <sys/types.h>
74 # include <sys/mman.h>
75 # include <sys/stat.h>
76 # include <sys/select.h>
77 # include <pthread.h>
78 # include <signal.h>
79 # include <errno.h>
80 # include <dlfcn.h>
81 # include <stdio.h>
82 # include <unistd.h>
83 # include <sys/resource.h>
84 # include <pthread.h>
85 # include <sys/stat.h>
86 # include <sys/time.h>
87 # include <sys/times.h>
88 # include <sys/utsname.h>
89 # include <sys/socket.h>
90 # include <sys/wait.h>
91 # include <pwd.h>
92 # include <poll.h>
93 # include <semaphore.h>
94 # include <fcntl.h>
95 # include <string.h>
96 # include <syscall.h>
97 # include <sys/sysinfo.h>
98 # include <gnu/libc-version.h>
99 # include <sys/ipc.h>
100 # include <sys/shm.h>
101 # include <link.h>
102 # include <stdint.h>
103 # include <inttypes.h>
104 # include <sys/ioctl.h>
106 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
108 #ifndef _GNU_SOURCE
109 #define _GNU_SOURCE
110 #include <sched.h>
111 #undef _GNU_SOURCE
112 #else
113 #include <sched.h>
114 #endif
116 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
117 // getrusage() is prepared to handle the associated failure.
118 #ifndef RUSAGE_THREAD
119 #define RUSAGE_THREAD (1) /* only the calling thread */
120 #endif
122 #define MAX_PATH (2 * K)
124 #define MAX_SECS 100000000
126 // for timer info max values which include all bits
127 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
129 #define LARGEPAGES_BIT (1 << 6)
130 ////////////////////////////////////////////////////////////////////////////////
131 // global variables
132 julong os::Linux::_physical_memory = 0;
134 address os::Linux::_initial_thread_stack_bottom = NULL;
135 uintptr_t os::Linux::_initial_thread_stack_size = 0;
137 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
138 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
139 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
140 Mutex* os::Linux::_createThread_lock = NULL;
141 pthread_t os::Linux::_main_thread;
142 int os::Linux::_page_size = -1;
143 const int os::Linux::_vm_default_page_size = (8 * K);
144 bool os::Linux::_is_floating_stack = false;
145 bool os::Linux::_is_NPTL = false;
146 bool os::Linux::_supports_fast_thread_cpu_time = false;
147 const char * os::Linux::_glibc_version = NULL;
148 const char * os::Linux::_libpthread_version = NULL;
149 pthread_condattr_t os::Linux::_condattr[1];
151 static jlong initial_time_count=0;
153 static int clock_tics_per_sec = 100;
155 // For diagnostics to print a message once. see run_periodic_checks
156 static sigset_t check_signal_done;
157 static bool check_signals = true;
159 static pid_t _initial_pid = 0;
161 /* Signal number used to suspend/resume a thread */
163 /* do not use any signal number less than SIGSEGV, see 4355769 */
164 static int SR_signum = SIGUSR2;
165 sigset_t SR_sigset;
167 /* Used to protect dlsym() calls */
168 static pthread_mutex_t dl_mutex;
170 // Declarations
171 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
173 // utility functions
175 static int SR_initialize();
177 julong os::available_memory() {
178 return Linux::available_memory();
179 }
181 julong os::Linux::available_memory() {
182 // values in struct sysinfo are "unsigned long"
183 struct sysinfo si;
184 julong avail_mem;
186 if (OSContainer::is_containerized()) {
187 jlong mem_limit, mem_usage;
188 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
189 if (PrintContainerInfo) {
190 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
191 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
192 }
193 }
195 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
196 if (PrintContainerInfo) {
197 tty->print_cr("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
198 }
199 }
201 if (mem_limit > 0 && mem_usage > 0 ) {
202 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
203 if (PrintContainerInfo) {
204 tty->print_cr("available container memory: " JULONG_FORMAT, avail_mem);
205 }
206 return avail_mem;
207 }
208 }
210 sysinfo(&si);
211 avail_mem = (julong)si.freeram * si.mem_unit;
212 if (Verbose) {
213 tty->print_cr("available memory: " JULONG_FORMAT, avail_mem);
214 }
215 return avail_mem;
216 }
218 julong os::physical_memory() {
219 jlong phys_mem = 0;
220 if (OSContainer::is_containerized()) {
221 jlong mem_limit;
222 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
223 if (PrintContainerInfo) {
224 tty->print_cr("total container memory: " JLONG_FORMAT, mem_limit);
225 }
226 return mem_limit;
227 }
229 if (PrintContainerInfo) {
230 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
231 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
232 }
233 }
235 phys_mem = Linux::physical_memory();
236 if (Verbose) {
237 tty->print_cr("total system memory: " JLONG_FORMAT, phys_mem);
238 }
239 return phys_mem;
240 }
242 ////////////////////////////////////////////////////////////////////////////////
243 // environment support
245 bool os::getenv(const char* name, char* buf, int len) {
246 const char* val = ::getenv(name);
247 if (val != NULL && strlen(val) < (size_t)len) {
248 strcpy(buf, val);
249 return true;
250 }
251 if (len > 0) buf[0] = 0; // return a null string
252 return false;
253 }
256 // Return true if user is running as root.
258 bool os::have_special_privileges() {
259 static bool init = false;
260 static bool privileges = false;
261 if (!init) {
262 privileges = (getuid() != geteuid()) || (getgid() != getegid());
263 init = true;
264 }
265 return privileges;
266 }
269 #ifndef SYS_gettid
270 // i386: 224, ia64: 1105, amd64: 186, sparc 143
271 #ifdef __ia64__
272 #define SYS_gettid 1105
273 #else
274 #ifdef __i386__
275 #define SYS_gettid 224
276 #else
277 #ifdef __amd64__
278 #define SYS_gettid 186
279 #else
280 #ifdef __sparc__
281 #define SYS_gettid 143
282 #else
283 #error define gettid for the arch
284 #endif
285 #endif
286 #endif
287 #endif
288 #endif
290 // Cpu architecture string
291 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
293 // pid_t gettid()
294 //
295 // Returns the kernel thread id of the currently running thread. Kernel
296 // thread id is used to access /proc.
297 //
298 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
299 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
300 //
301 pid_t os::Linux::gettid() {
302 int rslt = syscall(SYS_gettid);
303 if (rslt == -1) {
304 // old kernel, no NPTL support
305 return getpid();
306 } else {
307 return (pid_t)rslt;
308 }
309 }
311 // Most versions of linux have a bug where the number of processors are
312 // determined by looking at the /proc file system. In a chroot environment,
313 // the system call returns 1. This causes the VM to act as if it is
314 // a single processor and elide locking (see is_MP() call).
315 static bool unsafe_chroot_detected = false;
316 static const char *unstable_chroot_error = "/proc file system not found.\n"
317 "Java may be unstable running multithreaded in a chroot "
318 "environment on Linux when /proc filesystem is not mounted.";
320 void os::Linux::initialize_system_info() {
321 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
322 if (processor_count() == 1) {
323 pid_t pid = os::Linux::gettid();
324 char fname[32];
325 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
326 FILE *fp = fopen(fname, "r");
327 if (fp == NULL) {
328 unsafe_chroot_detected = true;
329 } else {
330 fclose(fp);
331 }
332 }
333 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
334 assert(processor_count() > 0, "linux error");
335 }
337 void os::init_system_properties_values() {
338 // The next steps are taken in the product version:
339 //
340 // Obtain the JAVA_HOME value from the location of libjvm.so.
341 // This library should be located at:
342 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
343 //
344 // If "/jre/lib/" appears at the right place in the path, then we
345 // assume libjvm.so is installed in a JDK and we use this path.
346 //
347 // Otherwise exit with message: "Could not create the Java virtual machine."
348 //
349 // The following extra steps are taken in the debugging version:
350 //
351 // If "/jre/lib/" does NOT appear at the right place in the path
352 // instead of exit check for $JAVA_HOME environment variable.
353 //
354 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
355 // then we append a fake suffix "hotspot/libjvm.so" to this path so
356 // it looks like libjvm.so is installed there
357 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
358 //
359 // Otherwise exit.
360 //
361 // Important note: if the location of libjvm.so changes this
362 // code needs to be changed accordingly.
364 // See ld(1):
365 // The linker uses the following search paths to locate required
366 // shared libraries:
367 // 1: ...
368 // ...
369 // 7: The default directories, normally /lib and /usr/lib.
370 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
371 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
372 #else
373 #define DEFAULT_LIBPATH "/lib:/usr/lib"
374 #endif
376 // Base path of extensions installed on the system.
377 #define SYS_EXT_DIR "/usr/java/packages"
378 #define EXTENSIONS_DIR "/lib/ext"
379 #define ENDORSED_DIR "/lib/endorsed"
381 // Buffer that fits several sprintfs.
382 // Note that the space for the colon and the trailing null are provided
383 // by the nulls included by the sizeof operator.
384 const size_t bufsize =
385 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
386 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
387 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
388 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
390 // sysclasspath, java_home, dll_dir
391 {
392 char *pslash;
393 os::jvm_path(buf, bufsize);
395 // Found the full path to libjvm.so.
396 // Now cut the path to <java_home>/jre if we can.
397 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
398 pslash = strrchr(buf, '/');
399 if (pslash != NULL) {
400 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
401 }
402 Arguments::set_dll_dir(buf);
404 if (pslash != NULL) {
405 pslash = strrchr(buf, '/');
406 if (pslash != NULL) {
407 *pslash = '\0'; // Get rid of /<arch>.
408 pslash = strrchr(buf, '/');
409 if (pslash != NULL) {
410 *pslash = '\0'; // Get rid of /lib.
411 }
412 }
413 }
414 Arguments::set_java_home(buf);
415 set_boot_path('/', ':');
416 }
418 // Where to look for native libraries.
419 //
420 // Note: Due to a legacy implementation, most of the library path
421 // is set in the launcher. This was to accomodate linking restrictions
422 // on legacy Linux implementations (which are no longer supported).
423 // Eventually, all the library path setting will be done here.
424 //
425 // However, to prevent the proliferation of improperly built native
426 // libraries, the new path component /usr/java/packages is added here.
427 // Eventually, all the library path setting will be done here.
428 {
429 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
430 // should always exist (until the legacy problem cited above is
431 // addressed).
432 const char *v = ::getenv("LD_LIBRARY_PATH");
433 const char *v_colon = ":";
434 if (v == NULL) { v = ""; v_colon = ""; }
435 // That's +1 for the colon and +1 for the trailing '\0'.
436 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
437 strlen(v) + 1 +
438 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
439 mtInternal);
440 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
441 Arguments::set_library_path(ld_library_path);
442 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
443 }
445 // Extensions directories.
446 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
447 Arguments::set_ext_dirs(buf);
449 // Endorsed standards default directory.
450 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
451 Arguments::set_endorsed_dirs(buf);
453 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
455 #undef DEFAULT_LIBPATH
456 #undef SYS_EXT_DIR
457 #undef EXTENSIONS_DIR
458 #undef ENDORSED_DIR
459 }
461 ////////////////////////////////////////////////////////////////////////////////
462 // breakpoint support
464 void os::breakpoint() {
465 BREAKPOINT;
466 }
468 extern "C" void breakpoint() {
469 // use debugger to set breakpoint here
470 }
472 ////////////////////////////////////////////////////////////////////////////////
473 // signal support
475 debug_only(static bool signal_sets_initialized = false);
476 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
478 bool os::Linux::is_sig_ignored(int sig) {
479 struct sigaction oact;
480 sigaction(sig, (struct sigaction*)NULL, &oact);
481 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
482 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
483 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
484 return true;
485 else
486 return false;
487 }
489 void os::Linux::signal_sets_init() {
490 // Should also have an assertion stating we are still single-threaded.
491 assert(!signal_sets_initialized, "Already initialized");
492 // Fill in signals that are necessarily unblocked for all threads in
493 // the VM. Currently, we unblock the following signals:
494 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
495 // by -Xrs (=ReduceSignalUsage));
496 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
497 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
498 // the dispositions or masks wrt these signals.
499 // Programs embedding the VM that want to use the above signals for their
500 // own purposes must, at this time, use the "-Xrs" option to prevent
501 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
502 // (See bug 4345157, and other related bugs).
503 // In reality, though, unblocking these signals is really a nop, since
504 // these signals are not blocked by default.
505 sigemptyset(&unblocked_sigs);
506 sigemptyset(&allowdebug_blocked_sigs);
507 sigaddset(&unblocked_sigs, SIGILL);
508 sigaddset(&unblocked_sigs, SIGSEGV);
509 sigaddset(&unblocked_sigs, SIGBUS);
510 sigaddset(&unblocked_sigs, SIGFPE);
511 #if defined(PPC64)
512 sigaddset(&unblocked_sigs, SIGTRAP);
513 #endif
514 sigaddset(&unblocked_sigs, SR_signum);
516 if (!ReduceSignalUsage) {
517 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
518 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
519 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
520 }
521 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
522 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
523 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
524 }
525 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
526 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
527 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
528 }
529 }
530 // Fill in signals that are blocked by all but the VM thread.
531 sigemptyset(&vm_sigs);
532 if (!ReduceSignalUsage)
533 sigaddset(&vm_sigs, BREAK_SIGNAL);
534 debug_only(signal_sets_initialized = true);
536 }
538 // These are signals that are unblocked while a thread is running Java.
539 // (For some reason, they get blocked by default.)
540 sigset_t* os::Linux::unblocked_signals() {
541 assert(signal_sets_initialized, "Not initialized");
542 return &unblocked_sigs;
543 }
545 // These are the signals that are blocked while a (non-VM) thread is
546 // running Java. Only the VM thread handles these signals.
547 sigset_t* os::Linux::vm_signals() {
548 assert(signal_sets_initialized, "Not initialized");
549 return &vm_sigs;
550 }
552 // These are signals that are blocked during cond_wait to allow debugger in
553 sigset_t* os::Linux::allowdebug_blocked_signals() {
554 assert(signal_sets_initialized, "Not initialized");
555 return &allowdebug_blocked_sigs;
556 }
558 void os::Linux::hotspot_sigmask(Thread* thread) {
560 //Save caller's signal mask before setting VM signal mask
561 sigset_t caller_sigmask;
562 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
564 OSThread* osthread = thread->osthread();
565 osthread->set_caller_sigmask(caller_sigmask);
567 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
569 if (!ReduceSignalUsage) {
570 if (thread->is_VM_thread()) {
571 // Only the VM thread handles BREAK_SIGNAL ...
572 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
573 } else {
574 // ... all other threads block BREAK_SIGNAL
575 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
576 }
577 }
578 }
580 //////////////////////////////////////////////////////////////////////////////
581 // detecting pthread library
583 void os::Linux::libpthread_init() {
584 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
585 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
586 // generic name for earlier versions.
587 // Define macros here so we can build HotSpot on old systems.
588 # ifndef _CS_GNU_LIBC_VERSION
589 # define _CS_GNU_LIBC_VERSION 2
590 # endif
591 # ifndef _CS_GNU_LIBPTHREAD_VERSION
592 # define _CS_GNU_LIBPTHREAD_VERSION 3
593 # endif
595 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
596 if (n > 0) {
597 char *str = (char *)malloc(n, mtInternal);
598 confstr(_CS_GNU_LIBC_VERSION, str, n);
599 os::Linux::set_glibc_version(str);
600 } else {
601 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
602 static char _gnu_libc_version[32];
603 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
604 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
605 os::Linux::set_glibc_version(_gnu_libc_version);
606 }
608 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
609 if (n > 0) {
610 char *str = (char *)malloc(n, mtInternal);
611 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
612 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
613 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
614 // is the case. LinuxThreads has a hard limit on max number of threads.
615 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
616 // On the other hand, NPTL does not have such a limit, sysconf()
617 // will return -1 and errno is not changed. Check if it is really NPTL.
618 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
619 strstr(str, "NPTL") &&
620 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
621 free(str);
622 os::Linux::set_libpthread_version("linuxthreads");
623 } else {
624 os::Linux::set_libpthread_version(str);
625 }
626 } else {
627 // glibc before 2.3.2 only has LinuxThreads.
628 os::Linux::set_libpthread_version("linuxthreads");
629 }
631 if (strstr(libpthread_version(), "NPTL")) {
632 os::Linux::set_is_NPTL();
633 } else {
634 os::Linux::set_is_LinuxThreads();
635 }
637 // LinuxThreads have two flavors: floating-stack mode, which allows variable
638 // stack size; and fixed-stack mode. NPTL is always floating-stack.
639 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
640 os::Linux::set_is_floating_stack();
641 }
642 }
644 /////////////////////////////////////////////////////////////////////////////
645 // thread stack
647 // Force Linux kernel to expand current thread stack. If "bottom" is close
648 // to the stack guard, caller should block all signals.
649 //
650 // MAP_GROWSDOWN:
651 // A special mmap() flag that is used to implement thread stacks. It tells
652 // kernel that the memory region should extend downwards when needed. This
653 // allows early versions of LinuxThreads to only mmap the first few pages
654 // when creating a new thread. Linux kernel will automatically expand thread
655 // stack as needed (on page faults).
656 //
657 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
658 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
659 // region, it's hard to tell if the fault is due to a legitimate stack
660 // access or because of reading/writing non-exist memory (e.g. buffer
661 // overrun). As a rule, if the fault happens below current stack pointer,
662 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
663 // application (see Linux kernel fault.c).
664 //
665 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
666 // stack overflow detection.
667 //
668 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
669 // not use this flag. However, the stack of initial thread is not created
670 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
671 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
672 // and then attach the thread to JVM.
673 //
674 // To get around the problem and allow stack banging on Linux, we need to
675 // manually expand thread stack after receiving the SIGSEGV.
676 //
677 // There are two ways to expand thread stack to address "bottom", we used
678 // both of them in JVM before 1.5:
679 // 1. adjust stack pointer first so that it is below "bottom", and then
680 // touch "bottom"
681 // 2. mmap() the page in question
682 //
683 // Now alternate signal stack is gone, it's harder to use 2. For instance,
684 // if current sp is already near the lower end of page 101, and we need to
685 // call mmap() to map page 100, it is possible that part of the mmap() frame
686 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
687 // That will destroy the mmap() frame and cause VM to crash.
688 //
689 // The following code works by adjusting sp first, then accessing the "bottom"
690 // page to force a page fault. Linux kernel will then automatically expand the
691 // stack mapping.
692 //
693 // _expand_stack_to() assumes its frame size is less than page size, which
694 // should always be true if the function is not inlined.
696 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
697 #define NOINLINE
698 #else
699 #define NOINLINE __attribute__ ((noinline))
700 #endif
702 static void _expand_stack_to(address bottom) NOINLINE;
704 static void _expand_stack_to(address bottom) {
705 address sp;
706 size_t size;
707 volatile char *p;
709 // Adjust bottom to point to the largest address within the same page, it
710 // gives us a one-page buffer if alloca() allocates slightly more memory.
711 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
712 bottom += os::Linux::page_size() - 1;
714 // sp might be slightly above current stack pointer; if that's the case, we
715 // will alloca() a little more space than necessary, which is OK. Don't use
716 // os::current_stack_pointer(), as its result can be slightly below current
717 // stack pointer, causing us to not alloca enough to reach "bottom".
718 sp = (address)&sp;
720 if (sp > bottom) {
721 size = sp - bottom;
722 p = (volatile char *)alloca(size);
723 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
724 p[0] = '\0';
725 }
726 }
728 void os::Linux::expand_stack_to(address bottom) {
729 _expand_stack_to(bottom);
730 }
732 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
733 assert(t!=NULL, "just checking");
734 assert(t->osthread()->expanding_stack(), "expand should be set");
735 assert(t->stack_base() != NULL, "stack_base was not initialized");
737 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
738 sigset_t mask_all, old_sigset;
739 sigfillset(&mask_all);
740 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
741 _expand_stack_to(addr);
742 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
743 return true;
744 }
745 return false;
746 }
748 //////////////////////////////////////////////////////////////////////////////
749 // create new thread
751 static address highest_vm_reserved_address();
753 // check if it's safe to start a new thread
754 static bool _thread_safety_check(Thread* thread) {
755 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
756 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
757 // Heap is mmap'ed at lower end of memory space. Thread stacks are
758 // allocated (MAP_FIXED) from high address space. Every thread stack
759 // occupies a fixed size slot (usually 2Mbytes, but user can change
760 // it to other values if they rebuild LinuxThreads).
761 //
762 // Problem with MAP_FIXED is that mmap() can still succeed even part of
763 // the memory region has already been mmap'ed. That means if we have too
764 // many threads and/or very large heap, eventually thread stack will
765 // collide with heap.
766 //
767 // Here we try to prevent heap/stack collision by comparing current
768 // stack bottom with the highest address that has been mmap'ed by JVM
769 // plus a safety margin for memory maps created by native code.
770 //
771 // This feature can be disabled by setting ThreadSafetyMargin to 0
772 //
773 if (ThreadSafetyMargin > 0) {
774 address stack_bottom = os::current_stack_base() - os::current_stack_size();
776 // not safe if our stack extends below the safety margin
777 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
778 } else {
779 return true;
780 }
781 } else {
782 // Floating stack LinuxThreads or NPTL:
783 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
784 // there's not enough space left, pthread_create() will fail. If we come
785 // here, that means enough space has been reserved for stack.
786 return true;
787 }
788 }
790 // Thread start routine for all newly created threads
791 static void *java_start(Thread *thread) {
792 // Try to randomize the cache line index of hot stack frames.
793 // This helps when threads of the same stack traces evict each other's
794 // cache lines. The threads can be either from the same JVM instance, or
795 // from different JVM instances. The benefit is especially true for
796 // processors with hyperthreading technology.
797 static int counter = 0;
798 int pid = os::current_process_id();
799 alloca(((pid ^ counter++) & 7) * 128);
801 ThreadLocalStorage::set_thread(thread);
803 OSThread* osthread = thread->osthread();
804 Monitor* sync = osthread->startThread_lock();
806 // non floating stack LinuxThreads needs extra check, see above
807 if (!_thread_safety_check(thread)) {
808 // notify parent thread
809 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
810 osthread->set_state(ZOMBIE);
811 sync->notify_all();
812 return NULL;
813 }
815 // thread_id is kernel thread id (similar to Solaris LWP id)
816 osthread->set_thread_id(os::Linux::gettid());
818 if (UseNUMA) {
819 int lgrp_id = os::numa_get_group_id();
820 if (lgrp_id != -1) {
821 thread->set_lgrp_id(lgrp_id);
822 }
823 }
824 // initialize signal mask for this thread
825 os::Linux::hotspot_sigmask(thread);
827 // initialize floating point control register
828 os::Linux::init_thread_fpu_state();
830 // handshaking with parent thread
831 {
832 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
834 // notify parent thread
835 osthread->set_state(INITIALIZED);
836 sync->notify_all();
838 // wait until os::start_thread()
839 while (osthread->get_state() == INITIALIZED) {
840 sync->wait(Mutex::_no_safepoint_check_flag);
841 }
842 }
844 // call one more level start routine
845 thread->run();
847 return 0;
848 }
850 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
851 assert(thread->osthread() == NULL, "caller responsible");
853 // Allocate the OSThread object
854 OSThread* osthread = new OSThread(NULL, NULL);
855 if (osthread == NULL) {
856 return false;
857 }
859 // set the correct thread state
860 osthread->set_thread_type(thr_type);
862 // Initial state is ALLOCATED but not INITIALIZED
863 osthread->set_state(ALLOCATED);
865 thread->set_osthread(osthread);
867 // init thread attributes
868 pthread_attr_t attr;
869 pthread_attr_init(&attr);
870 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
872 // stack size
873 if (os::Linux::supports_variable_stack_size()) {
874 // calculate stack size if it's not specified by caller
875 if (stack_size == 0) {
876 stack_size = os::Linux::default_stack_size(thr_type);
878 switch (thr_type) {
879 case os::java_thread:
880 // Java threads use ThreadStackSize which default value can be
881 // changed with the flag -Xss
882 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
883 stack_size = JavaThread::stack_size_at_create();
884 break;
885 case os::compiler_thread:
886 if (CompilerThreadStackSize > 0) {
887 stack_size = (size_t)(CompilerThreadStackSize * K);
888 break;
889 } // else fall through:
890 // use VMThreadStackSize if CompilerThreadStackSize is not defined
891 case os::vm_thread:
892 case os::pgc_thread:
893 case os::cgc_thread:
894 case os::watcher_thread:
895 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
896 break;
897 }
898 }
900 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
901 pthread_attr_setstacksize(&attr, stack_size);
902 } else {
903 // let pthread_create() pick the default value.
904 }
906 // glibc guard page
907 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
909 ThreadState state;
911 {
912 // Serialize thread creation if we are running with fixed stack LinuxThreads
913 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
914 if (lock) {
915 os::Linux::createThread_lock()->lock_without_safepoint_check();
916 }
918 pthread_t tid;
919 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
921 pthread_attr_destroy(&attr);
923 if (ret != 0) {
924 if (PrintMiscellaneous && (Verbose || WizardMode)) {
925 perror("pthread_create()");
926 }
927 // Need to clean up stuff we've allocated so far
928 thread->set_osthread(NULL);
929 delete osthread;
930 if (lock) os::Linux::createThread_lock()->unlock();
931 return false;
932 }
934 // Store pthread info into the OSThread
935 osthread->set_pthread_id(tid);
937 // Wait until child thread is either initialized or aborted
938 {
939 Monitor* sync_with_child = osthread->startThread_lock();
940 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
941 while ((state = osthread->get_state()) == ALLOCATED) {
942 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
943 }
944 }
946 if (lock) {
947 os::Linux::createThread_lock()->unlock();
948 }
949 }
951 // Aborted due to thread limit being reached
952 if (state == ZOMBIE) {
953 thread->set_osthread(NULL);
954 delete osthread;
955 return false;
956 }
958 // The thread is returned suspended (in state INITIALIZED),
959 // and is started higher up in the call chain
960 assert(state == INITIALIZED, "race condition");
961 return true;
962 }
964 /////////////////////////////////////////////////////////////////////////////
965 // attach existing thread
967 // bootstrap the main thread
968 bool os::create_main_thread(JavaThread* thread) {
969 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
970 return create_attached_thread(thread);
971 }
973 bool os::create_attached_thread(JavaThread* thread) {
974 #ifdef ASSERT
975 thread->verify_not_published();
976 #endif
978 // Allocate the OSThread object
979 OSThread* osthread = new OSThread(NULL, NULL);
981 if (osthread == NULL) {
982 return false;
983 }
985 // Store pthread info into the OSThread
986 osthread->set_thread_id(os::Linux::gettid());
987 osthread->set_pthread_id(::pthread_self());
989 // initialize floating point control register
990 os::Linux::init_thread_fpu_state();
992 // Initial thread state is RUNNABLE
993 osthread->set_state(RUNNABLE);
995 thread->set_osthread(osthread);
997 if (UseNUMA) {
998 int lgrp_id = os::numa_get_group_id();
999 if (lgrp_id != -1) {
1000 thread->set_lgrp_id(lgrp_id);
1001 }
1002 }
1004 if (os::is_primordial_thread()) {
1005 // If current thread is primordial thread, its stack is mapped on demand,
1006 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1007 // the entire stack region to avoid SEGV in stack banging.
1008 // It is also useful to get around the heap-stack-gap problem on SuSE
1009 // kernel (see 4821821 for details). We first expand stack to the top
1010 // of yellow zone, then enable stack yellow zone (order is significant,
1011 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1012 // is no gap between the last two virtual memory regions.
1014 JavaThread *jt = (JavaThread *)thread;
1015 address addr = jt->stack_yellow_zone_base();
1016 assert(addr != NULL, "initialization problem?");
1017 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1019 osthread->set_expanding_stack();
1020 os::Linux::manually_expand_stack(jt, addr);
1021 osthread->clear_expanding_stack();
1022 }
1024 // initialize signal mask for this thread
1025 // and save the caller's signal mask
1026 os::Linux::hotspot_sigmask(thread);
1028 return true;
1029 }
1031 void os::pd_start_thread(Thread* thread) {
1032 OSThread * osthread = thread->osthread();
1033 assert(osthread->get_state() != INITIALIZED, "just checking");
1034 Monitor* sync_with_child = osthread->startThread_lock();
1035 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1036 sync_with_child->notify();
1038 #if defined MIPS && !defined ZERO
1039 //To be accessed in NativeGeneralJump::patch_verified_entry()
1040 if (thread->is_Java_thread())
1041 {
1042 ((JavaThread*)thread)->set_handle_wrong_method_stub(SharedRuntime::get_handle_wrong_method_stub());
1043 }
1044 #endif
1045 }
1047 // Free Linux resources related to the OSThread
1048 void os::free_thread(OSThread* osthread) {
1049 assert(osthread != NULL, "osthread not set");
1051 if (Thread::current()->osthread() == osthread) {
1052 // Restore caller's signal mask
1053 sigset_t sigmask = osthread->caller_sigmask();
1054 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1055 }
1057 delete osthread;
1058 }
1060 //////////////////////////////////////////////////////////////////////////////
1061 // thread local storage
1063 // Restore the thread pointer if the destructor is called. This is in case
1064 // someone from JNI code sets up a destructor with pthread_key_create to run
1065 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1066 // will hang or crash. When detachCurrentThread is called the key will be set
1067 // to null and we will not be called again. If detachCurrentThread is never
1068 // called we could loop forever depending on the pthread implementation.
1069 static void restore_thread_pointer(void* p) {
1070 Thread* thread = (Thread*) p;
1071 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1072 }
1074 int os::allocate_thread_local_storage() {
1075 pthread_key_t key;
1076 int rslt = pthread_key_create(&key, restore_thread_pointer);
1077 assert(rslt == 0, "cannot allocate thread local storage");
1078 return (int)key;
1079 }
1081 // Note: This is currently not used by VM, as we don't destroy TLS key
1082 // on VM exit.
1083 void os::free_thread_local_storage(int index) {
1084 int rslt = pthread_key_delete((pthread_key_t)index);
1085 assert(rslt == 0, "invalid index");
1086 }
1088 void os::thread_local_storage_at_put(int index, void* value) {
1089 int rslt = pthread_setspecific((pthread_key_t)index, value);
1090 assert(rslt == 0, "pthread_setspecific failed");
1091 }
1093 extern "C" Thread* get_thread() {
1094 return ThreadLocalStorage::thread();
1095 }
1097 //////////////////////////////////////////////////////////////////////////////
1098 // primordial thread
1100 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1101 bool os::is_primordial_thread(void) {
1102 char dummy;
1103 // If called before init complete, thread stack bottom will be null.
1104 // Can be called if fatal error occurs before initialization.
1105 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1106 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1107 os::Linux::initial_thread_stack_size() != 0,
1108 "os::init did not locate primordial thread's stack region");
1109 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1110 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1111 os::Linux::initial_thread_stack_size()) {
1112 return true;
1113 } else {
1114 return false;
1115 }
1116 }
1118 // Find the virtual memory area that contains addr
1119 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1120 FILE *fp = fopen("/proc/self/maps", "r");
1121 if (fp) {
1122 address low, high;
1123 while (!feof(fp)) {
1124 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1125 if (low <= addr && addr < high) {
1126 if (vma_low) *vma_low = low;
1127 if (vma_high) *vma_high = high;
1128 fclose (fp);
1129 return true;
1130 }
1131 }
1132 for (;;) {
1133 int ch = fgetc(fp);
1134 if (ch == EOF || ch == (int)'\n') break;
1135 }
1136 }
1137 fclose(fp);
1138 }
1139 return false;
1140 }
1142 // Locate primordial thread stack. This special handling of primordial thread stack
1143 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1144 // bogus value for the primordial process thread. While the launcher has created
1145 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1146 // JNI invocation API from a primordial thread.
1147 void os::Linux::capture_initial_stack(size_t max_size) {
1149 // max_size is either 0 (which means accept OS default for thread stacks) or
1150 // a user-specified value known to be at least the minimum needed. If we
1151 // are actually on the primordial thread we can make it appear that we have a
1152 // smaller max_size stack by inserting the guard pages at that location. But we
1153 // cannot do anything to emulate a larger stack than what has been provided by
1154 // the OS or threading library. In fact if we try to use a stack greater than
1155 // what is set by rlimit then we will crash the hosting process.
1157 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1158 // If this is "unlimited" then it will be a huge value.
1159 struct rlimit rlim;
1160 getrlimit(RLIMIT_STACK, &rlim);
1161 size_t stack_size = rlim.rlim_cur;
1163 // 6308388: a bug in ld.so will relocate its own .data section to the
1164 // lower end of primordial stack; reduce ulimit -s value a little bit
1165 // so we won't install guard page on ld.so's data section.
1166 // But ensure we don't underflow the stack size - allow 1 page spare
1167 if (stack_size >= (size_t)(3 * page_size())) {
1168 stack_size -= 2 * page_size();
1169 }
1171 // Try to figure out where the stack base (top) is. This is harder.
1172 //
1173 // When an application is started, glibc saves the initial stack pointer in
1174 // a global variable "__libc_stack_end", which is then used by system
1175 // libraries. __libc_stack_end should be pretty close to stack top. The
1176 // variable is available since the very early days. However, because it is
1177 // a private interface, it could disappear in the future.
1178 //
1179 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1180 // to __libc_stack_end, it is very close to stack top, but isn't the real
1181 // stack top. Note that /proc may not exist if VM is running as a chroot
1182 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1183 // /proc/<pid>/stat could change in the future (though unlikely).
1184 //
1185 // We try __libc_stack_end first. If that doesn't work, look for
1186 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1187 // as a hint, which should work well in most cases.
1189 uintptr_t stack_start;
1191 // try __libc_stack_end first
1192 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1193 if (p && *p) {
1194 stack_start = *p;
1195 } else {
1196 // see if we can get the start_stack field from /proc/self/stat
1197 FILE *fp;
1198 int pid;
1199 char state;
1200 int ppid;
1201 int pgrp;
1202 int session;
1203 int nr;
1204 int tpgrp;
1205 unsigned long flags;
1206 unsigned long minflt;
1207 unsigned long cminflt;
1208 unsigned long majflt;
1209 unsigned long cmajflt;
1210 unsigned long utime;
1211 unsigned long stime;
1212 long cutime;
1213 long cstime;
1214 long prio;
1215 long nice;
1216 long junk;
1217 long it_real;
1218 uintptr_t start;
1219 uintptr_t vsize;
1220 intptr_t rss;
1221 uintptr_t rsslim;
1222 uintptr_t scodes;
1223 uintptr_t ecode;
1224 int i;
1226 // Figure what the primordial thread stack base is. Code is inspired
1227 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1228 // followed by command name surrounded by parentheses, state, etc.
1229 char stat[2048];
1230 int statlen;
1232 fp = fopen("/proc/self/stat", "r");
1233 if (fp) {
1234 statlen = fread(stat, 1, 2047, fp);
1235 stat[statlen] = '\0';
1236 fclose(fp);
1238 // Skip pid and the command string. Note that we could be dealing with
1239 // weird command names, e.g. user could decide to rename java launcher
1240 // to "java 1.4.2 :)", then the stat file would look like
1241 // 1234 (java 1.4.2 :)) R ... ...
1242 // We don't really need to know the command string, just find the last
1243 // occurrence of ")" and then start parsing from there. See bug 4726580.
1244 char * s = strrchr(stat, ')');
1246 i = 0;
1247 if (s) {
1248 // Skip blank chars
1249 do s++; while (isspace(*s));
1251 #define _UFM UINTX_FORMAT
1252 #define _DFM INTX_FORMAT
1254 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1255 /* 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 */
1256 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,
1257 &state, /* 3 %c */
1258 &ppid, /* 4 %d */
1259 &pgrp, /* 5 %d */
1260 &session, /* 6 %d */
1261 &nr, /* 7 %d */
1262 &tpgrp, /* 8 %d */
1263 &flags, /* 9 %lu */
1264 &minflt, /* 10 %lu */
1265 &cminflt, /* 11 %lu */
1266 &majflt, /* 12 %lu */
1267 &cmajflt, /* 13 %lu */
1268 &utime, /* 14 %lu */
1269 &stime, /* 15 %lu */
1270 &cutime, /* 16 %ld */
1271 &cstime, /* 17 %ld */
1272 &prio, /* 18 %ld */
1273 &nice, /* 19 %ld */
1274 &junk, /* 20 %ld */
1275 &it_real, /* 21 %ld */
1276 &start, /* 22 UINTX_FORMAT */
1277 &vsize, /* 23 UINTX_FORMAT */
1278 &rss, /* 24 INTX_FORMAT */
1279 &rsslim, /* 25 UINTX_FORMAT */
1280 &scodes, /* 26 UINTX_FORMAT */
1281 &ecode, /* 27 UINTX_FORMAT */
1282 &stack_start); /* 28 UINTX_FORMAT */
1283 }
1285 #undef _UFM
1286 #undef _DFM
1288 if (i != 28 - 2) {
1289 assert(false, "Bad conversion from /proc/self/stat");
1290 // product mode - assume we are the primordial thread, good luck in the
1291 // embedded case.
1292 warning("Can't detect primordial thread stack location - bad conversion");
1293 stack_start = (uintptr_t) &rlim;
1294 }
1295 } else {
1296 // For some reason we can't open /proc/self/stat (for example, running on
1297 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1298 // most cases, so don't abort:
1299 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1300 stack_start = (uintptr_t) &rlim;
1301 }
1302 }
1304 // Now we have a pointer (stack_start) very close to the stack top, the
1305 // next thing to do is to figure out the exact location of stack top. We
1306 // can find out the virtual memory area that contains stack_start by
1307 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1308 // and its upper limit is the real stack top. (again, this would fail if
1309 // running inside chroot, because /proc may not exist.)
1311 uintptr_t stack_top;
1312 address low, high;
1313 if (find_vma((address)stack_start, &low, &high)) {
1314 // success, "high" is the true stack top. (ignore "low", because initial
1315 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1316 stack_top = (uintptr_t)high;
1317 } else {
1318 // failed, likely because /proc/self/maps does not exist
1319 warning("Can't detect primordial thread stack location - find_vma failed");
1320 // best effort: stack_start is normally within a few pages below the real
1321 // stack top, use it as stack top, and reduce stack size so we won't put
1322 // guard page outside stack.
1323 stack_top = stack_start;
1324 stack_size -= 16 * page_size();
1325 }
1327 // stack_top could be partially down the page so align it
1328 stack_top = align_size_up(stack_top, page_size());
1330 // Allowed stack value is minimum of max_size and what we derived from rlimit
1331 if (max_size > 0) {
1332 _initial_thread_stack_size = MIN2(max_size, stack_size);
1333 } else {
1334 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1335 // clamp it at 8MB as we do on Solaris
1336 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1337 }
1339 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1340 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1341 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1342 }
1344 ////////////////////////////////////////////////////////////////////////////////
1345 // time support
1347 // Time since start-up in seconds to a fine granularity.
1348 // Used by VMSelfDestructTimer and the MemProfiler.
1349 double os::elapsedTime() {
1351 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1352 }
1354 jlong os::elapsed_counter() {
1355 return javaTimeNanos() - initial_time_count;
1356 }
1358 jlong os::elapsed_frequency() {
1359 return NANOSECS_PER_SEC; // nanosecond resolution
1360 }
1362 bool os::supports_vtime() { return true; }
1363 bool os::enable_vtime() { return false; }
1364 bool os::vtime_enabled() { return false; }
1366 double os::elapsedVTime() {
1367 struct rusage usage;
1368 int retval = getrusage(RUSAGE_THREAD, &usage);
1369 if (retval == 0) {
1370 return (double) (usage.ru_utime.tv_sec + usage.ru_stime.tv_sec) + (double) (usage.ru_utime.tv_usec + usage.ru_stime.tv_usec) / (1000 * 1000);
1371 } else {
1372 // better than nothing, but not much
1373 return elapsedTime();
1374 }
1375 }
1377 jlong os::javaTimeMillis() {
1378 timeval time;
1379 int status = gettimeofday(&time, NULL);
1380 assert(status != -1, "linux error");
1381 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1382 }
1384 #ifndef CLOCK_MONOTONIC
1385 #define CLOCK_MONOTONIC (1)
1386 #endif
1388 void os::Linux::clock_init() {
1389 // we do dlopen's in this particular order due to bug in linux
1390 // dynamical loader (see 6348968) leading to crash on exit
1391 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1392 if (handle == NULL) {
1393 handle = dlopen("librt.so", RTLD_LAZY);
1394 }
1396 if (handle) {
1397 int (*clock_getres_func)(clockid_t, struct timespec*) =
1398 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1399 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1400 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1401 if (clock_getres_func && clock_gettime_func) {
1402 // See if monotonic clock is supported by the kernel. Note that some
1403 // early implementations simply return kernel jiffies (updated every
1404 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1405 // for nano time (though the monotonic property is still nice to have).
1406 // It's fixed in newer kernels, however clock_getres() still returns
1407 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1408 // resolution for now. Hopefully as people move to new kernels, this
1409 // won't be a problem.
1410 struct timespec res;
1411 struct timespec tp;
1412 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1413 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1414 // yes, monotonic clock is supported
1415 _clock_gettime = clock_gettime_func;
1416 return;
1417 } else {
1418 // close librt if there is no monotonic clock
1419 dlclose(handle);
1420 }
1421 }
1422 }
1423 warning("No monotonic clock was available - timed services may " \
1424 "be adversely affected if the time-of-day clock changes");
1425 }
1427 #ifndef SYS_clock_getres
1429 #if defined(IA32) || defined(AMD64)
1430 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1431 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1432 #else
1433 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1434 #define sys_clock_getres(x,y) -1
1435 #endif
1437 #else
1438 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1439 #endif
1441 void os::Linux::fast_thread_clock_init() {
1442 if (!UseLinuxPosixThreadCPUClocks) {
1443 return;
1444 }
1445 clockid_t clockid;
1446 struct timespec tp;
1447 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1448 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1450 // Switch to using fast clocks for thread cpu time if
1451 // the sys_clock_getres() returns 0 error code.
1452 // Note, that some kernels may support the current thread
1453 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1454 // returned by the pthread_getcpuclockid().
1455 // If the fast Posix clocks are supported then the sys_clock_getres()
1456 // must return at least tp.tv_sec == 0 which means a resolution
1457 // better than 1 sec. This is extra check for reliability.
1459 if(pthread_getcpuclockid_func &&
1460 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1461 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1463 _supports_fast_thread_cpu_time = true;
1464 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1465 }
1466 }
1468 jlong os::javaTimeNanos() {
1469 if (Linux::supports_monotonic_clock()) {
1470 struct timespec tp;
1471 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1472 assert(status == 0, "gettime error");
1473 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1474 return result;
1475 } else {
1476 timeval time;
1477 int status = gettimeofday(&time, NULL);
1478 assert(status != -1, "linux error");
1479 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1480 return 1000 * usecs;
1481 }
1482 }
1484 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1485 if (Linux::supports_monotonic_clock()) {
1486 info_ptr->max_value = ALL_64_BITS;
1488 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1489 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1490 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1491 } else {
1492 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1493 info_ptr->max_value = ALL_64_BITS;
1495 // gettimeofday is a real time clock so it skips
1496 info_ptr->may_skip_backward = true;
1497 info_ptr->may_skip_forward = true;
1498 }
1500 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1501 }
1503 // Return the real, user, and system times in seconds from an
1504 // arbitrary fixed point in the past.
1505 bool os::getTimesSecs(double* process_real_time,
1506 double* process_user_time,
1507 double* process_system_time) {
1508 struct tms ticks;
1509 clock_t real_ticks = times(&ticks);
1511 if (real_ticks == (clock_t) (-1)) {
1512 return false;
1513 } else {
1514 double ticks_per_second = (double) clock_tics_per_sec;
1515 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1516 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1517 *process_real_time = ((double) real_ticks) / ticks_per_second;
1519 return true;
1520 }
1521 }
1524 char * os::local_time_string(char *buf, size_t buflen) {
1525 struct tm t;
1526 time_t long_time;
1527 time(&long_time);
1528 localtime_r(&long_time, &t);
1529 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1530 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1531 t.tm_hour, t.tm_min, t.tm_sec);
1532 return buf;
1533 }
1535 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1536 return localtime_r(clock, res);
1537 }
1539 ////////////////////////////////////////////////////////////////////////////////
1540 // runtime exit support
1542 // Note: os::shutdown() might be called very early during initialization, or
1543 // called from signal handler. Before adding something to os::shutdown(), make
1544 // sure it is async-safe and can handle partially initialized VM.
1545 void os::shutdown() {
1547 // allow PerfMemory to attempt cleanup of any persistent resources
1548 perfMemory_exit();
1550 // needs to remove object in file system
1551 AttachListener::abort();
1553 // flush buffered output, finish log files
1554 ostream_abort();
1556 // Check for abort hook
1557 abort_hook_t abort_hook = Arguments::abort_hook();
1558 if (abort_hook != NULL) {
1559 abort_hook();
1560 }
1562 }
1564 // Note: os::abort() might be called very early during initialization, or
1565 // called from signal handler. Before adding something to os::abort(), make
1566 // sure it is async-safe and can handle partially initialized VM.
1567 void os::abort(bool dump_core) {
1568 os::shutdown();
1569 if (dump_core) {
1570 #ifndef PRODUCT
1571 fdStream out(defaultStream::output_fd());
1572 out.print_raw("Current thread is ");
1573 char buf[16];
1574 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1575 out.print_raw_cr(buf);
1576 out.print_raw_cr("Dumping core ...");
1577 #endif
1578 ::abort(); // dump core
1579 }
1581 ::exit(1);
1582 }
1584 // Die immediately, no exit hook, no abort hook, no cleanup.
1585 void os::die() {
1586 // _exit() on LinuxThreads only kills current thread
1587 ::abort();
1588 }
1591 // This method is a copy of JDK's sysGetLastErrorString
1592 // from src/solaris/hpi/src/system_md.c
1594 size_t os::lasterror(char *buf, size_t len) {
1596 if (errno == 0) return 0;
1598 const char *s = ::strerror(errno);
1599 size_t n = ::strlen(s);
1600 if (n >= len) {
1601 n = len - 1;
1602 }
1603 ::strncpy(buf, s, n);
1604 buf[n] = '\0';
1605 return n;
1606 }
1608 intx os::current_thread_id() { return (intx)pthread_self(); }
1609 int os::current_process_id() {
1611 // Under the old linux thread library, linux gives each thread
1612 // its own process id. Because of this each thread will return
1613 // a different pid if this method were to return the result
1614 // of getpid(2). Linux provides no api that returns the pid
1615 // of the launcher thread for the vm. This implementation
1616 // returns a unique pid, the pid of the launcher thread
1617 // that starts the vm 'process'.
1619 // Under the NPTL, getpid() returns the same pid as the
1620 // launcher thread rather than a unique pid per thread.
1621 // Use gettid() if you want the old pre NPTL behaviour.
1623 // if you are looking for the result of a call to getpid() that
1624 // returns a unique pid for the calling thread, then look at the
1625 // OSThread::thread_id() method in osThread_linux.hpp file
1627 return (int)(_initial_pid ? _initial_pid : getpid());
1628 }
1630 // DLL functions
1632 const char* os::dll_file_extension() { return ".so"; }
1634 // This must be hard coded because it's the system's temporary
1635 // directory not the java application's temp directory, ala java.io.tmpdir.
1636 const char* os::get_temp_directory() { return "/tmp"; }
1638 static bool file_exists(const char* filename) {
1639 struct stat statbuf;
1640 if (filename == NULL || strlen(filename) == 0) {
1641 return false;
1642 }
1643 return os::stat(filename, &statbuf) == 0;
1644 }
1646 bool os::dll_build_name(char* buffer, size_t buflen,
1647 const char* pname, const char* fname) {
1648 bool retval = false;
1649 // Copied from libhpi
1650 const size_t pnamelen = pname ? strlen(pname) : 0;
1652 // Return error on buffer overflow.
1653 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1654 return retval;
1655 }
1657 if (pnamelen == 0) {
1658 snprintf(buffer, buflen, "lib%s.so", fname);
1659 retval = true;
1660 } else if (strchr(pname, *os::path_separator()) != NULL) {
1661 int n;
1662 char** pelements = split_path(pname, &n);
1663 if (pelements == NULL) {
1664 return false;
1665 }
1666 for (int i = 0 ; i < n ; i++) {
1667 // Really shouldn't be NULL, but check can't hurt
1668 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1669 continue; // skip the empty path values
1670 }
1671 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1672 if (file_exists(buffer)) {
1673 retval = true;
1674 break;
1675 }
1676 }
1677 // release the storage
1678 for (int i = 0 ; i < n ; i++) {
1679 if (pelements[i] != NULL) {
1680 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1681 }
1682 }
1683 if (pelements != NULL) {
1684 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1685 }
1686 } else {
1687 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1688 retval = true;
1689 }
1690 return retval;
1691 }
1693 // check if addr is inside libjvm.so
1694 bool os::address_is_in_vm(address addr) {
1695 static address libjvm_base_addr;
1696 Dl_info dlinfo;
1698 if (libjvm_base_addr == NULL) {
1699 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1700 libjvm_base_addr = (address)dlinfo.dli_fbase;
1701 }
1702 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1703 }
1705 if (dladdr((void *)addr, &dlinfo) != 0) {
1706 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1707 }
1709 return false;
1710 }
1712 bool os::dll_address_to_function_name(address addr, char *buf,
1713 int buflen, int *offset) {
1714 // buf is not optional, but offset is optional
1715 assert(buf != NULL, "sanity check");
1717 Dl_info dlinfo;
1719 if (dladdr((void*)addr, &dlinfo) != 0) {
1720 // see if we have a matching symbol
1721 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1722 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1723 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1724 }
1725 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1726 return true;
1727 }
1728 // no matching symbol so try for just file info
1729 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1730 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1731 buf, buflen, offset, dlinfo.dli_fname)) {
1732 return true;
1733 }
1734 }
1735 }
1737 buf[0] = '\0';
1738 if (offset != NULL) *offset = -1;
1739 return false;
1740 }
1742 struct _address_to_library_name {
1743 address addr; // input : memory address
1744 size_t buflen; // size of fname
1745 char* fname; // output: library name
1746 address base; // library base addr
1747 };
1749 static int address_to_library_name_callback(struct dl_phdr_info *info,
1750 size_t size, void *data) {
1751 int i;
1752 bool found = false;
1753 address libbase = NULL;
1754 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1756 // iterate through all loadable segments
1757 for (i = 0; i < info->dlpi_phnum; i++) {
1758 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1759 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1760 // base address of a library is the lowest address of its loaded
1761 // segments.
1762 if (libbase == NULL || libbase > segbase) {
1763 libbase = segbase;
1764 }
1765 // see if 'addr' is within current segment
1766 if (segbase <= d->addr &&
1767 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1768 found = true;
1769 }
1770 }
1771 }
1773 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1774 // so dll_address_to_library_name() can fall through to use dladdr() which
1775 // can figure out executable name from argv[0].
1776 if (found && info->dlpi_name && info->dlpi_name[0]) {
1777 d->base = libbase;
1778 if (d->fname) {
1779 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1780 }
1781 return 1;
1782 }
1783 return 0;
1784 }
1786 bool os::dll_address_to_library_name(address addr, char* buf,
1787 int buflen, int* offset) {
1788 // buf is not optional, but offset is optional
1789 assert(buf != NULL, "sanity check");
1791 Dl_info dlinfo;
1792 struct _address_to_library_name data;
1794 // There is a bug in old glibc dladdr() implementation that it could resolve
1795 // to wrong library name if the .so file has a base address != NULL. Here
1796 // we iterate through the program headers of all loaded libraries to find
1797 // out which library 'addr' really belongs to. This workaround can be
1798 // removed once the minimum requirement for glibc is moved to 2.3.x.
1799 data.addr = addr;
1800 data.fname = buf;
1801 data.buflen = buflen;
1802 data.base = NULL;
1803 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1805 if (rslt) {
1806 // buf already contains library name
1807 if (offset) *offset = addr - data.base;
1808 return true;
1809 }
1810 if (dladdr((void*)addr, &dlinfo) != 0) {
1811 if (dlinfo.dli_fname != NULL) {
1812 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1813 }
1814 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1815 *offset = addr - (address)dlinfo.dli_fbase;
1816 }
1817 return true;
1818 }
1820 buf[0] = '\0';
1821 if (offset) *offset = -1;
1822 return false;
1823 }
1825 // Loads .dll/.so and
1826 // in case of error it checks if .dll/.so was built for the
1827 // same architecture as Hotspot is running on
1830 // Remember the stack's state. The Linux dynamic linker will change
1831 // the stack to 'executable' at most once, so we must safepoint only once.
1832 bool os::Linux::_stack_is_executable = false;
1834 // VM operation that loads a library. This is necessary if stack protection
1835 // of the Java stacks can be lost during loading the library. If we
1836 // do not stop the Java threads, they can stack overflow before the stacks
1837 // are protected again.
1838 class VM_LinuxDllLoad: public VM_Operation {
1839 private:
1840 const char *_filename;
1841 char *_ebuf;
1842 int _ebuflen;
1843 void *_lib;
1844 public:
1845 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1846 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1847 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1848 void doit() {
1849 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1850 os::Linux::_stack_is_executable = true;
1851 }
1852 void* loaded_library() { return _lib; }
1853 };
1855 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1856 {
1857 void * result = NULL;
1858 bool load_attempted = false;
1860 // Check whether the library to load might change execution rights
1861 // of the stack. If they are changed, the protection of the stack
1862 // guard pages will be lost. We need a safepoint to fix this.
1863 //
1864 // See Linux man page execstack(8) for more info.
1865 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1866 ElfFile ef(filename);
1867 if (!ef.specifies_noexecstack()) {
1868 if (!is_init_completed()) {
1869 os::Linux::_stack_is_executable = true;
1870 // This is OK - No Java threads have been created yet, and hence no
1871 // stack guard pages to fix.
1872 //
1873 // This should happen only when you are building JDK7 using a very
1874 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1875 //
1876 // Dynamic loader will make all stacks executable after
1877 // this function returns, and will not do that again.
1878 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1879 } else {
1880 warning("You have loaded library %s which might have disabled stack guard. "
1881 "The VM will try to fix the stack guard now.\n"
1882 "It's highly recommended that you fix the library with "
1883 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1884 filename);
1886 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1887 JavaThread *jt = JavaThread::current();
1888 if (jt->thread_state() != _thread_in_native) {
1889 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1890 // that requires ExecStack. Cannot enter safe point. Let's give up.
1891 warning("Unable to fix stack guard. Giving up.");
1892 } else {
1893 if (!LoadExecStackDllInVMThread) {
1894 // This is for the case where the DLL has an static
1895 // constructor function that executes JNI code. We cannot
1896 // load such DLLs in the VMThread.
1897 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1898 }
1900 ThreadInVMfromNative tiv(jt);
1901 debug_only(VMNativeEntryWrapper vew;)
1903 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1904 VMThread::execute(&op);
1905 if (LoadExecStackDllInVMThread) {
1906 result = op.loaded_library();
1907 }
1908 load_attempted = true;
1909 }
1910 }
1911 }
1912 }
1914 if (!load_attempted) {
1915 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1916 }
1918 if (result != NULL) {
1919 // Successful loading
1920 return result;
1921 }
1923 Elf32_Ehdr elf_head;
1924 int diag_msg_max_length=ebuflen-strlen(ebuf);
1925 char* diag_msg_buf=ebuf+strlen(ebuf);
1927 if (diag_msg_max_length==0) {
1928 // No more space in ebuf for additional diagnostics message
1929 return NULL;
1930 }
1933 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1935 if (file_descriptor < 0) {
1936 // Can't open library, report dlerror() message
1937 return NULL;
1938 }
1940 bool failed_to_read_elf_head=
1941 (sizeof(elf_head)!=
1942 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1944 ::close(file_descriptor);
1945 if (failed_to_read_elf_head) {
1946 // file i/o error - report dlerror() msg
1947 return NULL;
1948 }
1950 typedef struct {
1951 Elf32_Half code; // Actual value as defined in elf.h
1952 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1953 char elf_class; // 32 or 64 bit
1954 char endianess; // MSB or LSB
1955 char* name; // String representation
1956 } arch_t;
1958 #ifndef EM_486
1959 #define EM_486 6 /* Intel 80486 */
1960 #endif
1962 static const arch_t arch_array[]={
1963 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1964 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1965 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1966 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1967 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1968 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1969 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1970 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1971 #if defined(VM_LITTLE_ENDIAN)
1972 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1973 #else
1974 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1975 #endif
1976 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1977 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1978 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1979 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1980 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1981 {EM_MIPS, EM_MIPS, ELFCLASS64, ELFDATA2LSB, (char*)"MIPS64 LE"},
1982 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1983 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1984 };
1986 #if (defined IA32)
1987 static Elf32_Half running_arch_code=EM_386;
1988 #elif (defined AMD64)
1989 static Elf32_Half running_arch_code=EM_X86_64;
1990 #elif (defined IA64)
1991 static Elf32_Half running_arch_code=EM_IA_64;
1992 #elif (defined __sparc) && (defined _LP64)
1993 static Elf32_Half running_arch_code=EM_SPARCV9;
1994 #elif (defined __sparc) && (!defined _LP64)
1995 static Elf32_Half running_arch_code=EM_SPARC;
1996 #elif (defined MIPS64)
1997 static Elf32_Half running_arch_code=EM_MIPS;
1998 #elif (defined __powerpc64__)
1999 static Elf32_Half running_arch_code=EM_PPC64;
2000 #elif (defined __powerpc__)
2001 static Elf32_Half running_arch_code=EM_PPC;
2002 #elif (defined ARM)
2003 static Elf32_Half running_arch_code=EM_ARM;
2004 #elif (defined S390)
2005 static Elf32_Half running_arch_code=EM_S390;
2006 #elif (defined ALPHA)
2007 static Elf32_Half running_arch_code=EM_ALPHA;
2008 #elif (defined MIPSEL)
2009 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2010 #elif (defined PARISC)
2011 static Elf32_Half running_arch_code=EM_PARISC;
2012 #elif (defined MIPS)
2013 static Elf32_Half running_arch_code=EM_MIPS;
2014 #elif (defined M68K)
2015 static Elf32_Half running_arch_code=EM_68K;
2016 #else
2017 #error Method os::dll_load requires that one of following is defined:\
2018 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, __mips64, PARISC, M68K
2019 #endif
2021 // Identify compatability class for VM's architecture and library's architecture
2022 // Obtain string descriptions for architectures
2024 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2025 int running_arch_index=-1;
2027 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2028 if (running_arch_code == arch_array[i].code) {
2029 running_arch_index = i;
2030 }
2031 if (lib_arch.code == arch_array[i].code) {
2032 lib_arch.compat_class = arch_array[i].compat_class;
2033 lib_arch.name = arch_array[i].name;
2034 }
2035 }
2037 assert(running_arch_index != -1,
2038 "Didn't find running architecture code (running_arch_code) in arch_array");
2039 if (running_arch_index == -1) {
2040 // Even though running architecture detection failed
2041 // we may still continue with reporting dlerror() message
2042 return NULL;
2043 }
2045 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2046 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2047 return NULL;
2048 }
2050 #ifndef S390
2051 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2052 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2053 return NULL;
2054 }
2055 #endif // !S390
2057 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2058 if ( lib_arch.name!=NULL ) {
2059 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2060 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2061 lib_arch.name, arch_array[running_arch_index].name);
2062 } else {
2063 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2064 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2065 lib_arch.code,
2066 arch_array[running_arch_index].name);
2067 }
2068 }
2070 return NULL;
2071 }
2073 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2074 void * result = ::dlopen(filename, RTLD_LAZY);
2075 if (result == NULL) {
2076 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2077 ebuf[ebuflen-1] = '\0';
2078 }
2079 return result;
2080 }
2082 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2083 void * result = NULL;
2084 if (LoadExecStackDllInVMThread) {
2085 result = dlopen_helper(filename, ebuf, ebuflen);
2086 }
2088 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2089 // library that requires an executable stack, or which does not have this
2090 // stack attribute set, dlopen changes the stack attribute to executable. The
2091 // read protection of the guard pages gets lost.
2092 //
2093 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2094 // may have been queued at the same time.
2096 if (!_stack_is_executable) {
2097 JavaThread *jt = Threads::first();
2099 while (jt) {
2100 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2101 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2102 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2103 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2104 warning("Attempt to reguard stack yellow zone failed.");
2105 }
2106 }
2107 jt = jt->next();
2108 }
2109 }
2111 return result;
2112 }
2114 /*
2115 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2116 * chances are you might want to run the generated bits against glibc-2.0
2117 * libdl.so, so always use locking for any version of glibc.
2118 */
2119 void* os::dll_lookup(void* handle, const char* name) {
2120 pthread_mutex_lock(&dl_mutex);
2121 void* res = dlsym(handle, name);
2122 pthread_mutex_unlock(&dl_mutex);
2123 return res;
2124 }
2126 void* os::get_default_process_handle() {
2127 return (void*)::dlopen(NULL, RTLD_LAZY);
2128 }
2130 static bool _print_ascii_file(const char* filename, outputStream* st) {
2131 int fd = ::open(filename, O_RDONLY);
2132 if (fd == -1) {
2133 return false;
2134 }
2136 char buf[32];
2137 int bytes;
2138 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2139 st->print_raw(buf, bytes);
2140 }
2142 ::close(fd);
2144 return true;
2145 }
2147 void os::print_dll_info(outputStream *st) {
2148 st->print_cr("Dynamic libraries:");
2150 char fname[32];
2151 pid_t pid = os::Linux::gettid();
2153 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2155 if (!_print_ascii_file(fname, st)) {
2156 st->print("Can not get library information for pid = %d\n", pid);
2157 }
2158 }
2160 void os::print_os_info_brief(outputStream* st) {
2161 os::Linux::print_distro_info(st);
2163 os::Posix::print_uname_info(st);
2165 os::Linux::print_libversion_info(st);
2167 }
2169 void os::print_os_info(outputStream* st) {
2170 st->print("OS:");
2172 os::Linux::print_distro_info(st);
2174 os::Posix::print_uname_info(st);
2176 // Print warning if unsafe chroot environment detected
2177 if (unsafe_chroot_detected) {
2178 st->print("WARNING!! ");
2179 st->print_cr("%s", unstable_chroot_error);
2180 }
2182 os::Linux::print_libversion_info(st);
2184 os::Posix::print_rlimit_info(st);
2186 os::Posix::print_load_average(st);
2188 os::Linux::print_full_memory_info(st);
2190 os::Linux::print_container_info(st);
2191 }
2193 // Try to identify popular distros.
2194 // Most Linux distributions have a /etc/XXX-release file, which contains
2195 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2196 // file that also contains the OS version string. Some have more than one
2197 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2198 // /etc/redhat-release.), so the order is important.
2199 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2200 // their own specific XXX-release file as well as a redhat-release file.
2201 // Because of this the XXX-release file needs to be searched for before the
2202 // redhat-release file.
2203 // Since Red Hat has a lsb-release file that is not very descriptive the
2204 // search for redhat-release needs to be before lsb-release.
2205 // Since the lsb-release file is the new standard it needs to be searched
2206 // before the older style release files.
2207 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2208 // next to last resort. The os-release file is a new standard that contains
2209 // distribution information and the system-release file seems to be an old
2210 // standard that has been replaced by the lsb-release and os-release files.
2211 // Searching for the debian_version file is the last resort. It contains
2212 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2213 // "Debian " is printed before the contents of the debian_version file.
2214 void os::Linux::print_distro_info(outputStream* st) {
2215 if (!_print_ascii_file("/etc/oracle-release", st) &&
2216 !_print_ascii_file("/etc/mandriva-release", st) &&
2217 !_print_ascii_file("/etc/mandrake-release", st) &&
2218 !_print_ascii_file("/etc/sun-release", st) &&
2219 !_print_ascii_file("/etc/redhat-release", st) &&
2220 !_print_ascii_file("/etc/lsb-release", st) &&
2221 !_print_ascii_file("/etc/SuSE-release", st) &&
2222 !_print_ascii_file("/etc/turbolinux-release", st) &&
2223 !_print_ascii_file("/etc/gentoo-release", st) &&
2224 !_print_ascii_file("/etc/ltib-release", st) &&
2225 !_print_ascii_file("/etc/angstrom-version", st) &&
2226 !_print_ascii_file("/etc/system-release", st) &&
2227 !_print_ascii_file("/etc/os-release", st)) {
2229 if (file_exists("/etc/debian_version")) {
2230 st->print("Debian ");
2231 _print_ascii_file("/etc/debian_version", st);
2232 } else {
2233 st->print("Linux");
2234 }
2235 }
2236 st->cr();
2237 }
2239 void os::Linux::print_libversion_info(outputStream* st) {
2240 // libc, pthread
2241 st->print("libc:");
2242 st->print("%s ", os::Linux::glibc_version());
2243 st->print("%s ", os::Linux::libpthread_version());
2244 if (os::Linux::is_LinuxThreads()) {
2245 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2246 }
2247 st->cr();
2248 }
2250 void os::Linux::print_full_memory_info(outputStream* st) {
2251 st->print("\n/proc/meminfo:\n");
2252 _print_ascii_file("/proc/meminfo", st);
2253 st->cr();
2254 }
2256 void os::Linux::print_container_info(outputStream* st) {
2257 if (!OSContainer::is_containerized()) {
2258 return;
2259 }
2261 st->print("container (cgroup) information:\n");
2263 const char *p_ct = OSContainer::container_type();
2264 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2266 char *p = OSContainer::cpu_cpuset_cpus();
2267 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2268 free(p);
2270 p = OSContainer::cpu_cpuset_memory_nodes();
2271 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2272 free(p);
2274 int i = OSContainer::active_processor_count();
2275 if (i > 0) {
2276 st->print("active_processor_count: %d\n", i);
2277 } else {
2278 st->print("active_processor_count: failed\n");
2279 }
2281 i = OSContainer::cpu_quota();
2282 st->print("cpu_quota: %d\n", i);
2284 i = OSContainer::cpu_period();
2285 st->print("cpu_period: %d\n", i);
2287 i = OSContainer::cpu_shares();
2288 st->print("cpu_shares: %d\n", i);
2290 jlong j = OSContainer::memory_limit_in_bytes();
2291 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2293 j = OSContainer::memory_and_swap_limit_in_bytes();
2294 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2296 j = OSContainer::memory_soft_limit_in_bytes();
2297 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2299 j = OSContainer::OSContainer::memory_usage_in_bytes();
2300 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2302 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2303 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2304 st->cr();
2305 }
2307 void os::print_memory_info(outputStream* st) {
2309 st->print("Memory:");
2310 st->print(" %dk page", os::vm_page_size()>>10);
2312 // values in struct sysinfo are "unsigned long"
2313 struct sysinfo si;
2314 sysinfo(&si);
2316 st->print(", physical " UINT64_FORMAT "k",
2317 os::physical_memory() >> 10);
2318 st->print("(" UINT64_FORMAT "k free)",
2319 os::available_memory() >> 10);
2320 st->print(", swap " UINT64_FORMAT "k",
2321 ((jlong)si.totalswap * si.mem_unit) >> 10);
2322 st->print("(" UINT64_FORMAT "k free)",
2323 ((jlong)si.freeswap * si.mem_unit) >> 10);
2324 st->cr();
2325 }
2327 void os::pd_print_cpu_info(outputStream* st) {
2328 st->print("\n/proc/cpuinfo:\n");
2329 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2330 st->print(" <Not Available>");
2331 }
2332 st->cr();
2333 }
2335 void os::print_siginfo(outputStream* st, void* siginfo) {
2336 const siginfo_t* si = (const siginfo_t*)siginfo;
2338 os::Posix::print_siginfo_brief(st, si);
2339 #if INCLUDE_CDS
2340 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2341 UseSharedSpaces) {
2342 FileMapInfo* mapinfo = FileMapInfo::current_info();
2343 if (mapinfo->is_in_shared_space(si->si_addr)) {
2344 st->print("\n\nError accessing class data sharing archive." \
2345 " Mapped file inaccessible during execution, " \
2346 " possible disk/network problem.");
2347 }
2348 }
2349 #endif
2350 st->cr();
2351 }
2354 static void print_signal_handler(outputStream* st, int sig,
2355 char* buf, size_t buflen);
2357 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2358 st->print_cr("Signal Handlers:");
2359 print_signal_handler(st, SIGSEGV, buf, buflen);
2360 print_signal_handler(st, SIGBUS , buf, buflen);
2361 print_signal_handler(st, SIGFPE , buf, buflen);
2362 print_signal_handler(st, SIGPIPE, buf, buflen);
2363 print_signal_handler(st, SIGXFSZ, buf, buflen);
2364 print_signal_handler(st, SIGILL , buf, buflen);
2365 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2366 print_signal_handler(st, SR_signum, buf, buflen);
2367 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2368 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2369 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2370 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2371 #if defined(PPC64)
2372 print_signal_handler(st, SIGTRAP, buf, buflen);
2373 #endif
2374 }
2376 static char saved_jvm_path[MAXPATHLEN] = {0};
2378 // Find the full path to the current module, libjvm.so
2379 void os::jvm_path(char *buf, jint buflen) {
2380 // Error checking.
2381 if (buflen < MAXPATHLEN) {
2382 assert(false, "must use a large-enough buffer");
2383 buf[0] = '\0';
2384 return;
2385 }
2386 // Lazy resolve the path to current module.
2387 if (saved_jvm_path[0] != 0) {
2388 strcpy(buf, saved_jvm_path);
2389 return;
2390 }
2392 char dli_fname[MAXPATHLEN];
2393 bool ret = dll_address_to_library_name(
2394 CAST_FROM_FN_PTR(address, os::jvm_path),
2395 dli_fname, sizeof(dli_fname), NULL);
2396 assert(ret, "cannot locate libjvm");
2397 char *rp = NULL;
2398 if (ret && dli_fname[0] != '\0') {
2399 rp = realpath(dli_fname, buf);
2400 }
2401 if (rp == NULL)
2402 return;
2404 if (Arguments::created_by_gamma_launcher()) {
2405 // Support for the gamma launcher. Typical value for buf is
2406 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2407 // the right place in the string, then assume we are installed in a JDK and
2408 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2409 // up the path so it looks like libjvm.so is installed there (append a
2410 // fake suffix hotspot/libjvm.so).
2411 const char *p = buf + strlen(buf) - 1;
2412 for (int count = 0; p > buf && count < 5; ++count) {
2413 for (--p; p > buf && *p != '/'; --p)
2414 /* empty */ ;
2415 }
2417 if (strncmp(p, "/jre/lib/", 9) != 0) {
2418 // Look for JAVA_HOME in the environment.
2419 char* java_home_var = ::getenv("JAVA_HOME");
2420 if (java_home_var != NULL && java_home_var[0] != 0) {
2421 char* jrelib_p;
2422 int len;
2424 // Check the current module name "libjvm.so".
2425 p = strrchr(buf, '/');
2426 assert(strstr(p, "/libjvm") == p, "invalid library name");
2428 rp = realpath(java_home_var, buf);
2429 if (rp == NULL)
2430 return;
2432 // determine if this is a legacy image or modules image
2433 // modules image doesn't have "jre" subdirectory
2434 len = strlen(buf);
2435 assert(len < buflen, "Ran out of buffer room");
2436 jrelib_p = buf + len;
2437 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2438 if (0 != access(buf, F_OK)) {
2439 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2440 }
2442 if (0 == access(buf, F_OK)) {
2443 // Use current module name "libjvm.so"
2444 len = strlen(buf);
2445 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2446 } else {
2447 // Go back to path of .so
2448 rp = realpath(dli_fname, buf);
2449 if (rp == NULL)
2450 return;
2451 }
2452 }
2453 }
2454 }
2456 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2457 }
2459 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2460 // no prefix required, not even "_"
2461 }
2463 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2464 // no suffix required
2465 }
2467 ////////////////////////////////////////////////////////////////////////////////
2468 // sun.misc.Signal support
2470 static volatile jint sigint_count = 0;
2472 static void
2473 UserHandler(int sig, void *siginfo, void *context) {
2474 // 4511530 - sem_post is serialized and handled by the manager thread. When
2475 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2476 // don't want to flood the manager thread with sem_post requests.
2477 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2478 return;
2480 // Ctrl-C is pressed during error reporting, likely because the error
2481 // handler fails to abort. Let VM die immediately.
2482 if (sig == SIGINT && is_error_reported()) {
2483 os::die();
2484 }
2486 os::signal_notify(sig);
2487 }
2489 void* os::user_handler() {
2490 return CAST_FROM_FN_PTR(void*, UserHandler);
2491 }
2493 class Semaphore : public StackObj {
2494 public:
2495 Semaphore();
2496 ~Semaphore();
2497 void signal();
2498 void wait();
2499 bool trywait();
2500 bool timedwait(unsigned int sec, int nsec);
2501 private:
2502 sem_t _semaphore;
2503 };
2505 Semaphore::Semaphore() {
2506 sem_init(&_semaphore, 0, 0);
2507 }
2509 Semaphore::~Semaphore() {
2510 sem_destroy(&_semaphore);
2511 }
2513 void Semaphore::signal() {
2514 sem_post(&_semaphore);
2515 }
2517 void Semaphore::wait() {
2518 sem_wait(&_semaphore);
2519 }
2521 bool Semaphore::trywait() {
2522 return sem_trywait(&_semaphore) == 0;
2523 }
2525 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2527 struct timespec ts;
2528 // Semaphore's are always associated with CLOCK_REALTIME
2529 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2530 // see unpackTime for discussion on overflow checking
2531 if (sec >= MAX_SECS) {
2532 ts.tv_sec += MAX_SECS;
2533 ts.tv_nsec = 0;
2534 } else {
2535 ts.tv_sec += sec;
2536 ts.tv_nsec += nsec;
2537 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2538 ts.tv_nsec -= NANOSECS_PER_SEC;
2539 ++ts.tv_sec; // note: this must be <= max_secs
2540 }
2541 }
2543 while (1) {
2544 int result = sem_timedwait(&_semaphore, &ts);
2545 if (result == 0) {
2546 return true;
2547 } else if (errno == EINTR) {
2548 continue;
2549 } else if (errno == ETIMEDOUT) {
2550 return false;
2551 } else {
2552 return false;
2553 }
2554 }
2555 }
2557 extern "C" {
2558 typedef void (*sa_handler_t)(int);
2559 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2560 }
2562 void* os::signal(int signal_number, void* handler) {
2563 struct sigaction sigAct, oldSigAct;
2565 sigfillset(&(sigAct.sa_mask));
2566 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2567 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2569 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2570 // -1 means registration failed
2571 return (void *)-1;
2572 }
2574 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2575 }
2577 void os::signal_raise(int signal_number) {
2578 ::raise(signal_number);
2579 }
2581 /*
2582 * The following code is moved from os.cpp for making this
2583 * code platform specific, which it is by its very nature.
2584 */
2586 // Will be modified when max signal is changed to be dynamic
2587 int os::sigexitnum_pd() {
2588 return NSIG;
2589 }
2591 // a counter for each possible signal value
2592 static volatile jint pending_signals[NSIG+1] = { 0 };
2594 // Linux(POSIX) specific hand shaking semaphore.
2595 static sem_t sig_sem;
2596 static Semaphore sr_semaphore;
2598 void os::signal_init_pd() {
2599 // Initialize signal structures
2600 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2602 // Initialize signal semaphore
2603 ::sem_init(&sig_sem, 0, 0);
2604 }
2606 void os::signal_notify(int sig) {
2607 Atomic::inc(&pending_signals[sig]);
2608 ::sem_post(&sig_sem);
2609 }
2611 static int check_pending_signals(bool wait) {
2612 Atomic::store(0, &sigint_count);
2613 for (;;) {
2614 for (int i = 0; i < NSIG + 1; i++) {
2615 jint n = pending_signals[i];
2616 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2617 return i;
2618 }
2619 }
2620 if (!wait) {
2621 return -1;
2622 }
2623 JavaThread *thread = JavaThread::current();
2624 ThreadBlockInVM tbivm(thread);
2626 bool threadIsSuspended;
2627 do {
2628 thread->set_suspend_equivalent();
2629 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2630 ::sem_wait(&sig_sem);
2632 // were we externally suspended while we were waiting?
2633 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2634 if (threadIsSuspended) {
2635 //
2636 // The semaphore has been incremented, but while we were waiting
2637 // another thread suspended us. We don't want to continue running
2638 // while suspended because that would surprise the thread that
2639 // suspended us.
2640 //
2641 ::sem_post(&sig_sem);
2643 thread->java_suspend_self();
2644 }
2645 } while (threadIsSuspended);
2646 }
2647 }
2649 int os::signal_lookup() {
2650 return check_pending_signals(false);
2651 }
2653 int os::signal_wait() {
2654 return check_pending_signals(true);
2655 }
2657 ////////////////////////////////////////////////////////////////////////////////
2658 // Virtual Memory
2660 int os::vm_page_size() {
2661 // Seems redundant as all get out
2662 assert(os::Linux::page_size() != -1, "must call os::init");
2663 return os::Linux::page_size();
2664 }
2666 // Solaris allocates memory by pages.
2667 int os::vm_allocation_granularity() {
2668 assert(os::Linux::page_size() != -1, "must call os::init");
2669 return os::Linux::page_size();
2670 }
2672 // Rationale behind this function:
2673 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2674 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2675 // samples for JITted code. Here we create private executable mapping over the code cache
2676 // and then we can use standard (well, almost, as mapping can change) way to provide
2677 // info for the reporting script by storing timestamp and location of symbol
2678 void linux_wrap_code(char* base, size_t size) {
2679 static volatile jint cnt = 0;
2681 if (!UseOprofile) {
2682 return;
2683 }
2685 char buf[PATH_MAX+1];
2686 int num = Atomic::add(1, &cnt);
2688 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2689 os::get_temp_directory(), os::current_process_id(), num);
2690 unlink(buf);
2692 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2694 if (fd != -1) {
2695 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2696 if (rv != (off_t)-1) {
2697 if (::write(fd, "", 1) == 1) {
2698 mmap(base, size,
2699 PROT_READ|PROT_WRITE|PROT_EXEC,
2700 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2701 }
2702 }
2703 ::close(fd);
2704 unlink(buf);
2705 }
2706 }
2708 static bool recoverable_mmap_error(int err) {
2709 // See if the error is one we can let the caller handle. This
2710 // list of errno values comes from JBS-6843484. I can't find a
2711 // Linux man page that documents this specific set of errno
2712 // values so while this list currently matches Solaris, it may
2713 // change as we gain experience with this failure mode.
2714 switch (err) {
2715 case EBADF:
2716 case EINVAL:
2717 case ENOTSUP:
2718 // let the caller deal with these errors
2719 return true;
2721 default:
2722 // Any remaining errors on this OS can cause our reserved mapping
2723 // to be lost. That can cause confusion where different data
2724 // structures think they have the same memory mapped. The worst
2725 // scenario is if both the VM and a library think they have the
2726 // same memory mapped.
2727 return false;
2728 }
2729 }
2731 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2732 int err) {
2733 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2734 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2735 strerror(err), err);
2736 }
2738 static void warn_fail_commit_memory(char* addr, size_t size,
2739 size_t alignment_hint, bool exec,
2740 int err) {
2741 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2742 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2743 alignment_hint, exec, strerror(err), err);
2744 }
2746 // NOTE: Linux kernel does not really reserve the pages for us.
2747 // All it does is to check if there are enough free pages
2748 // left at the time of mmap(). This could be a potential
2749 // problem.
2750 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2751 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2752 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2753 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2754 if (res != (uintptr_t) MAP_FAILED) {
2755 if (UseNUMAInterleaving) {
2756 numa_make_global(addr, size);
2757 }
2758 return 0;
2759 }
2761 int err = errno; // save errno from mmap() call above
2763 if (!recoverable_mmap_error(err)) {
2764 warn_fail_commit_memory(addr, size, exec, err);
2765 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2766 }
2768 return err;
2769 }
2771 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2772 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2773 }
2775 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2776 const char* mesg) {
2777 assert(mesg != NULL, "mesg must be specified");
2778 int err = os::Linux::commit_memory_impl(addr, size, exec);
2779 if (err != 0) {
2780 // the caller wants all commit errors to exit with the specified mesg:
2781 warn_fail_commit_memory(addr, size, exec, err);
2782 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2783 }
2784 }
2786 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2787 #ifndef MAP_HUGETLB
2788 #define MAP_HUGETLB 0x40000
2789 #endif
2791 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2792 #ifndef MADV_HUGEPAGE
2793 #define MADV_HUGEPAGE 14
2794 #endif
2796 int os::Linux::commit_memory_impl(char* addr, size_t size,
2797 size_t alignment_hint, bool exec) {
2798 int err = os::Linux::commit_memory_impl(addr, size, exec);
2799 if (err == 0) {
2800 realign_memory(addr, size, alignment_hint);
2801 }
2802 return err;
2803 }
2805 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2806 bool exec) {
2807 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2808 }
2810 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2811 size_t alignment_hint, bool exec,
2812 const char* mesg) {
2813 assert(mesg != NULL, "mesg must be specified");
2814 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2815 if (err != 0) {
2816 // the caller wants all commit errors to exit with the specified mesg:
2817 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2818 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2819 }
2820 }
2822 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2823 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2824 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2825 // be supported or the memory may already be backed by huge pages.
2826 ::madvise(addr, bytes, MADV_HUGEPAGE);
2827 }
2828 }
2830 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2831 // This method works by doing an mmap over an existing mmaping and effectively discarding
2832 // the existing pages. However it won't work for SHM-based large pages that cannot be
2833 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2834 // small pages on top of the SHM segment. This method always works for small pages, so we
2835 // allow that in any case.
2836 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2837 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2838 }
2839 }
2841 void os::numa_make_global(char *addr, size_t bytes) {
2842 Linux::numa_interleave_memory(addr, bytes);
2843 }
2845 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2846 // bind policy to MPOL_PREFERRED for the current thread.
2847 #define USE_MPOL_PREFERRED 0
2849 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2850 // To make NUMA and large pages more robust when both enabled, we need to ease
2851 // the requirements on where the memory should be allocated. MPOL_BIND is the
2852 // default policy and it will force memory to be allocated on the specified
2853 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2854 // the specified node, but will not force it. Using this policy will prevent
2855 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2856 // free large pages.
2857 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2858 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2859 }
2861 bool os::numa_topology_changed() { return false; }
2863 size_t os::numa_get_groups_num() {
2864 // Return just the number of nodes in which it's possible to allocate memory
2865 // (in numa terminology, configured nodes).
2866 return Linux::numa_num_configured_nodes();
2867 }
2869 int os::numa_get_group_id() {
2870 int cpu_id = Linux::sched_getcpu();
2871 if (cpu_id != -1) {
2872 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2873 if (lgrp_id != -1) {
2874 return lgrp_id;
2875 }
2876 }
2877 return 0;
2878 }
2880 int os::Linux::get_existing_num_nodes() {
2881 size_t node;
2882 size_t highest_node_number = Linux::numa_max_node();
2883 int num_nodes = 0;
2885 // Get the total number of nodes in the system including nodes without memory.
2886 for (node = 0; node <= highest_node_number; node++) {
2887 if (isnode_in_existing_nodes(node)) {
2888 num_nodes++;
2889 }
2890 }
2891 return num_nodes;
2892 }
2894 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2895 size_t highest_node_number = Linux::numa_max_node();
2896 size_t i = 0;
2898 // Map all node ids in which is possible to allocate memory. Also nodes are
2899 // not always consecutively available, i.e. available from 0 to the highest
2900 // node number.
2901 for (size_t node = 0; node <= highest_node_number; node++) {
2902 if (Linux::isnode_in_configured_nodes(node)) {
2903 ids[i++] = node;
2904 }
2905 }
2906 return i;
2907 }
2909 bool os::get_page_info(char *start, page_info* info) {
2910 return false;
2911 }
2913 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2914 return end;
2915 }
2918 int os::Linux::sched_getcpu_syscall(void) {
2919 unsigned int cpu = 0;
2920 int retval = -1;
2922 #if defined(IA32)
2923 # ifndef SYS_getcpu
2924 # define SYS_getcpu 318
2925 # endif
2926 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2927 #elif defined(AMD64)
2928 // Unfortunately we have to bring all these macros here from vsyscall.h
2929 // to be able to compile on old linuxes.
2930 # define __NR_vgetcpu 2
2931 # define VSYSCALL_START (-10UL << 20)
2932 # define VSYSCALL_SIZE 1024
2933 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2934 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2935 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2936 retval = vgetcpu(&cpu, NULL, NULL);
2937 #endif
2939 return (retval == -1) ? retval : cpu;
2940 }
2942 // Something to do with the numa-aware allocator needs these symbols
2943 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2944 extern "C" JNIEXPORT void numa_error(char *where) { }
2945 extern "C" JNIEXPORT int fork1() { return fork(); }
2947 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2948 // load symbol from base version instead.
2949 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2950 void *f = dlvsym(handle, name, "libnuma_1.1");
2951 if (f == NULL) {
2952 f = dlsym(handle, name);
2953 }
2954 return f;
2955 }
2957 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2958 // Return NULL if the symbol is not defined in this particular version.
2959 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2960 return dlvsym(handle, name, "libnuma_1.2");
2961 }
2963 bool os::Linux::libnuma_init() {
2964 // sched_getcpu() should be in libc.
2965 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2966 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2968 // If it's not, try a direct syscall.
2969 if (sched_getcpu() == -1)
2970 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2972 if (sched_getcpu() != -1) { // Does it work?
2973 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2974 if (handle != NULL) {
2975 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2976 libnuma_dlsym(handle, "numa_node_to_cpus")));
2977 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2978 libnuma_dlsym(handle, "numa_max_node")));
2979 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2980 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2981 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2982 libnuma_dlsym(handle, "numa_available")));
2983 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2984 libnuma_dlsym(handle, "numa_tonode_memory")));
2985 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2986 libnuma_dlsym(handle, "numa_interleave_memory")));
2987 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2988 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2989 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2990 libnuma_dlsym(handle, "numa_set_bind_policy")));
2991 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2992 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2993 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2994 libnuma_dlsym(handle, "numa_distance")));
2996 if (numa_available() != -1) {
2997 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2998 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2999 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3000 // Create an index -> node mapping, since nodes are not always consecutive
3001 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3002 rebuild_nindex_to_node_map();
3003 // Create a cpu -> node mapping
3004 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3005 rebuild_cpu_to_node_map();
3006 return true;
3007 }
3008 }
3009 }
3010 return false;
3011 }
3013 void os::Linux::rebuild_nindex_to_node_map() {
3014 int highest_node_number = Linux::numa_max_node();
3016 nindex_to_node()->clear();
3017 for (int node = 0; node <= highest_node_number; node++) {
3018 if (Linux::isnode_in_existing_nodes(node)) {
3019 nindex_to_node()->append(node);
3020 }
3021 }
3022 }
3024 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3025 // The table is later used in get_node_by_cpu().
3026 void os::Linux::rebuild_cpu_to_node_map() {
3027 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3028 // in libnuma (possible values are starting from 16,
3029 // and continuing up with every other power of 2, but less
3030 // than the maximum number of CPUs supported by kernel), and
3031 // is a subject to change (in libnuma version 2 the requirements
3032 // are more reasonable) we'll just hardcode the number they use
3033 // in the library.
3034 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3036 size_t cpu_num = processor_count();
3037 size_t cpu_map_size = NCPUS / BitsPerCLong;
3038 size_t cpu_map_valid_size =
3039 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3041 cpu_to_node()->clear();
3042 cpu_to_node()->at_grow(cpu_num - 1);
3044 size_t node_num = get_existing_num_nodes();
3046 int distance = 0;
3047 int closest_distance = INT_MAX;
3048 int closest_node = 0;
3049 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3050 for (size_t i = 0; i < node_num; i++) {
3051 // Check if node is configured (not a memory-less node). If it is not, find
3052 // the closest configured node.
3053 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3054 closest_distance = INT_MAX;
3055 // Check distance from all remaining nodes in the system. Ignore distance
3056 // from itself and from another non-configured node.
3057 for (size_t m = 0; m < node_num; m++) {
3058 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3059 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3060 // If a closest node is found, update. There is always at least one
3061 // configured node in the system so there is always at least one node
3062 // close.
3063 if (distance != 0 && distance < closest_distance) {
3064 closest_distance = distance;
3065 closest_node = nindex_to_node()->at(m);
3066 }
3067 }
3068 }
3069 } else {
3070 // Current node is already a configured node.
3071 closest_node = nindex_to_node()->at(i);
3072 }
3074 // Get cpus from the original node and map them to the closest node. If node
3075 // is a configured node (not a memory-less node), then original node and
3076 // closest node are the same.
3077 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3078 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3079 if (cpu_map[j] != 0) {
3080 for (size_t k = 0; k < BitsPerCLong; k++) {
3081 if (cpu_map[j] & (1UL << k)) {
3082 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3083 }
3084 }
3085 }
3086 }
3087 }
3088 }
3089 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
3090 }
3092 int os::Linux::get_node_by_cpu(int cpu_id) {
3093 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3094 return cpu_to_node()->at(cpu_id);
3095 }
3096 return -1;
3097 }
3099 GrowableArray<int>* os::Linux::_cpu_to_node;
3100 GrowableArray<int>* os::Linux::_nindex_to_node;
3101 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3102 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3103 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3104 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3105 os::Linux::numa_available_func_t os::Linux::_numa_available;
3106 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3107 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3108 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3109 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3110 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3111 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3112 unsigned long* os::Linux::_numa_all_nodes;
3113 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3114 struct bitmask* os::Linux::_numa_nodes_ptr;
3116 bool os::pd_uncommit_memory(char* addr, size_t size) {
3117 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3118 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3119 return res != (uintptr_t) MAP_FAILED;
3120 }
3122 static
3123 address get_stack_commited_bottom(address bottom, size_t size) {
3124 address nbot = bottom;
3125 address ntop = bottom + size;
3127 size_t page_sz = os::vm_page_size();
3128 unsigned pages = size / page_sz;
3130 unsigned char vec[1];
3131 unsigned imin = 1, imax = pages + 1, imid;
3132 int mincore_return_value = 0;
3134 assert(imin <= imax, "Unexpected page size");
3136 while (imin < imax) {
3137 imid = (imax + imin) / 2;
3138 nbot = ntop - (imid * page_sz);
3140 // Use a trick with mincore to check whether the page is mapped or not.
3141 // mincore sets vec to 1 if page resides in memory and to 0 if page
3142 // is swapped output but if page we are asking for is unmapped
3143 // it returns -1,ENOMEM
3144 mincore_return_value = mincore(nbot, page_sz, vec);
3146 if (mincore_return_value == -1) {
3147 // Page is not mapped go up
3148 // to find first mapped page
3149 if (errno != EAGAIN) {
3150 assert(errno == ENOMEM, "Unexpected mincore errno");
3151 imax = imid;
3152 }
3153 } else {
3154 // Page is mapped go down
3155 // to find first not mapped page
3156 imin = imid + 1;
3157 }
3158 }
3160 nbot = nbot + page_sz;
3162 // Adjust stack bottom one page up if last checked page is not mapped
3163 if (mincore_return_value == -1) {
3164 nbot = nbot + page_sz;
3165 }
3167 return nbot;
3168 }
3171 // Linux uses a growable mapping for the stack, and if the mapping for
3172 // the stack guard pages is not removed when we detach a thread the
3173 // stack cannot grow beyond the pages where the stack guard was
3174 // mapped. If at some point later in the process the stack expands to
3175 // that point, the Linux kernel cannot expand the stack any further
3176 // because the guard pages are in the way, and a segfault occurs.
3177 //
3178 // However, it's essential not to split the stack region by unmapping
3179 // a region (leaving a hole) that's already part of the stack mapping,
3180 // so if the stack mapping has already grown beyond the guard pages at
3181 // the time we create them, we have to truncate the stack mapping.
3182 // So, we need to know the extent of the stack mapping when
3183 // create_stack_guard_pages() is called.
3185 // We only need this for stacks that are growable: at the time of
3186 // writing thread stacks don't use growable mappings (i.e. those
3187 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3188 // only applies to the main thread.
3190 // If the (growable) stack mapping already extends beyond the point
3191 // where we're going to put our guard pages, truncate the mapping at
3192 // that point by munmap()ping it. This ensures that when we later
3193 // munmap() the guard pages we don't leave a hole in the stack
3194 // mapping. This only affects the main/primordial thread
3196 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3198 if (os::is_primordial_thread()) {
3199 // As we manually grow stack up to bottom inside create_attached_thread(),
3200 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3201 // we don't need to do anything special.
3202 // Check it first, before calling heavy function.
3203 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3204 unsigned char vec[1];
3206 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3207 // Fallback to slow path on all errors, including EAGAIN
3208 stack_extent = (uintptr_t) get_stack_commited_bottom(
3209 os::Linux::initial_thread_stack_bottom(),
3210 (size_t)addr - stack_extent);
3211 }
3213 if (stack_extent < (uintptr_t)addr) {
3214 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3215 }
3216 }
3218 return os::commit_memory(addr, size, !ExecMem);
3219 }
3221 // If this is a growable mapping, remove the guard pages entirely by
3222 // munmap()ping them. If not, just call uncommit_memory(). This only
3223 // affects the main/primordial thread, but guard against future OS changes.
3224 // It's safe to always unmap guard pages for primordial thread because we
3225 // always place it right after end of the mapped region.
3227 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3228 uintptr_t stack_extent, stack_base;
3230 if (os::is_primordial_thread()) {
3231 return ::munmap(addr, size) == 0;
3232 }
3234 return os::uncommit_memory(addr, size);
3235 }
3237 static address _highest_vm_reserved_address = NULL;
3239 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3240 // at 'requested_addr'. If there are existing memory mappings at the same
3241 // location, however, they will be overwritten. If 'fixed' is false,
3242 // 'requested_addr' is only treated as a hint, the return value may or
3243 // may not start from the requested address. Unlike Linux mmap(), this
3244 // function returns NULL to indicate failure.
3245 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3246 char * addr;
3247 int flags;
3249 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3250 if (fixed) {
3251 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3252 flags |= MAP_FIXED;
3253 }
3255 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3256 // touch an uncommitted page. Otherwise, the read/write might
3257 // succeed if we have enough swap space to back the physical page.
3258 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3259 flags, -1, 0);
3261 if (addr != MAP_FAILED) {
3262 // anon_mmap() should only get called during VM initialization,
3263 // don't need lock (actually we can skip locking even it can be called
3264 // from multiple threads, because _highest_vm_reserved_address is just a
3265 // hint about the upper limit of non-stack memory regions.)
3266 if ((address)addr + bytes > _highest_vm_reserved_address) {
3267 _highest_vm_reserved_address = (address)addr + bytes;
3268 }
3269 }
3271 return addr == MAP_FAILED ? NULL : addr;
3272 }
3274 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3275 // (req_addr != NULL) or with a given alignment.
3276 // - bytes shall be a multiple of alignment.
3277 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3278 // - alignment sets the alignment at which memory shall be allocated.
3279 // It must be a multiple of allocation granularity.
3280 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3281 // req_addr or NULL.
3282 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3284 size_t extra_size = bytes;
3285 if (req_addr == NULL && alignment > 0) {
3286 extra_size += alignment;
3287 }
3289 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3290 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3291 -1, 0);
3292 if (start == MAP_FAILED) {
3293 start = NULL;
3294 } else {
3295 if (req_addr != NULL) {
3296 if (start != req_addr) {
3297 ::munmap(start, extra_size);
3298 start = NULL;
3299 }
3300 } else {
3301 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3302 char* const end_aligned = start_aligned + bytes;
3303 char* const end = start + extra_size;
3304 if (start_aligned > start) {
3305 ::munmap(start, start_aligned - start);
3306 }
3307 if (end_aligned < end) {
3308 ::munmap(end_aligned, end - end_aligned);
3309 }
3310 start = start_aligned;
3311 }
3312 }
3313 return start;
3314 }
3316 // Don't update _highest_vm_reserved_address, because there might be memory
3317 // regions above addr + size. If so, releasing a memory region only creates
3318 // a hole in the address space, it doesn't help prevent heap-stack collision.
3319 //
3320 static int anon_munmap(char * addr, size_t size) {
3321 return ::munmap(addr, size) == 0;
3322 }
3324 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3325 size_t alignment_hint) {
3326 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3327 }
3329 bool os::pd_release_memory(char* addr, size_t size) {
3330 return anon_munmap(addr, size);
3331 }
3333 static address highest_vm_reserved_address() {
3334 return _highest_vm_reserved_address;
3335 }
3337 static bool linux_mprotect(char* addr, size_t size, int prot) {
3338 // Linux wants the mprotect address argument to be page aligned.
3339 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3341 // According to SUSv3, mprotect() should only be used with mappings
3342 // established by mmap(), and mmap() always maps whole pages. Unaligned
3343 // 'addr' likely indicates problem in the VM (e.g. trying to change
3344 // protection of malloc'ed or statically allocated memory). Check the
3345 // caller if you hit this assert.
3346 assert(addr == bottom, "sanity check");
3348 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3349 return ::mprotect(bottom, size, prot) == 0;
3350 }
3352 // Set protections specified
3353 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3354 bool is_committed) {
3355 unsigned int p = 0;
3356 switch (prot) {
3357 case MEM_PROT_NONE: p = PROT_NONE; break;
3358 case MEM_PROT_READ: p = PROT_READ; break;
3359 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3360 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3361 default:
3362 ShouldNotReachHere();
3363 }
3364 // is_committed is unused.
3365 return linux_mprotect(addr, bytes, p);
3366 }
3368 bool os::guard_memory(char* addr, size_t size) {
3369 return linux_mprotect(addr, size, PROT_NONE);
3370 }
3372 bool os::unguard_memory(char* addr, size_t size) {
3373 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3374 }
3376 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3377 bool result = false;
3378 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3379 MAP_ANONYMOUS|MAP_PRIVATE,
3380 -1, 0);
3381 if (p != MAP_FAILED) {
3382 void *aligned_p = align_ptr_up(p, page_size);
3384 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3386 munmap(p, page_size * 2);
3387 }
3389 if (warn && !result) {
3390 warning("TransparentHugePages is not supported by the operating system.");
3391 }
3393 return result;
3394 }
3396 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3397 bool result = false;
3398 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3399 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3400 -1, 0);
3402 if (p != MAP_FAILED) {
3403 // We don't know if this really is a huge page or not.
3404 FILE *fp = fopen("/proc/self/maps", "r");
3405 if (fp) {
3406 while (!feof(fp)) {
3407 char chars[257];
3408 long x = 0;
3409 if (fgets(chars, sizeof(chars), fp)) {
3410 if (sscanf(chars, "%lx-%*x", &x) == 1
3411 && x == (long)p) {
3412 if (strstr (chars, "hugepage")) {
3413 result = true;
3414 break;
3415 }
3416 }
3417 }
3418 }
3419 fclose(fp);
3420 }
3421 munmap(p, page_size);
3422 }
3424 if (warn && !result) {
3425 warning("HugeTLBFS is not supported by the operating system.");
3426 }
3428 return result;
3429 }
3431 /*
3432 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3433 *
3434 * From the coredump_filter documentation:
3435 *
3436 * - (bit 0) anonymous private memory
3437 * - (bit 1) anonymous shared memory
3438 * - (bit 2) file-backed private memory
3439 * - (bit 3) file-backed shared memory
3440 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3441 * effective only if the bit 2 is cleared)
3442 * - (bit 5) hugetlb private memory
3443 * - (bit 6) hugetlb shared memory
3444 */
3445 static void set_coredump_filter(void) {
3446 FILE *f;
3447 long cdm;
3449 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3450 return;
3451 }
3453 if (fscanf(f, "%lx", &cdm) != 1) {
3454 fclose(f);
3455 return;
3456 }
3458 rewind(f);
3460 if ((cdm & LARGEPAGES_BIT) == 0) {
3461 cdm |= LARGEPAGES_BIT;
3462 fprintf(f, "%#lx", cdm);
3463 }
3465 fclose(f);
3466 }
3468 // Large page support
3470 static size_t _large_page_size = 0;
3472 size_t os::Linux::find_large_page_size() {
3473 size_t large_page_size = 0;
3475 // large_page_size on Linux is used to round up heap size. x86 uses either
3476 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3477 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3478 // page as large as 256M.
3479 //
3480 // Here we try to figure out page size by parsing /proc/meminfo and looking
3481 // for a line with the following format:
3482 // Hugepagesize: 2048 kB
3483 //
3484 // If we can't determine the value (e.g. /proc is not mounted, or the text
3485 // format has been changed), we'll use the largest page size supported by
3486 // the processor.
3488 #ifndef ZERO
3489 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3490 ARM_ONLY(2 * M) PPC_ONLY(4 * M) MIPS64_ONLY(4 * M); //In MIPS _large_page_size is seted 4*M.
3491 #endif // ZERO
3493 FILE *fp = fopen("/proc/meminfo", "r");
3494 if (fp) {
3495 while (!feof(fp)) {
3496 int x = 0;
3497 char buf[16];
3498 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3499 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3500 large_page_size = x * K;
3501 break;
3502 }
3503 } else {
3504 // skip to next line
3505 for (;;) {
3506 int ch = fgetc(fp);
3507 if (ch == EOF || ch == (int)'\n') break;
3508 }
3509 }
3510 }
3511 fclose(fp);
3512 }
3514 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3515 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3516 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3517 proper_unit_for_byte_size(large_page_size));
3518 }
3520 return large_page_size;
3521 }
3523 size_t os::Linux::setup_large_page_size() {
3524 _large_page_size = Linux::find_large_page_size();
3525 const size_t default_page_size = (size_t)Linux::page_size();
3526 if (_large_page_size > default_page_size) {
3527 _page_sizes[0] = _large_page_size;
3528 _page_sizes[1] = default_page_size;
3529 _page_sizes[2] = 0;
3530 }
3532 return _large_page_size;
3533 }
3535 bool os::Linux::setup_large_page_type(size_t page_size) {
3536 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3537 FLAG_IS_DEFAULT(UseSHM) &&
3538 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3540 // The type of large pages has not been specified by the user.
3542 // Try UseHugeTLBFS and then UseSHM.
3543 UseHugeTLBFS = UseSHM = true;
3545 // Don't try UseTransparentHugePages since there are known
3546 // performance issues with it turned on. This might change in the future.
3547 UseTransparentHugePages = false;
3548 }
3550 if (UseTransparentHugePages) {
3551 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3552 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3553 UseHugeTLBFS = false;
3554 UseSHM = false;
3555 return true;
3556 }
3557 UseTransparentHugePages = false;
3558 }
3560 if (UseHugeTLBFS) {
3561 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3562 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3563 UseSHM = false;
3564 return true;
3565 }
3566 UseHugeTLBFS = false;
3567 }
3569 return UseSHM;
3570 }
3572 void os::large_page_init() {
3573 if (!UseLargePages &&
3574 !UseTransparentHugePages &&
3575 !UseHugeTLBFS &&
3576 !UseSHM) {
3577 // Not using large pages.
3578 return;
3579 }
3581 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3582 // The user explicitly turned off large pages.
3583 // Ignore the rest of the large pages flags.
3584 UseTransparentHugePages = false;
3585 UseHugeTLBFS = false;
3586 UseSHM = false;
3587 return;
3588 }
3590 size_t large_page_size = Linux::setup_large_page_size();
3591 UseLargePages = Linux::setup_large_page_type(large_page_size);
3593 set_coredump_filter();
3594 }
3596 #ifndef SHM_HUGETLB
3597 #define SHM_HUGETLB 04000
3598 #endif
3600 #define shm_warning_format(format, ...) \
3601 do { \
3602 if (UseLargePages && \
3603 (!FLAG_IS_DEFAULT(UseLargePages) || \
3604 !FLAG_IS_DEFAULT(UseSHM) || \
3605 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3606 warning(format, __VA_ARGS__); \
3607 } \
3608 } while (0)
3610 #define shm_warning(str) shm_warning_format("%s", str)
3612 #define shm_warning_with_errno(str) \
3613 do { \
3614 int err = errno; \
3615 shm_warning_format(str " (error = %d)", err); \
3616 } while (0)
3618 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3619 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3621 if (!is_size_aligned(alignment, SHMLBA)) {
3622 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3623 return NULL;
3624 }
3626 // To ensure that we get 'alignment' aligned memory from shmat,
3627 // we pre-reserve aligned virtual memory and then attach to that.
3629 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3630 if (pre_reserved_addr == NULL) {
3631 // Couldn't pre-reserve aligned memory.
3632 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3633 return NULL;
3634 }
3636 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3637 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3639 if ((intptr_t)addr == -1) {
3640 int err = errno;
3641 shm_warning_with_errno("Failed to attach shared memory.");
3643 assert(err != EACCES, "Unexpected error");
3644 assert(err != EIDRM, "Unexpected error");
3645 assert(err != EINVAL, "Unexpected error");
3647 // Since we don't know if the kernel unmapped the pre-reserved memory area
3648 // we can't unmap it, since that would potentially unmap memory that was
3649 // mapped from other threads.
3650 return NULL;
3651 }
3653 return addr;
3654 }
3656 static char* shmat_at_address(int shmid, char* req_addr) {
3657 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3658 assert(false, "Requested address needs to be SHMLBA aligned");
3659 return NULL;
3660 }
3662 char* addr = (char*)shmat(shmid, req_addr, 0);
3664 if ((intptr_t)addr == -1) {
3665 shm_warning_with_errno("Failed to attach shared memory.");
3666 return NULL;
3667 }
3669 return addr;
3670 }
3672 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3673 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3674 if (req_addr != NULL) {
3675 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3676 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3677 return shmat_at_address(shmid, req_addr);
3678 }
3680 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3681 // return large page size aligned memory addresses when req_addr == NULL.
3682 // However, if the alignment is larger than the large page size, we have
3683 // to manually ensure that the memory returned is 'alignment' aligned.
3684 if (alignment > os::large_page_size()) {
3685 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3686 return shmat_with_alignment(shmid, bytes, alignment);
3687 } else {
3688 return shmat_at_address(shmid, NULL);
3689 }
3690 }
3692 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3693 // "exec" is passed in but not used. Creating the shared image for
3694 // the code cache doesn't have an SHM_X executable permission to check.
3695 assert(UseLargePages && UseSHM, "only for SHM large pages");
3696 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3697 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3699 if (!is_size_aligned(bytes, os::large_page_size())) {
3700 return NULL; // Fallback to small pages.
3701 }
3703 // Create a large shared memory region to attach to based on size.
3704 // Currently, size is the total size of the heap.
3705 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3706 if (shmid == -1) {
3707 // Possible reasons for shmget failure:
3708 // 1. shmmax is too small for Java heap.
3709 // > check shmmax value: cat /proc/sys/kernel/shmmax
3710 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3711 // 2. not enough large page memory.
3712 // > check available large pages: cat /proc/meminfo
3713 // > increase amount of large pages:
3714 // echo new_value > /proc/sys/vm/nr_hugepages
3715 // Note 1: different Linux may use different name for this property,
3716 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3717 // Note 2: it's possible there's enough physical memory available but
3718 // they are so fragmented after a long run that they can't
3719 // coalesce into large pages. Try to reserve large pages when
3720 // the system is still "fresh".
3721 shm_warning_with_errno("Failed to reserve shared memory.");
3722 return NULL;
3723 }
3725 // Attach to the region.
3726 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3728 // Remove shmid. If shmat() is successful, the actual shared memory segment
3729 // will be deleted when it's detached by shmdt() or when the process
3730 // terminates. If shmat() is not successful this will remove the shared
3731 // segment immediately.
3732 shmctl(shmid, IPC_RMID, NULL);
3734 return addr;
3735 }
3737 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3738 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3740 bool warn_on_failure = UseLargePages &&
3741 (!FLAG_IS_DEFAULT(UseLargePages) ||
3742 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3743 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3745 if (warn_on_failure) {
3746 char msg[128];
3747 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3748 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3749 warning("%s", msg);
3750 }
3751 }
3753 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3754 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3755 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3756 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3758 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3759 char* addr = (char*)::mmap(req_addr, bytes, prot,
3760 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3761 -1, 0);
3763 if (addr == MAP_FAILED) {
3764 warn_on_large_pages_failure(req_addr, bytes, errno);
3765 return NULL;
3766 }
3768 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3770 return addr;
3771 }
3773 // Reserve memory using mmap(MAP_HUGETLB).
3774 // - bytes shall be a multiple of alignment.
3775 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3776 // - alignment sets the alignment at which memory shall be allocated.
3777 // It must be a multiple of allocation granularity.
3778 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3779 // req_addr or NULL.
3780 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3781 size_t large_page_size = os::large_page_size();
3782 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3784 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3785 assert(is_size_aligned(bytes, alignment), "Must be");
3787 // First reserve - but not commit - the address range in small pages.
3788 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3790 if (start == NULL) {
3791 return NULL;
3792 }
3794 assert(is_ptr_aligned(start, alignment), "Must be");
3796 char* end = start + bytes;
3798 // Find the regions of the allocated chunk that can be promoted to large pages.
3799 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3800 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3802 size_t lp_bytes = lp_end - lp_start;
3804 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3806 if (lp_bytes == 0) {
3807 // The mapped region doesn't even span the start and the end of a large page.
3808 // Fall back to allocate a non-special area.
3809 ::munmap(start, end - start);
3810 return NULL;
3811 }
3813 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3815 void* result;
3817 // Commit small-paged leading area.
3818 if (start != lp_start) {
3819 result = ::mmap(start, lp_start - start, prot,
3820 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3821 -1, 0);
3822 if (result == MAP_FAILED) {
3823 ::munmap(lp_start, end - lp_start);
3824 return NULL;
3825 }
3826 }
3828 // Commit large-paged area.
3829 result = ::mmap(lp_start, lp_bytes, prot,
3830 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3831 -1, 0);
3832 if (result == MAP_FAILED) {
3833 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3834 // If the mmap above fails, the large pages region will be unmapped and we
3835 // have regions before and after with small pages. Release these regions.
3836 //
3837 // | mapped | unmapped | mapped |
3838 // ^ ^ ^ ^
3839 // start lp_start lp_end end
3840 //
3841 ::munmap(start, lp_start - start);
3842 ::munmap(lp_end, end - lp_end);
3843 return NULL;
3844 }
3846 // Commit small-paged trailing area.
3847 if (lp_end != end) {
3848 result = ::mmap(lp_end, end - lp_end, prot,
3849 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3850 -1, 0);
3851 if (result == MAP_FAILED) {
3852 ::munmap(start, lp_end - start);
3853 return NULL;
3854 }
3855 }
3857 return start;
3858 }
3860 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3861 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3862 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3863 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3864 assert(is_power_of_2(os::large_page_size()), "Must be");
3865 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3867 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3868 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3869 } else {
3870 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3871 }
3872 }
3874 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3875 assert(UseLargePages, "only for large pages");
3877 char* addr;
3878 if (UseSHM) {
3879 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3880 } else {
3881 assert(UseHugeTLBFS, "must be");
3882 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3883 }
3885 if (addr != NULL) {
3886 if (UseNUMAInterleaving) {
3887 numa_make_global(addr, bytes);
3888 }
3890 // The memory is committed
3891 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3892 }
3894 return addr;
3895 }
3897 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3898 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3899 return shmdt(base) == 0;
3900 }
3902 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3903 return pd_release_memory(base, bytes);
3904 }
3906 bool os::release_memory_special(char* base, size_t bytes) {
3907 bool res;
3908 if (MemTracker::tracking_level() > NMT_minimal) {
3909 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3910 res = os::Linux::release_memory_special_impl(base, bytes);
3911 if (res) {
3912 tkr.record((address)base, bytes);
3913 }
3915 } else {
3916 res = os::Linux::release_memory_special_impl(base, bytes);
3917 }
3918 return res;
3919 }
3921 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3922 assert(UseLargePages, "only for large pages");
3923 bool res;
3925 if (UseSHM) {
3926 res = os::Linux::release_memory_special_shm(base, bytes);
3927 } else {
3928 assert(UseHugeTLBFS, "must be");
3929 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3930 }
3931 return res;
3932 }
3934 size_t os::large_page_size() {
3935 return _large_page_size;
3936 }
3938 // With SysV SHM the entire memory region must be allocated as shared
3939 // memory.
3940 // HugeTLBFS allows application to commit large page memory on demand.
3941 // However, when committing memory with HugeTLBFS fails, the region
3942 // that was supposed to be committed will lose the old reservation
3943 // and allow other threads to steal that memory region. Because of this
3944 // behavior we can't commit HugeTLBFS memory.
3945 bool os::can_commit_large_page_memory() {
3946 return UseTransparentHugePages;
3947 }
3949 bool os::can_execute_large_page_memory() {
3950 return UseTransparentHugePages || UseHugeTLBFS;
3951 }
3953 // Reserve memory at an arbitrary address, only if that area is
3954 // available (and not reserved for something else).
3956 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3957 const int max_tries = 10;
3958 char* base[max_tries];
3959 size_t size[max_tries];
3960 const size_t gap = 0x000000;
3962 // Assert only that the size is a multiple of the page size, since
3963 // that's all that mmap requires, and since that's all we really know
3964 // about at this low abstraction level. If we need higher alignment,
3965 // we can either pass an alignment to this method or verify alignment
3966 // in one of the methods further up the call chain. See bug 5044738.
3967 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3969 // Repeatedly allocate blocks until the block is allocated at the
3970 // right spot. Give up after max_tries. Note that reserve_memory() will
3971 // automatically update _highest_vm_reserved_address if the call is
3972 // successful. The variable tracks the highest memory address every reserved
3973 // by JVM. It is used to detect heap-stack collision if running with
3974 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3975 // space than needed, it could confuse the collision detecting code. To
3976 // solve the problem, save current _highest_vm_reserved_address and
3977 // calculate the correct value before return.
3978 address old_highest = _highest_vm_reserved_address;
3980 // Linux mmap allows caller to pass an address as hint; give it a try first,
3981 // if kernel honors the hint then we can return immediately.
3982 char * addr = anon_mmap(requested_addr, bytes, false);
3983 if (addr == requested_addr) {
3984 return requested_addr;
3985 }
3987 if (addr != NULL) {
3988 // mmap() is successful but it fails to reserve at the requested address
3989 anon_munmap(addr, bytes);
3990 }
3992 int i;
3993 for (i = 0; i < max_tries; ++i) {
3994 base[i] = reserve_memory(bytes);
3996 if (base[i] != NULL) {
3997 // Is this the block we wanted?
3998 if (base[i] == requested_addr) {
3999 size[i] = bytes;
4000 break;
4001 }
4003 // Does this overlap the block we wanted? Give back the overlapped
4004 // parts and try again.
4006 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
4007 if (top_overlap >= 0 && top_overlap < bytes) {
4008 unmap_memory(base[i], top_overlap);
4009 base[i] += top_overlap;
4010 size[i] = bytes - top_overlap;
4011 } else {
4012 size_t bottom_overlap = base[i] + bytes - requested_addr;
4013 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
4014 unmap_memory(requested_addr, bottom_overlap);
4015 size[i] = bytes - bottom_overlap;
4016 } else {
4017 size[i] = bytes;
4018 }
4019 }
4020 }
4021 }
4023 // Give back the unused reserved pieces.
4025 for (int j = 0; j < i; ++j) {
4026 if (base[j] != NULL) {
4027 unmap_memory(base[j], size[j]);
4028 }
4029 }
4031 if (i < max_tries) {
4032 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
4033 return requested_addr;
4034 } else {
4035 _highest_vm_reserved_address = old_highest;
4036 return NULL;
4037 }
4038 }
4040 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4041 return ::read(fd, buf, nBytes);
4042 }
4044 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
4045 // Solaris uses poll(), linux uses park().
4046 // Poll() is likely a better choice, assuming that Thread.interrupt()
4047 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
4048 // SIGSEGV, see 4355769.
4050 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
4051 assert(thread == Thread::current(), "thread consistency check");
4053 ParkEvent * const slp = thread->_SleepEvent ;
4054 slp->reset() ;
4055 OrderAccess::fence() ;
4057 if (interruptible) {
4058 jlong prevtime = javaTimeNanos();
4060 for (;;) {
4061 if (os::is_interrupted(thread, true)) {
4062 return OS_INTRPT;
4063 }
4065 jlong newtime = javaTimeNanos();
4067 if (newtime - prevtime < 0) {
4068 // time moving backwards, should only happen if no monotonic clock
4069 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4070 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4071 } else {
4072 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4073 }
4075 if(millis <= 0) {
4076 return OS_OK;
4077 }
4079 prevtime = newtime;
4081 {
4082 assert(thread->is_Java_thread(), "sanity check");
4083 JavaThread *jt = (JavaThread *) thread;
4084 ThreadBlockInVM tbivm(jt);
4085 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
4087 jt->set_suspend_equivalent();
4088 // cleared by handle_special_suspend_equivalent_condition() or
4089 // java_suspend_self() via check_and_wait_while_suspended()
4091 slp->park(millis);
4093 // were we externally suspended while we were waiting?
4094 jt->check_and_wait_while_suspended();
4095 }
4096 }
4097 } else {
4098 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4099 jlong prevtime = javaTimeNanos();
4101 for (;;) {
4102 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
4103 // the 1st iteration ...
4104 jlong newtime = javaTimeNanos();
4106 if (newtime - prevtime < 0) {
4107 // time moving backwards, should only happen if no monotonic clock
4108 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4109 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4110 } else {
4111 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4112 }
4114 if(millis <= 0) break ;
4116 prevtime = newtime;
4117 slp->park(millis);
4118 }
4119 return OS_OK ;
4120 }
4121 }
4123 //
4124 // Short sleep, direct OS call.
4125 //
4126 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4127 // sched_yield(2) will actually give up the CPU:
4128 //
4129 // * Alone on this pariticular CPU, keeps running.
4130 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4131 // (pre 2.6.39).
4132 //
4133 // So calling this with 0 is an alternative.
4134 //
4135 void os::naked_short_sleep(jlong ms) {
4136 struct timespec req;
4138 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4139 req.tv_sec = 0;
4140 if (ms > 0) {
4141 req.tv_nsec = (ms % 1000) * 1000000;
4142 }
4143 else {
4144 req.tv_nsec = 1;
4145 }
4147 nanosleep(&req, NULL);
4149 return;
4150 }
4152 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4153 void os::infinite_sleep() {
4154 while (true) { // sleep forever ...
4155 ::sleep(100); // ... 100 seconds at a time
4156 }
4157 }
4159 // Used to convert frequent JVM_Yield() to nops
4160 bool os::dont_yield() {
4161 return DontYieldALot;
4162 }
4164 void os::yield() {
4165 sched_yield();
4166 }
4168 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
4170 void os::yield_all(int attempts) {
4171 // Yields to all threads, including threads with lower priorities
4172 // Threads on Linux are all with same priority. The Solaris style
4173 // os::yield_all() with nanosleep(1ms) is not necessary.
4174 sched_yield();
4175 }
4177 // Called from the tight loops to possibly influence time-sharing heuristics
4178 void os::loop_breaker(int attempts) {
4179 os::yield_all(attempts);
4180 }
4182 ////////////////////////////////////////////////////////////////////////////////
4183 // thread priority support
4185 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4186 // only supports dynamic priority, static priority must be zero. For real-time
4187 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4188 // However, for large multi-threaded applications, SCHED_RR is not only slower
4189 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4190 // of 5 runs - Sep 2005).
4191 //
4192 // The following code actually changes the niceness of kernel-thread/LWP. It
4193 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4194 // not the entire user process, and user level threads are 1:1 mapped to kernel
4195 // threads. It has always been the case, but could change in the future. For
4196 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4197 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4199 int os::java_to_os_priority[CriticalPriority + 1] = {
4200 19, // 0 Entry should never be used
4202 4, // 1 MinPriority
4203 3, // 2
4204 2, // 3
4206 1, // 4
4207 0, // 5 NormPriority
4208 -1, // 6
4210 -2, // 7
4211 -3, // 8
4212 -4, // 9 NearMaxPriority
4214 -5, // 10 MaxPriority
4216 -5 // 11 CriticalPriority
4217 };
4219 static int prio_init() {
4220 if (ThreadPriorityPolicy == 1) {
4221 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4222 // if effective uid is not root. Perhaps, a more elegant way of doing
4223 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4224 if (geteuid() != 0) {
4225 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4226 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4227 }
4228 ThreadPriorityPolicy = 0;
4229 }
4230 }
4231 if (UseCriticalJavaThreadPriority) {
4232 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4233 }
4234 return 0;
4235 }
4237 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4238 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4240 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4241 return (ret == 0) ? OS_OK : OS_ERR;
4242 }
4244 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4245 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4246 *priority_ptr = java_to_os_priority[NormPriority];
4247 return OS_OK;
4248 }
4250 errno = 0;
4251 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4252 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4253 }
4255 // Hint to the underlying OS that a task switch would not be good.
4256 // Void return because it's a hint and can fail.
4257 void os::hint_no_preempt() {}
4259 ////////////////////////////////////////////////////////////////////////////////
4260 // suspend/resume support
4262 // the low-level signal-based suspend/resume support is a remnant from the
4263 // old VM-suspension that used to be for java-suspension, safepoints etc,
4264 // within hotspot. Now there is a single use-case for this:
4265 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4266 // that runs in the watcher thread.
4267 // The remaining code is greatly simplified from the more general suspension
4268 // code that used to be used.
4269 //
4270 // The protocol is quite simple:
4271 // - suspend:
4272 // - sends a signal to the target thread
4273 // - polls the suspend state of the osthread using a yield loop
4274 // - target thread signal handler (SR_handler) sets suspend state
4275 // and blocks in sigsuspend until continued
4276 // - resume:
4277 // - sets target osthread state to continue
4278 // - sends signal to end the sigsuspend loop in the SR_handler
4279 //
4280 // Note that the SR_lock plays no role in this suspend/resume protocol.
4281 //
4283 static void resume_clear_context(OSThread *osthread) {
4284 osthread->set_ucontext(NULL);
4285 osthread->set_siginfo(NULL);
4286 }
4288 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4289 osthread->set_ucontext(context);
4290 osthread->set_siginfo(siginfo);
4291 }
4293 //
4294 // Handler function invoked when a thread's execution is suspended or
4295 // resumed. We have to be careful that only async-safe functions are
4296 // called here (Note: most pthread functions are not async safe and
4297 // should be avoided.)
4298 //
4299 // Note: sigwait() is a more natural fit than sigsuspend() from an
4300 // interface point of view, but sigwait() prevents the signal hander
4301 // from being run. libpthread would get very confused by not having
4302 // its signal handlers run and prevents sigwait()'s use with the
4303 // mutex granting granting signal.
4304 //
4305 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4306 //
4307 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4308 // Save and restore errno to avoid confusing native code with EINTR
4309 // after sigsuspend.
4310 int old_errno = errno;
4312 Thread* thread = Thread::current();
4313 OSThread* osthread = thread->osthread();
4314 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4316 os::SuspendResume::State current = osthread->sr.state();
4317 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4318 suspend_save_context(osthread, siginfo, context);
4320 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4321 os::SuspendResume::State state = osthread->sr.suspended();
4322 if (state == os::SuspendResume::SR_SUSPENDED) {
4323 sigset_t suspend_set; // signals for sigsuspend()
4325 // get current set of blocked signals and unblock resume signal
4326 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4327 sigdelset(&suspend_set, SR_signum);
4329 sr_semaphore.signal();
4330 // wait here until we are resumed
4331 while (1) {
4332 sigsuspend(&suspend_set);
4334 os::SuspendResume::State result = osthread->sr.running();
4335 if (result == os::SuspendResume::SR_RUNNING) {
4336 sr_semaphore.signal();
4337 break;
4338 }
4339 }
4341 } else if (state == os::SuspendResume::SR_RUNNING) {
4342 // request was cancelled, continue
4343 } else {
4344 ShouldNotReachHere();
4345 }
4347 resume_clear_context(osthread);
4348 } else if (current == os::SuspendResume::SR_RUNNING) {
4349 // request was cancelled, continue
4350 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4351 // ignore
4352 } else {
4353 // ignore
4354 }
4356 errno = old_errno;
4357 }
4360 static int SR_initialize() {
4361 struct sigaction act;
4362 char *s;
4363 /* Get signal number to use for suspend/resume */
4364 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4365 int sig = ::strtol(s, 0, 10);
4366 if (sig > 0 || sig < _NSIG) {
4367 SR_signum = sig;
4368 }
4369 }
4371 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4372 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4374 sigemptyset(&SR_sigset);
4375 sigaddset(&SR_sigset, SR_signum);
4377 /* Set up signal handler for suspend/resume */
4378 act.sa_flags = SA_RESTART|SA_SIGINFO;
4379 act.sa_handler = (void (*)(int)) SR_handler;
4381 // SR_signum is blocked by default.
4382 // 4528190 - We also need to block pthread restart signal (32 on all
4383 // supported Linux platforms). Note that LinuxThreads need to block
4384 // this signal for all threads to work properly. So we don't have
4385 // to use hard-coded signal number when setting up the mask.
4386 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4388 if (sigaction(SR_signum, &act, 0) == -1) {
4389 return -1;
4390 }
4392 // Save signal flag
4393 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4394 return 0;
4395 }
4397 static int sr_notify(OSThread* osthread) {
4398 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4399 assert_status(status == 0, status, "pthread_kill");
4400 return status;
4401 }
4403 // "Randomly" selected value for how long we want to spin
4404 // before bailing out on suspending a thread, also how often
4405 // we send a signal to a thread we want to resume
4406 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4407 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4409 // returns true on success and false on error - really an error is fatal
4410 // but this seems the normal response to library errors
4411 static bool do_suspend(OSThread* osthread) {
4412 assert(osthread->sr.is_running(), "thread should be running");
4413 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4415 // mark as suspended and send signal
4416 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4417 // failed to switch, state wasn't running?
4418 ShouldNotReachHere();
4419 return false;
4420 }
4422 if (sr_notify(osthread) != 0) {
4423 ShouldNotReachHere();
4424 }
4426 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4427 while (true) {
4428 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4429 break;
4430 } else {
4431 // timeout
4432 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4433 if (cancelled == os::SuspendResume::SR_RUNNING) {
4434 return false;
4435 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4436 // make sure that we consume the signal on the semaphore as well
4437 sr_semaphore.wait();
4438 break;
4439 } else {
4440 ShouldNotReachHere();
4441 return false;
4442 }
4443 }
4444 }
4446 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4447 return true;
4448 }
4450 static void do_resume(OSThread* osthread) {
4451 assert(osthread->sr.is_suspended(), "thread should be suspended");
4452 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4454 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4455 // failed to switch to WAKEUP_REQUEST
4456 ShouldNotReachHere();
4457 return;
4458 }
4460 while (true) {
4461 if (sr_notify(osthread) == 0) {
4462 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4463 if (osthread->sr.is_running()) {
4464 return;
4465 }
4466 }
4467 } else {
4468 ShouldNotReachHere();
4469 }
4470 }
4472 guarantee(osthread->sr.is_running(), "Must be running!");
4473 }
4475 ////////////////////////////////////////////////////////////////////////////////
4476 // interrupt support
4478 void os::interrupt(Thread* thread) {
4479 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4480 "possibility of dangling Thread pointer");
4482 OSThread* osthread = thread->osthread();
4484 if (!osthread->interrupted()) {
4485 osthread->set_interrupted(true);
4486 // More than one thread can get here with the same value of osthread,
4487 // resulting in multiple notifications. We do, however, want the store
4488 // to interrupted() to be visible to other threads before we execute unpark().
4489 OrderAccess::fence();
4490 ParkEvent * const slp = thread->_SleepEvent ;
4491 if (slp != NULL) slp->unpark() ;
4492 }
4494 // For JSR166. Unpark even if interrupt status already was set
4495 if (thread->is_Java_thread())
4496 ((JavaThread*)thread)->parker()->unpark();
4498 ParkEvent * ev = thread->_ParkEvent ;
4499 if (ev != NULL) ev->unpark() ;
4501 }
4503 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4504 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4505 "possibility of dangling Thread pointer");
4507 OSThread* osthread = thread->osthread();
4509 bool interrupted = osthread->interrupted();
4511 if (interrupted && clear_interrupted) {
4512 osthread->set_interrupted(false);
4513 // consider thread->_SleepEvent->reset() ... optional optimization
4514 }
4516 return interrupted;
4517 }
4519 ///////////////////////////////////////////////////////////////////////////////////
4520 // signal handling (except suspend/resume)
4522 // This routine may be used by user applications as a "hook" to catch signals.
4523 // The user-defined signal handler must pass unrecognized signals to this
4524 // routine, and if it returns true (non-zero), then the signal handler must
4525 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4526 // routine will never retun false (zero), but instead will execute a VM panic
4527 // routine kill the process.
4528 //
4529 // If this routine returns false, it is OK to call it again. This allows
4530 // the user-defined signal handler to perform checks either before or after
4531 // the VM performs its own checks. Naturally, the user code would be making
4532 // a serious error if it tried to handle an exception (such as a null check
4533 // or breakpoint) that the VM was generating for its own correct operation.
4534 //
4535 // This routine may recognize any of the following kinds of signals:
4536 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4537 // It should be consulted by handlers for any of those signals.
4538 //
4539 // The caller of this routine must pass in the three arguments supplied
4540 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4541 // field of the structure passed to sigaction(). This routine assumes that
4542 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4543 //
4544 // Note that the VM will print warnings if it detects conflicting signal
4545 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4546 //
4547 extern "C" JNIEXPORT int
4548 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4549 void* ucontext, int abort_if_unrecognized);
4551 void signalHandler(int sig, siginfo_t* info, void* uc) {
4552 assert(info != NULL && uc != NULL, "it must be old kernel");
4553 int orig_errno = errno; // Preserve errno value over signal handler.
4554 JVM_handle_linux_signal(sig, info, uc, true);
4555 errno = orig_errno;
4556 }
4559 // This boolean allows users to forward their own non-matching signals
4560 // to JVM_handle_linux_signal, harmlessly.
4561 bool os::Linux::signal_handlers_are_installed = false;
4563 // For signal-chaining
4564 struct sigaction os::Linux::sigact[MAXSIGNUM];
4565 unsigned int os::Linux::sigs = 0;
4566 bool os::Linux::libjsig_is_loaded = false;
4567 typedef struct sigaction *(*get_signal_t)(int);
4568 get_signal_t os::Linux::get_signal_action = NULL;
4570 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4571 struct sigaction *actp = NULL;
4573 if (libjsig_is_loaded) {
4574 // Retrieve the old signal handler from libjsig
4575 actp = (*get_signal_action)(sig);
4576 }
4577 if (actp == NULL) {
4578 // Retrieve the preinstalled signal handler from jvm
4579 actp = get_preinstalled_handler(sig);
4580 }
4582 return actp;
4583 }
4585 static bool call_chained_handler(struct sigaction *actp, int sig,
4586 siginfo_t *siginfo, void *context) {
4587 // Call the old signal handler
4588 if (actp->sa_handler == SIG_DFL) {
4589 // It's more reasonable to let jvm treat it as an unexpected exception
4590 // instead of taking the default action.
4591 return false;
4592 } else if (actp->sa_handler != SIG_IGN) {
4593 if ((actp->sa_flags & SA_NODEFER) == 0) {
4594 // automaticlly block the signal
4595 sigaddset(&(actp->sa_mask), sig);
4596 }
4598 sa_handler_t hand = NULL;
4599 sa_sigaction_t sa = NULL;
4600 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4601 // retrieve the chained handler
4602 if (siginfo_flag_set) {
4603 sa = actp->sa_sigaction;
4604 } else {
4605 hand = actp->sa_handler;
4606 }
4608 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4609 actp->sa_handler = SIG_DFL;
4610 }
4612 // try to honor the signal mask
4613 sigset_t oset;
4614 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4616 // call into the chained handler
4617 if (siginfo_flag_set) {
4618 (*sa)(sig, siginfo, context);
4619 } else {
4620 (*hand)(sig);
4621 }
4623 // restore the signal mask
4624 pthread_sigmask(SIG_SETMASK, &oset, 0);
4625 }
4626 // Tell jvm's signal handler the signal is taken care of.
4627 return true;
4628 }
4630 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4631 bool chained = false;
4632 // signal-chaining
4633 if (UseSignalChaining) {
4634 struct sigaction *actp = get_chained_signal_action(sig);
4635 if (actp != NULL) {
4636 chained = call_chained_handler(actp, sig, siginfo, context);
4637 }
4638 }
4639 return chained;
4640 }
4642 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4643 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4644 return &sigact[sig];
4645 }
4646 return NULL;
4647 }
4649 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4650 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4651 sigact[sig] = oldAct;
4652 sigs |= (unsigned int)1 << sig;
4653 }
4655 // for diagnostic
4656 int os::Linux::sigflags[MAXSIGNUM];
4658 int os::Linux::get_our_sigflags(int sig) {
4659 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4660 return sigflags[sig];
4661 }
4663 void os::Linux::set_our_sigflags(int sig, int flags) {
4664 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4665 sigflags[sig] = flags;
4666 }
4668 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4669 // Check for overwrite.
4670 struct sigaction oldAct;
4671 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4673 void* oldhand = oldAct.sa_sigaction
4674 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4675 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4676 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4677 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4678 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4679 if (AllowUserSignalHandlers || !set_installed) {
4680 // Do not overwrite; user takes responsibility to forward to us.
4681 return;
4682 } else if (UseSignalChaining) {
4683 // save the old handler in jvm
4684 save_preinstalled_handler(sig, oldAct);
4685 // libjsig also interposes the sigaction() call below and saves the
4686 // old sigaction on it own.
4687 } else {
4688 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4689 "%#lx for signal %d.", (long)oldhand, sig));
4690 }
4691 }
4693 struct sigaction sigAct;
4694 sigfillset(&(sigAct.sa_mask));
4695 sigAct.sa_handler = SIG_DFL;
4696 if (!set_installed) {
4697 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4698 } else {
4699 sigAct.sa_sigaction = signalHandler;
4700 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4701 }
4702 // Save flags, which are set by ours
4703 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4704 sigflags[sig] = sigAct.sa_flags;
4706 int ret = sigaction(sig, &sigAct, &oldAct);
4707 assert(ret == 0, "check");
4709 void* oldhand2 = oldAct.sa_sigaction
4710 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4711 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4712 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4713 }
4715 // install signal handlers for signals that HotSpot needs to
4716 // handle in order to support Java-level exception handling.
4718 void os::Linux::install_signal_handlers() {
4719 if (!signal_handlers_are_installed) {
4720 signal_handlers_are_installed = true;
4722 // signal-chaining
4723 typedef void (*signal_setting_t)();
4724 signal_setting_t begin_signal_setting = NULL;
4725 signal_setting_t end_signal_setting = NULL;
4726 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4727 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4728 if (begin_signal_setting != NULL) {
4729 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4730 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4731 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4732 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4733 libjsig_is_loaded = true;
4734 assert(UseSignalChaining, "should enable signal-chaining");
4735 }
4736 if (libjsig_is_loaded) {
4737 // Tell libjsig jvm is setting signal handlers
4738 (*begin_signal_setting)();
4739 }
4741 set_signal_handler(SIGSEGV, true);
4742 set_signal_handler(SIGPIPE, true);
4743 set_signal_handler(SIGBUS, true);
4744 set_signal_handler(SIGILL, true);
4745 set_signal_handler(SIGFPE, true);
4746 #if defined(PPC64)
4747 set_signal_handler(SIGTRAP, true);
4748 #endif
4749 set_signal_handler(SIGXFSZ, true);
4751 if (libjsig_is_loaded) {
4752 // Tell libjsig jvm finishes setting signal handlers
4753 (*end_signal_setting)();
4754 }
4756 // We don't activate signal checker if libjsig is in place, we trust ourselves
4757 // and if UserSignalHandler is installed all bets are off.
4758 // Log that signal checking is off only if -verbose:jni is specified.
4759 if (CheckJNICalls) {
4760 if (libjsig_is_loaded) {
4761 if (PrintJNIResolving) {
4762 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4763 }
4764 check_signals = false;
4765 }
4766 if (AllowUserSignalHandlers) {
4767 if (PrintJNIResolving) {
4768 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4769 }
4770 check_signals = false;
4771 }
4772 }
4773 }
4774 }
4776 // This is the fastest way to get thread cpu time on Linux.
4777 // Returns cpu time (user+sys) for any thread, not only for current.
4778 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4779 // It might work on 2.6.10+ with a special kernel/glibc patch.
4780 // For reference, please, see IEEE Std 1003.1-2004:
4781 // http://www.unix.org/single_unix_specification
4783 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4784 struct timespec tp;
4785 int rc = os::Linux::clock_gettime(clockid, &tp);
4786 assert(rc == 0, "clock_gettime is expected to return 0 code");
4788 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4789 }
4791 /////
4792 // glibc on Linux platform uses non-documented flag
4793 // to indicate, that some special sort of signal
4794 // trampoline is used.
4795 // We will never set this flag, and we should
4796 // ignore this flag in our diagnostic
4797 #ifdef SIGNIFICANT_SIGNAL_MASK
4798 #undef SIGNIFICANT_SIGNAL_MASK
4799 #endif
4800 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4802 static const char* get_signal_handler_name(address handler,
4803 char* buf, int buflen) {
4804 int offset = 0;
4805 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4806 if (found) {
4807 // skip directory names
4808 const char *p1, *p2;
4809 p1 = buf;
4810 size_t len = strlen(os::file_separator());
4811 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4812 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4813 } else {
4814 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4815 }
4816 return buf;
4817 }
4819 static void print_signal_handler(outputStream* st, int sig,
4820 char* buf, size_t buflen) {
4821 struct sigaction sa;
4823 sigaction(sig, NULL, &sa);
4825 // See comment for SIGNIFICANT_SIGNAL_MASK define
4826 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4828 st->print("%s: ", os::exception_name(sig, buf, buflen));
4830 address handler = (sa.sa_flags & SA_SIGINFO)
4831 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4832 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4834 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4835 st->print("SIG_DFL");
4836 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4837 st->print("SIG_IGN");
4838 } else {
4839 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4840 }
4842 st->print(", sa_mask[0]=");
4843 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4845 address rh = VMError::get_resetted_sighandler(sig);
4846 // May be, handler was resetted by VMError?
4847 if(rh != NULL) {
4848 handler = rh;
4849 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4850 }
4852 st->print(", sa_flags=");
4853 os::Posix::print_sa_flags(st, sa.sa_flags);
4855 // Check: is it our handler?
4856 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4857 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4858 // It is our signal handler
4859 // check for flags, reset system-used one!
4860 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4861 st->print(
4862 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4863 os::Linux::get_our_sigflags(sig));
4864 }
4865 }
4866 st->cr();
4867 }
4870 #define DO_SIGNAL_CHECK(sig) \
4871 if (!sigismember(&check_signal_done, sig)) \
4872 os::Linux::check_signal_handler(sig)
4874 // This method is a periodic task to check for misbehaving JNI applications
4875 // under CheckJNI, we can add any periodic checks here
4877 void os::run_periodic_checks() {
4879 if (check_signals == false) return;
4881 // SEGV and BUS if overridden could potentially prevent
4882 // generation of hs*.log in the event of a crash, debugging
4883 // such a case can be very challenging, so we absolutely
4884 // check the following for a good measure:
4885 DO_SIGNAL_CHECK(SIGSEGV);
4886 DO_SIGNAL_CHECK(SIGILL);
4887 DO_SIGNAL_CHECK(SIGFPE);
4888 DO_SIGNAL_CHECK(SIGBUS);
4889 DO_SIGNAL_CHECK(SIGPIPE);
4890 DO_SIGNAL_CHECK(SIGXFSZ);
4891 #if defined(PPC64)
4892 DO_SIGNAL_CHECK(SIGTRAP);
4893 #endif
4895 // ReduceSignalUsage allows the user to override these handlers
4896 // see comments at the very top and jvm_solaris.h
4897 if (!ReduceSignalUsage) {
4898 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4899 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4900 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4901 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4902 }
4904 DO_SIGNAL_CHECK(SR_signum);
4905 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4906 }
4908 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4910 static os_sigaction_t os_sigaction = NULL;
4912 void os::Linux::check_signal_handler(int sig) {
4913 char buf[O_BUFLEN];
4914 address jvmHandler = NULL;
4917 struct sigaction act;
4918 if (os_sigaction == NULL) {
4919 // only trust the default sigaction, in case it has been interposed
4920 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4921 if (os_sigaction == NULL) return;
4922 }
4924 os_sigaction(sig, (struct sigaction*)NULL, &act);
4927 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4929 address thisHandler = (act.sa_flags & SA_SIGINFO)
4930 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4931 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4934 switch(sig) {
4935 case SIGSEGV:
4936 case SIGBUS:
4937 case SIGFPE:
4938 case SIGPIPE:
4939 case SIGILL:
4940 case SIGXFSZ:
4941 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4942 break;
4944 case SHUTDOWN1_SIGNAL:
4945 case SHUTDOWN2_SIGNAL:
4946 case SHUTDOWN3_SIGNAL:
4947 case BREAK_SIGNAL:
4948 jvmHandler = (address)user_handler();
4949 break;
4951 case INTERRUPT_SIGNAL:
4952 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4953 break;
4955 default:
4956 if (sig == SR_signum) {
4957 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4958 } else {
4959 return;
4960 }
4961 break;
4962 }
4964 if (thisHandler != jvmHandler) {
4965 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4966 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4967 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4968 // No need to check this sig any longer
4969 sigaddset(&check_signal_done, sig);
4970 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4971 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4972 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4973 exception_name(sig, buf, O_BUFLEN));
4974 }
4975 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4976 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4977 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4978 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4979 // No need to check this sig any longer
4980 sigaddset(&check_signal_done, sig);
4981 }
4983 // Dump all the signal
4984 if (sigismember(&check_signal_done, sig)) {
4985 print_signal_handlers(tty, buf, O_BUFLEN);
4986 }
4987 }
4989 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4991 extern bool signal_name(int signo, char* buf, size_t len);
4993 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4994 if (0 < exception_code && exception_code <= SIGRTMAX) {
4995 // signal
4996 if (!signal_name(exception_code, buf, size)) {
4997 jio_snprintf(buf, size, "SIG%d", exception_code);
4998 }
4999 return buf;
5000 } else {
5001 return NULL;
5002 }
5003 }
5005 // this is called _before_ most of the global arguments have been parsed
5006 void os::init(void) {
5007 char dummy; /* used to get a guess on initial stack address */
5009 // With LinuxThreads the JavaMain thread pid (primordial thread)
5010 // is different than the pid of the java launcher thread.
5011 // So, on Linux, the launcher thread pid is passed to the VM
5012 // via the sun.java.launcher.pid property.
5013 // Use this property instead of getpid() if it was correctly passed.
5014 // See bug 6351349.
5015 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
5017 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
5019 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5021 init_random(1234567);
5023 ThreadCritical::initialize();
5025 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5026 if (Linux::page_size() == -1) {
5027 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
5028 strerror(errno)));
5029 }
5030 init_page_sizes((size_t) Linux::page_size());
5032 Linux::initialize_system_info();
5034 // _main_thread points to the thread that created/loaded the JVM.
5035 Linux::_main_thread = pthread_self();
5037 Linux::clock_init();
5038 initial_time_count = javaTimeNanos();
5040 // pthread_condattr initialization for monotonic clock
5041 int status;
5042 pthread_condattr_t* _condattr = os::Linux::condAttr();
5043 if ((status = pthread_condattr_init(_condattr)) != 0) {
5044 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
5045 }
5046 // Only set the clock if CLOCK_MONOTONIC is available
5047 if (Linux::supports_monotonic_clock()) {
5048 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
5049 if (status == EINVAL) {
5050 warning("Unable to use monotonic clock with relative timed-waits" \
5051 " - changes to the time-of-day clock may have adverse affects");
5052 } else {
5053 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
5054 }
5055 }
5056 }
5057 // else it defaults to CLOCK_REALTIME
5059 pthread_mutex_init(&dl_mutex, NULL);
5061 // If the pagesize of the VM is greater than 8K determine the appropriate
5062 // number of initial guard pages. The user can change this with the
5063 // command line arguments, if needed.
5064 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
5065 StackYellowPages = 1;
5066 StackRedPages = 1;
5067 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
5068 }
5070 // retrieve entry point for pthread_setname_np
5071 Linux::_pthread_setname_np =
5072 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5074 }
5076 // To install functions for atexit system call
5077 extern "C" {
5078 static void perfMemory_exit_helper() {
5079 perfMemory_exit();
5080 }
5081 }
5083 void os::pd_init_container_support() {
5084 OSContainer::init();
5085 }
5087 // this is called _after_ the global arguments have been parsed
5088 jint os::init_2(void)
5089 {
5090 Linux::fast_thread_clock_init();
5092 // Allocate a single page and mark it as readable for safepoint polling
5093 #ifdef OPT_SAFEPOINT
5094 void * p = (void *)(0x10000);
5095 address polling_page = (address) ::mmap(p, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5096 #else
5097 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5098 #endif
5099 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
5101 os::set_polling_page( polling_page );
5103 #ifndef PRODUCT
5104 if(Verbose && PrintMiscellaneous)
5105 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5106 #endif
5108 if (!UseMembar) {
5109 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5110 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
5111 os::set_memory_serialize_page( mem_serialize_page );
5113 #ifndef PRODUCT
5114 if(Verbose && PrintMiscellaneous)
5115 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5116 #endif
5117 }
5119 // initialize suspend/resume support - must do this before signal_sets_init()
5120 if (SR_initialize() != 0) {
5121 perror("SR_initialize failed");
5122 return JNI_ERR;
5123 }
5125 Linux::signal_sets_init();
5126 Linux::install_signal_handlers();
5128 // Check minimum allowable stack size for thread creation and to initialize
5129 // the java system classes, including StackOverflowError - depends on page
5130 // size. Add a page for compiler2 recursion in main thread.
5131 // Add in 2*BytesPerWord times page size to account for VM stack during
5132 // class initialization depending on 32 or 64 bit VM.
5134 /* 2014/1/2 Liao: JDK8 requires larger -Xss option.
5135 * TongWeb cannot run with -Xss192K.
5136 * We are not sure whether this causes errors, so simply print a warning. */
5137 size_t min_stack_allowed_jdk6 = os::Linux::min_stack_allowed;
5138 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
5139 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
5140 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
5142 size_t threadStackSizeInBytes = ThreadStackSize * K;
5143 if (threadStackSizeInBytes != 0 &&
5144 threadStackSizeInBytes < min_stack_allowed_jdk6) {
5145 tty->print_cr("\nThe stack size specified is too small, "
5146 "Specify at least %dk",
5147 os::Linux::min_stack_allowed/ K);
5148 return JNI_ERR;
5149 }
5151 // Make the stack size a multiple of the page size so that
5152 // the yellow/red zones can be guarded.
5153 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5154 vm_page_size()));
5156 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5158 #if defined(IA32)
5159 workaround_expand_exec_shield_cs_limit();
5160 #endif
5162 Linux::libpthread_init();
5163 if (PrintMiscellaneous && (Verbose || WizardMode)) {
5164 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
5165 Linux::glibc_version(), Linux::libpthread_version(),
5166 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
5167 }
5169 if (UseNUMA) {
5170 if (!Linux::libnuma_init()) {
5171 UseNUMA = false;
5172 } else {
5173 if ((Linux::numa_max_node() < 1)) {
5174 // There's only one node(they start from 0), disable NUMA.
5175 UseNUMA = false;
5176 }
5177 }
5178 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5179 // we can make the adaptive lgrp chunk resizing work. If the user specified
5180 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
5181 // disable adaptive resizing.
5182 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5183 if (FLAG_IS_DEFAULT(UseNUMA)) {
5184 UseNUMA = false;
5185 } else {
5186 if (FLAG_IS_DEFAULT(UseLargePages) &&
5187 FLAG_IS_DEFAULT(UseSHM) &&
5188 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
5189 UseLargePages = false;
5190 } else {
5191 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
5192 UseAdaptiveSizePolicy = false;
5193 UseAdaptiveNUMAChunkSizing = false;
5194 }
5195 }
5196 }
5197 if (!UseNUMA && ForceNUMA) {
5198 UseNUMA = true;
5199 }
5200 }
5202 if (MaxFDLimit) {
5203 // set the number of file descriptors to max. print out error
5204 // if getrlimit/setrlimit fails but continue regardless.
5205 struct rlimit nbr_files;
5206 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5207 if (status != 0) {
5208 if (PrintMiscellaneous && (Verbose || WizardMode))
5209 perror("os::init_2 getrlimit failed");
5210 } else {
5211 nbr_files.rlim_cur = nbr_files.rlim_max;
5212 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5213 if (status != 0) {
5214 if (PrintMiscellaneous && (Verbose || WizardMode))
5215 perror("os::init_2 setrlimit failed");
5216 }
5217 }
5218 }
5220 // Initialize lock used to serialize thread creation (see os::create_thread)
5221 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5223 // at-exit methods are called in the reverse order of their registration.
5224 // atexit functions are called on return from main or as a result of a
5225 // call to exit(3C). There can be only 32 of these functions registered
5226 // and atexit() does not set errno.
5228 if (PerfAllowAtExitRegistration) {
5229 // only register atexit functions if PerfAllowAtExitRegistration is set.
5230 // atexit functions can be delayed until process exit time, which
5231 // can be problematic for embedded VM situations. Embedded VMs should
5232 // call DestroyJavaVM() to assure that VM resources are released.
5234 // note: perfMemory_exit_helper atexit function may be removed in
5235 // the future if the appropriate cleanup code can be added to the
5236 // VM_Exit VMOperation's doit method.
5237 if (atexit(perfMemory_exit_helper) != 0) {
5238 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5239 }
5240 }
5242 // initialize thread priority policy
5243 prio_init();
5245 return JNI_OK;
5246 }
5248 // Mark the polling page as unreadable
5249 void os::make_polling_page_unreadable(void) {
5250 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5251 fatal("Could not disable polling page");
5252 };
5254 // Mark the polling page as readable
5255 void os::make_polling_page_readable(void) {
5256 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5257 fatal("Could not enable polling page");
5258 }
5259 };
5261 static int os_cpu_count(const cpu_set_t* cpus) {
5262 int count = 0;
5263 // only look up to the number of configured processors
5264 for (int i = 0; i < os::processor_count(); i++) {
5265 if (CPU_ISSET(i, cpus)) {
5266 count++;
5267 }
5268 }
5269 return count;
5270 }
5272 // Get the current number of available processors for this process.
5273 // This value can change at any time during a process's lifetime.
5274 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5275 // If anything goes wrong we fallback to returning the number of online
5276 // processors - which can be greater than the number available to the process.
5277 int os::Linux::active_processor_count() {
5278 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5279 int cpus_size = sizeof(cpu_set_t);
5280 int cpu_count = 0;
5282 // pid 0 means the current thread - which we have to assume represents the process
5283 if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
5284 cpu_count = os_cpu_count(&cpus);
5285 if (PrintActiveCpus) {
5286 tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5287 }
5288 }
5289 else {
5290 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5291 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5292 "which may exceed available processors", strerror(errno), cpu_count);
5293 }
5295 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5296 return cpu_count;
5297 }
5299 // Determine the active processor count from one of
5300 // three different sources:
5301 //
5302 // 1. User option -XX:ActiveProcessorCount
5303 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5304 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5305 //
5306 // Option 1, if specified, will always override.
5307 // If the cgroup subsystem is active and configured, we
5308 // will return the min of the cgroup and option 2 results.
5309 // This is required since tools, such as numactl, that
5310 // alter cpu affinity do not update cgroup subsystem
5311 // cpuset configuration files.
5312 int os::active_processor_count() {
5313 // User has overridden the number of active processors
5314 if (ActiveProcessorCount > 0) {
5315 if (PrintActiveCpus) {
5316 tty->print_cr("active_processor_count: "
5317 "active processor count set by user : %d",
5318 ActiveProcessorCount);
5319 }
5320 return ActiveProcessorCount;
5321 }
5323 int active_cpus;
5324 if (OSContainer::is_containerized()) {
5325 active_cpus = OSContainer::active_processor_count();
5326 if (PrintActiveCpus) {
5327 tty->print_cr("active_processor_count: determined by OSContainer: %d",
5328 active_cpus);
5329 }
5330 } else {
5331 active_cpus = os::Linux::active_processor_count();
5332 }
5334 return active_cpus;
5335 }
5337 void os::set_native_thread_name(const char *name) {
5338 if (Linux::_pthread_setname_np) {
5339 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5340 snprintf(buf, sizeof(buf), "%s", name);
5341 buf[sizeof(buf) - 1] = '\0';
5342 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5343 // ERANGE should not happen; all other errors should just be ignored.
5344 assert(rc != ERANGE, "pthread_setname_np failed");
5345 }
5346 }
5348 bool os::distribute_processes(uint length, uint* distribution) {
5349 // Not yet implemented.
5350 return false;
5351 }
5353 bool os::bind_to_processor(uint processor_id) {
5354 // Not yet implemented.
5355 return false;
5356 }
5358 ///
5360 void os::SuspendedThreadTask::internal_do_task() {
5361 if (do_suspend(_thread->osthread())) {
5362 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5363 do_task(context);
5364 do_resume(_thread->osthread());
5365 }
5366 }
5368 class PcFetcher : public os::SuspendedThreadTask {
5369 public:
5370 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5371 ExtendedPC result();
5372 protected:
5373 void do_task(const os::SuspendedThreadTaskContext& context);
5374 private:
5375 ExtendedPC _epc;
5376 };
5378 ExtendedPC PcFetcher::result() {
5379 guarantee(is_done(), "task is not done yet.");
5380 return _epc;
5381 }
5383 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5384 Thread* thread = context.thread();
5385 OSThread* osthread = thread->osthread();
5386 if (osthread->ucontext() != NULL) {
5387 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5388 } else {
5389 // NULL context is unexpected, double-check this is the VMThread
5390 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5391 }
5392 }
5394 // Suspends the target using the signal mechanism and then grabs the PC before
5395 // resuming the target. Used by the flat-profiler only
5396 ExtendedPC os::get_thread_pc(Thread* thread) {
5397 // Make sure that it is called by the watcher for the VMThread
5398 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5399 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5401 PcFetcher fetcher(thread);
5402 fetcher.run();
5403 return fetcher.result();
5404 }
5406 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5407 {
5408 if (is_NPTL()) {
5409 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5410 } else {
5411 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5412 // word back to default 64bit precision if condvar is signaled. Java
5413 // wants 53bit precision. Save and restore current value.
5414 int fpu = get_fpu_control_word();
5415 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5416 set_fpu_control_word(fpu);
5417 return status;
5418 }
5419 }
5421 ////////////////////////////////////////////////////////////////////////////////
5422 // debug support
5424 bool os::find(address addr, outputStream* st) {
5425 Dl_info dlinfo;
5426 memset(&dlinfo, 0, sizeof(dlinfo));
5427 if (dladdr(addr, &dlinfo) != 0) {
5428 st->print(PTR_FORMAT ": ", addr);
5429 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5430 st->print("%s+%#x", dlinfo.dli_sname,
5431 addr - (intptr_t)dlinfo.dli_saddr);
5432 } else if (dlinfo.dli_fbase != NULL) {
5433 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5434 } else {
5435 st->print("<absolute address>");
5436 }
5437 if (dlinfo.dli_fname != NULL) {
5438 st->print(" in %s", dlinfo.dli_fname);
5439 }
5440 if (dlinfo.dli_fbase != NULL) {
5441 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5442 }
5443 st->cr();
5445 if (Verbose) {
5446 // decode some bytes around the PC
5447 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5448 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5449 address lowest = (address) dlinfo.dli_sname;
5450 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5451 if (begin < lowest) begin = lowest;
5452 Dl_info dlinfo2;
5453 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5454 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5455 end = (address) dlinfo2.dli_saddr;
5456 Disassembler::decode(begin, end, st);
5457 }
5458 return true;
5459 }
5460 return false;
5461 }
5463 ////////////////////////////////////////////////////////////////////////////////
5464 // misc
5466 // This does not do anything on Linux. This is basically a hook for being
5467 // able to use structured exception handling (thread-local exception filters)
5468 // on, e.g., Win32.
5469 void
5470 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5471 JavaCallArguments* args, Thread* thread) {
5472 f(value, method, args, thread);
5473 }
5475 void os::print_statistics() {
5476 }
5478 int os::message_box(const char* title, const char* message) {
5479 int i;
5480 fdStream err(defaultStream::error_fd());
5481 for (i = 0; i < 78; i++) err.print_raw("=");
5482 err.cr();
5483 err.print_raw_cr(title);
5484 for (i = 0; i < 78; i++) err.print_raw("-");
5485 err.cr();
5486 err.print_raw_cr(message);
5487 for (i = 0; i < 78; i++) err.print_raw("=");
5488 err.cr();
5490 char buf[16];
5491 // Prevent process from exiting upon "read error" without consuming all CPU
5492 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5494 return buf[0] == 'y' || buf[0] == 'Y';
5495 }
5497 int os::stat(const char *path, struct stat *sbuf) {
5498 char pathbuf[MAX_PATH];
5499 if (strlen(path) > MAX_PATH - 1) {
5500 errno = ENAMETOOLONG;
5501 return -1;
5502 }
5503 os::native_path(strcpy(pathbuf, path));
5504 return ::stat(pathbuf, sbuf);
5505 }
5507 bool os::check_heap(bool force) {
5508 return true;
5509 }
5511 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5512 return ::vsnprintf(buf, count, format, args);
5513 }
5515 // Is a (classpath) directory empty?
5516 bool os::dir_is_empty(const char* path) {
5517 DIR *dir = NULL;
5518 struct dirent *ptr;
5520 dir = opendir(path);
5521 if (dir == NULL) return true;
5523 /* Scan the directory */
5524 bool result = true;
5525 while (result && (ptr = readdir(dir)) != NULL) {
5526 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5527 result = false;
5528 }
5529 }
5530 closedir(dir);
5531 return result;
5532 }
5534 // This code originates from JDK's sysOpen and open64_w
5535 // from src/solaris/hpi/src/system_md.c
5537 #ifndef O_DELETE
5538 #define O_DELETE 0x10000
5539 #endif
5541 // Open a file. Unlink the file immediately after open returns
5542 // if the specified oflag has the O_DELETE flag set.
5543 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5545 int os::open(const char *path, int oflag, int mode) {
5547 if (strlen(path) > MAX_PATH - 1) {
5548 errno = ENAMETOOLONG;
5549 return -1;
5550 }
5551 int fd;
5552 int o_delete = (oflag & O_DELETE);
5553 oflag = oflag & ~O_DELETE;
5555 fd = ::open64(path, oflag, mode);
5556 if (fd == -1) return -1;
5558 //If the open succeeded, the file might still be a directory
5559 {
5560 struct stat64 buf64;
5561 int ret = ::fstat64(fd, &buf64);
5562 int st_mode = buf64.st_mode;
5564 if (ret != -1) {
5565 if ((st_mode & S_IFMT) == S_IFDIR) {
5566 errno = EISDIR;
5567 ::close(fd);
5568 return -1;
5569 }
5570 } else {
5571 ::close(fd);
5572 return -1;
5573 }
5574 }
5576 /*
5577 * All file descriptors that are opened in the JVM and not
5578 * specifically destined for a subprocess should have the
5579 * close-on-exec flag set. If we don't set it, then careless 3rd
5580 * party native code might fork and exec without closing all
5581 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5582 * UNIXProcess.c), and this in turn might:
5583 *
5584 * - cause end-of-file to fail to be detected on some file
5585 * descriptors, resulting in mysterious hangs, or
5586 *
5587 * - might cause an fopen in the subprocess to fail on a system
5588 * suffering from bug 1085341.
5589 *
5590 * (Yes, the default setting of the close-on-exec flag is a Unix
5591 * design flaw)
5592 *
5593 * See:
5594 * 1085341: 32-bit stdio routines should support file descriptors >255
5595 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5596 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5597 */
5598 #ifdef FD_CLOEXEC
5599 {
5600 int flags = ::fcntl(fd, F_GETFD);
5601 if (flags != -1)
5602 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5603 }
5604 #endif
5606 if (o_delete != 0) {
5607 ::unlink(path);
5608 }
5609 return fd;
5610 }
5613 // create binary file, rewriting existing file if required
5614 int os::create_binary_file(const char* path, bool rewrite_existing) {
5615 int oflags = O_WRONLY | O_CREAT;
5616 if (!rewrite_existing) {
5617 oflags |= O_EXCL;
5618 }
5619 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5620 }
5622 // return current position of file pointer
5623 jlong os::current_file_offset(int fd) {
5624 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5625 }
5627 // move file pointer to the specified offset
5628 jlong os::seek_to_file_offset(int fd, jlong offset) {
5629 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5630 }
5632 // This code originates from JDK's sysAvailable
5633 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5635 int os::available(int fd, jlong *bytes) {
5636 jlong cur, end;
5637 int mode;
5638 struct stat64 buf64;
5640 if (::fstat64(fd, &buf64) >= 0) {
5641 mode = buf64.st_mode;
5642 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5643 /*
5644 * XXX: is the following call interruptible? If so, this might
5645 * need to go through the INTERRUPT_IO() wrapper as for other
5646 * blocking, interruptible calls in this file.
5647 */
5648 int n;
5649 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5650 *bytes = n;
5651 return 1;
5652 }
5653 }
5654 }
5655 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5656 return 0;
5657 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5658 return 0;
5659 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5660 return 0;
5661 }
5662 *bytes = end - cur;
5663 return 1;
5664 }
5666 int os::socket_available(int fd, jint *pbytes) {
5667 // Linux doc says EINTR not returned, unlike Solaris
5668 int ret = ::ioctl(fd, FIONREAD, pbytes);
5670 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5671 // is expected to return 0 on failure and 1 on success to the jdk.
5672 return (ret < 0) ? 0 : 1;
5673 }
5675 // Map a block of memory.
5676 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5677 char *addr, size_t bytes, bool read_only,
5678 bool allow_exec) {
5679 int prot;
5680 int flags = MAP_PRIVATE;
5682 if (read_only) {
5683 prot = PROT_READ;
5684 } else {
5685 prot = PROT_READ | PROT_WRITE;
5686 }
5688 if (allow_exec) {
5689 prot |= PROT_EXEC;
5690 }
5692 if (addr != NULL) {
5693 flags |= MAP_FIXED;
5694 }
5696 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5697 fd, file_offset);
5698 if (mapped_address == MAP_FAILED) {
5699 return NULL;
5700 }
5701 return mapped_address;
5702 }
5705 // Remap a block of memory.
5706 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5707 char *addr, size_t bytes, bool read_only,
5708 bool allow_exec) {
5709 // same as map_memory() on this OS
5710 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5711 allow_exec);
5712 }
5715 // Unmap a block of memory.
5716 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5717 return munmap(addr, bytes) == 0;
5718 }
5720 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5722 static clockid_t thread_cpu_clockid(Thread* thread) {
5723 pthread_t tid = thread->osthread()->pthread_id();
5724 clockid_t clockid;
5726 // Get thread clockid
5727 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5728 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5729 return clockid;
5730 }
5732 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5733 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5734 // of a thread.
5735 //
5736 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5737 // the fast estimate available on the platform.
5739 jlong os::current_thread_cpu_time() {
5740 if (os::Linux::supports_fast_thread_cpu_time()) {
5741 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5742 } else {
5743 // return user + sys since the cost is the same
5744 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5745 }
5746 }
5748 jlong os::thread_cpu_time(Thread* thread) {
5749 // consistent with what current_thread_cpu_time() returns
5750 if (os::Linux::supports_fast_thread_cpu_time()) {
5751 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5752 } else {
5753 return slow_thread_cpu_time(thread, true /* user + sys */);
5754 }
5755 }
5757 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5758 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5759 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5760 } else {
5761 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5762 }
5763 }
5765 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5766 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5767 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5768 } else {
5769 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5770 }
5771 }
5773 //
5774 // -1 on error.
5775 //
5777 PRAGMA_DIAG_PUSH
5778 PRAGMA_FORMAT_NONLITERAL_IGNORED
5779 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5780 static bool proc_task_unchecked = true;
5781 static const char *proc_stat_path = "/proc/%d/stat";
5782 pid_t tid = thread->osthread()->thread_id();
5783 char *s;
5784 char stat[2048];
5785 int statlen;
5786 char proc_name[64];
5787 int count;
5788 long sys_time, user_time;
5789 char cdummy;
5790 int idummy;
5791 long ldummy;
5792 FILE *fp;
5794 // The /proc/<tid>/stat aggregates per-process usage on
5795 // new Linux kernels 2.6+ where NPTL is supported.
5796 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5797 // See bug 6328462.
5798 // There possibly can be cases where there is no directory
5799 // /proc/self/task, so we check its availability.
5800 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5801 // This is executed only once
5802 proc_task_unchecked = false;
5803 fp = fopen("/proc/self/task", "r");
5804 if (fp != NULL) {
5805 proc_stat_path = "/proc/self/task/%d/stat";
5806 fclose(fp);
5807 }
5808 }
5810 sprintf(proc_name, proc_stat_path, tid);
5811 fp = fopen(proc_name, "r");
5812 if ( fp == NULL ) return -1;
5813 statlen = fread(stat, 1, 2047, fp);
5814 stat[statlen] = '\0';
5815 fclose(fp);
5817 // Skip pid and the command string. Note that we could be dealing with
5818 // weird command names, e.g. user could decide to rename java launcher
5819 // to "java 1.4.2 :)", then the stat file would look like
5820 // 1234 (java 1.4.2 :)) R ... ...
5821 // We don't really need to know the command string, just find the last
5822 // occurrence of ")" and then start parsing from there. See bug 4726580.
5823 s = strrchr(stat, ')');
5824 if (s == NULL ) return -1;
5826 // Skip blank chars
5827 do s++; while (isspace(*s));
5829 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5830 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5831 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5832 &user_time, &sys_time);
5833 if ( count != 13 ) return -1;
5834 if (user_sys_cpu_time) {
5835 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5836 } else {
5837 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5838 }
5839 }
5840 PRAGMA_DIAG_POP
5842 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5843 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5844 info_ptr->may_skip_backward = false; // elapsed time not wall time
5845 info_ptr->may_skip_forward = false; // elapsed time not wall time
5846 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5847 }
5849 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5850 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5851 info_ptr->may_skip_backward = false; // elapsed time not wall time
5852 info_ptr->may_skip_forward = false; // elapsed time not wall time
5853 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5854 }
5856 bool os::is_thread_cpu_time_supported() {
5857 return true;
5858 }
5860 // System loadavg support. Returns -1 if load average cannot be obtained.
5861 // Linux doesn't yet have a (official) notion of processor sets,
5862 // so just return the system wide load average.
5863 int os::loadavg(double loadavg[], int nelem) {
5864 return ::getloadavg(loadavg, nelem);
5865 }
5867 void os::pause() {
5868 char filename[MAX_PATH];
5869 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5870 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5871 } else {
5872 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5873 }
5875 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5876 if (fd != -1) {
5877 struct stat buf;
5878 ::close(fd);
5879 while (::stat(filename, &buf) == 0) {
5880 (void)::poll(NULL, 0, 100);
5881 }
5882 } else {
5883 jio_fprintf(stderr,
5884 "Could not open pause file '%s', continuing immediately.\n", filename);
5885 }
5886 }
5889 // Refer to the comments in os_solaris.cpp park-unpark.
5890 //
5891 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5892 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5893 // For specifics regarding the bug see GLIBC BUGID 261237 :
5894 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5895 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5896 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5897 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5898 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5899 // and monitorenter when we're using 1-0 locking. All those operations may result in
5900 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5901 // of libpthread avoids the problem, but isn't practical.
5902 //
5903 // Possible remedies:
5904 //
5905 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5906 // This is palliative and probabilistic, however. If the thread is preempted
5907 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5908 // than the minimum period may have passed, and the abstime may be stale (in the
5909 // past) resultin in a hang. Using this technique reduces the odds of a hang
5910 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5911 //
5912 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5913 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5914 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5915 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5916 // thread.
5917 //
5918 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5919 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5920 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5921 // This also works well. In fact it avoids kernel-level scalability impediments
5922 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5923 // timers in a graceful fashion.
5924 //
5925 // 4. When the abstime value is in the past it appears that control returns
5926 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5927 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5928 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5929 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5930 // It may be possible to avoid reinitialization by checking the return
5931 // value from pthread_cond_timedwait(). In addition to reinitializing the
5932 // condvar we must establish the invariant that cond_signal() is only called
5933 // within critical sections protected by the adjunct mutex. This prevents
5934 // cond_signal() from "seeing" a condvar that's in the midst of being
5935 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5936 // desirable signal-after-unlock optimization that avoids futile context switching.
5937 //
5938 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5939 // structure when a condvar is used or initialized. cond_destroy() would
5940 // release the helper structure. Our reinitialize-after-timedwait fix
5941 // put excessive stress on malloc/free and locks protecting the c-heap.
5942 //
5943 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5944 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5945 // and only enabling the work-around for vulnerable environments.
5947 // utility to compute the abstime argument to timedwait:
5948 // millis is the relative timeout time
5949 // abstime will be the absolute timeout time
5950 // TODO: replace compute_abstime() with unpackTime()
5952 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5953 if (millis < 0) millis = 0;
5955 jlong seconds = millis / 1000;
5956 millis %= 1000;
5957 if (seconds > 50000000) { // see man cond_timedwait(3T)
5958 seconds = 50000000;
5959 }
5961 if (os::Linux::supports_monotonic_clock()) {
5962 struct timespec now;
5963 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5964 assert_status(status == 0, status, "clock_gettime");
5965 abstime->tv_sec = now.tv_sec + seconds;
5966 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5967 if (nanos >= NANOSECS_PER_SEC) {
5968 abstime->tv_sec += 1;
5969 nanos -= NANOSECS_PER_SEC;
5970 }
5971 abstime->tv_nsec = nanos;
5972 } else {
5973 struct timeval now;
5974 int status = gettimeofday(&now, NULL);
5975 assert(status == 0, "gettimeofday");
5976 abstime->tv_sec = now.tv_sec + seconds;
5977 long usec = now.tv_usec + millis * 1000;
5978 if (usec >= 1000000) {
5979 abstime->tv_sec += 1;
5980 usec -= 1000000;
5981 }
5982 abstime->tv_nsec = usec * 1000;
5983 }
5984 return abstime;
5985 }
5988 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5989 // Conceptually TryPark() should be equivalent to park(0).
5991 int os::PlatformEvent::TryPark() {
5992 for (;;) {
5993 const int v = _Event ;
5994 guarantee ((v == 0) || (v == 1), "invariant") ;
5995 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5996 }
5997 }
5999 void os::PlatformEvent::park() { // AKA "down()"
6000 // Invariant: Only the thread associated with the Event/PlatformEvent
6001 // may call park().
6002 // TODO: assert that _Assoc != NULL or _Assoc == Self
6003 int v ;
6004 for (;;) {
6005 v = _Event ;
6006 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6007 }
6008 guarantee (v >= 0, "invariant") ;
6009 if (v == 0) {
6010 // Do this the hard way by blocking ...
6011 int status = pthread_mutex_lock(_mutex);
6012 assert_status(status == 0, status, "mutex_lock");
6013 guarantee (_nParked == 0, "invariant") ;
6014 ++ _nParked ;
6015 while (_Event < 0) {
6016 status = pthread_cond_wait(_cond, _mutex);
6017 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6018 // Treat this the same as if the wait was interrupted
6019 if (status == ETIME) { status = EINTR; }
6020 assert_status(status == 0 || status == EINTR, status, "cond_wait");
6021 }
6022 -- _nParked ;
6024 _Event = 0 ;
6025 status = pthread_mutex_unlock(_mutex);
6026 assert_status(status == 0, status, "mutex_unlock");
6027 // Paranoia to ensure our locked and lock-free paths interact
6028 // correctly with each other.
6029 OrderAccess::fence();
6030 }
6031 guarantee (_Event >= 0, "invariant") ;
6032 }
6034 int os::PlatformEvent::park(jlong millis) {
6035 guarantee (_nParked == 0, "invariant") ;
6037 int v ;
6038 for (;;) {
6039 v = _Event ;
6040 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6041 }
6042 guarantee (v >= 0, "invariant") ;
6043 if (v != 0) return OS_OK ;
6045 // We do this the hard way, by blocking the thread.
6046 // Consider enforcing a minimum timeout value.
6047 struct timespec abst;
6048 compute_abstime(&abst, millis);
6050 int ret = OS_TIMEOUT;
6051 int status = pthread_mutex_lock(_mutex);
6052 assert_status(status == 0, status, "mutex_lock");
6053 guarantee (_nParked == 0, "invariant") ;
6054 ++_nParked ;
6056 // Object.wait(timo) will return because of
6057 // (a) notification
6058 // (b) timeout
6059 // (c) thread.interrupt
6060 //
6061 // Thread.interrupt and object.notify{All} both call Event::set.
6062 // That is, we treat thread.interrupt as a special case of notification.
6063 // The underlying Solaris implementation, cond_timedwait, admits
6064 // spurious/premature wakeups, but the JLS/JVM spec prevents the
6065 // JVM from making those visible to Java code. As such, we must
6066 // filter out spurious wakeups. We assume all ETIME returns are valid.
6067 //
6068 // TODO: properly differentiate simultaneous notify+interrupt.
6069 // In that case, we should propagate the notify to another waiter.
6071 while (_Event < 0) {
6072 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
6073 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6074 pthread_cond_destroy (_cond);
6075 pthread_cond_init (_cond, os::Linux::condAttr()) ;
6076 }
6077 assert_status(status == 0 || status == EINTR ||
6078 status == ETIME || status == ETIMEDOUT,
6079 status, "cond_timedwait");
6080 if (!FilterSpuriousWakeups) break ; // previous semantics
6081 if (status == ETIME || status == ETIMEDOUT) break ;
6082 // We consume and ignore EINTR and spurious wakeups.
6083 }
6084 --_nParked ;
6085 if (_Event >= 0) {
6086 ret = OS_OK;
6087 }
6088 _Event = 0 ;
6089 status = pthread_mutex_unlock(_mutex);
6090 assert_status(status == 0, status, "mutex_unlock");
6091 assert (_nParked == 0, "invariant") ;
6092 // Paranoia to ensure our locked and lock-free paths interact
6093 // correctly with each other.
6094 OrderAccess::fence();
6095 return ret;
6096 }
6098 void os::PlatformEvent::unpark() {
6099 // Transitions for _Event:
6100 // 0 :=> 1
6101 // 1 :=> 1
6102 // -1 :=> either 0 or 1; must signal target thread
6103 // That is, we can safely transition _Event from -1 to either
6104 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
6105 // unpark() calls.
6106 // See also: "Semaphores in Plan 9" by Mullender & Cox
6107 //
6108 // Note: Forcing a transition from "-1" to "1" on an unpark() means
6109 // that it will take two back-to-back park() calls for the owning
6110 // thread to block. This has the benefit of forcing a spurious return
6111 // from the first park() call after an unpark() call which will help
6112 // shake out uses of park() and unpark() without condition variables.
6114 if (Atomic::xchg(1, &_Event) >= 0) return;
6116 // Wait for the thread associated with the event to vacate
6117 int status = pthread_mutex_lock(_mutex);
6118 assert_status(status == 0, status, "mutex_lock");
6119 int AnyWaiters = _nParked;
6120 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6121 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
6122 AnyWaiters = 0;
6123 pthread_cond_signal(_cond);
6124 }
6125 status = pthread_mutex_unlock(_mutex);
6126 assert_status(status == 0, status, "mutex_unlock");
6127 if (AnyWaiters != 0) {
6128 status = pthread_cond_signal(_cond);
6129 assert_status(status == 0, status, "cond_signal");
6130 }
6132 // Note that we signal() _after dropping the lock for "immortal" Events.
6133 // This is safe and avoids a common class of futile wakeups. In rare
6134 // circumstances this can cause a thread to return prematurely from
6135 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
6136 // simply re-test the condition and re-park itself.
6137 }
6140 // JSR166
6141 // -------------------------------------------------------
6143 /*
6144 * The solaris and linux implementations of park/unpark are fairly
6145 * conservative for now, but can be improved. They currently use a
6146 * mutex/condvar pair, plus a a count.
6147 * Park decrements count if > 0, else does a condvar wait. Unpark
6148 * sets count to 1 and signals condvar. Only one thread ever waits
6149 * on the condvar. Contention seen when trying to park implies that someone
6150 * is unparking you, so don't wait. And spurious returns are fine, so there
6151 * is no need to track notifications.
6152 */
6154 /*
6155 * This code is common to linux and solaris and will be moved to a
6156 * common place in dolphin.
6157 *
6158 * The passed in time value is either a relative time in nanoseconds
6159 * or an absolute time in milliseconds. Either way it has to be unpacked
6160 * into suitable seconds and nanoseconds components and stored in the
6161 * given timespec structure.
6162 * Given time is a 64-bit value and the time_t used in the timespec is only
6163 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6164 * overflow if times way in the future are given. Further on Solaris versions
6165 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6166 * number of seconds, in abstime, is less than current_time + 100,000,000.
6167 * As it will be 28 years before "now + 100000000" will overflow we can
6168 * ignore overflow and just impose a hard-limit on seconds using the value
6169 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6170 * years from "now".
6171 */
6173 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6174 assert (time > 0, "convertTime");
6175 time_t max_secs = 0;
6177 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
6178 struct timeval now;
6179 int status = gettimeofday(&now, NULL);
6180 assert(status == 0, "gettimeofday");
6182 max_secs = now.tv_sec + MAX_SECS;
6184 if (isAbsolute) {
6185 jlong secs = time / 1000;
6186 if (secs > max_secs) {
6187 absTime->tv_sec = max_secs;
6188 } else {
6189 absTime->tv_sec = secs;
6190 }
6191 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6192 } else {
6193 jlong secs = time / NANOSECS_PER_SEC;
6194 if (secs >= MAX_SECS) {
6195 absTime->tv_sec = max_secs;
6196 absTime->tv_nsec = 0;
6197 } else {
6198 absTime->tv_sec = now.tv_sec + secs;
6199 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6200 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6201 absTime->tv_nsec -= NANOSECS_PER_SEC;
6202 ++absTime->tv_sec; // note: this must be <= max_secs
6203 }
6204 }
6205 }
6206 } else {
6207 // must be relative using monotonic clock
6208 struct timespec now;
6209 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6210 assert_status(status == 0, status, "clock_gettime");
6211 max_secs = now.tv_sec + MAX_SECS;
6212 jlong secs = time / NANOSECS_PER_SEC;
6213 if (secs >= MAX_SECS) {
6214 absTime->tv_sec = max_secs;
6215 absTime->tv_nsec = 0;
6216 } else {
6217 absTime->tv_sec = now.tv_sec + secs;
6218 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
6219 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6220 absTime->tv_nsec -= NANOSECS_PER_SEC;
6221 ++absTime->tv_sec; // note: this must be <= max_secs
6222 }
6223 }
6224 }
6225 assert(absTime->tv_sec >= 0, "tv_sec < 0");
6226 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6227 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6228 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6229 }
6231 void Parker::park(bool isAbsolute, jlong time) {
6232 // Ideally we'd do something useful while spinning, such
6233 // as calling unpackTime().
6235 // Optional fast-path check:
6236 // Return immediately if a permit is available.
6237 // We depend on Atomic::xchg() having full barrier semantics
6238 // since we are doing a lock-free update to _counter.
6239 if (Atomic::xchg(0, &_counter) > 0) return;
6241 Thread* thread = Thread::current();
6242 assert(thread->is_Java_thread(), "Must be JavaThread");
6243 JavaThread *jt = (JavaThread *)thread;
6245 // Optional optimization -- avoid state transitions if there's an interrupt pending.
6246 // Check interrupt before trying to wait
6247 if (Thread::is_interrupted(thread, false)) {
6248 return;
6249 }
6251 // Next, demultiplex/decode time arguments
6252 timespec absTime;
6253 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6254 return;
6255 }
6256 if (time > 0) {
6257 unpackTime(&absTime, isAbsolute, time);
6258 }
6261 // Enter safepoint region
6262 // Beware of deadlocks such as 6317397.
6263 // The per-thread Parker:: mutex is a classic leaf-lock.
6264 // In particular a thread must never block on the Threads_lock while
6265 // holding the Parker:: mutex. If safepoints are pending both the
6266 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6267 ThreadBlockInVM tbivm(jt);
6269 // Don't wait if cannot get lock since interference arises from
6270 // unblocking. Also. check interrupt before trying wait
6271 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6272 return;
6273 }
6275 int status ;
6276 if (_counter > 0) { // no wait needed
6277 _counter = 0;
6278 status = pthread_mutex_unlock(_mutex);
6279 assert (status == 0, "invariant") ;
6280 // Paranoia to ensure our locked and lock-free paths interact
6281 // correctly with each other and Java-level accesses.
6282 OrderAccess::fence();
6283 return;
6284 }
6286 #ifdef ASSERT
6287 // Don't catch signals while blocked; let the running threads have the signals.
6288 // (This allows a debugger to break into the running thread.)
6289 sigset_t oldsigs;
6290 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6291 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6292 #endif
6294 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6295 jt->set_suspend_equivalent();
6296 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6298 assert(_cur_index == -1, "invariant");
6299 if (time == 0) {
6300 _cur_index = REL_INDEX; // arbitrary choice when not timed
6301 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6302 } else {
6303 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6304 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6305 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6306 pthread_cond_destroy (&_cond[_cur_index]) ;
6307 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6308 }
6309 }
6310 _cur_index = -1;
6311 assert_status(status == 0 || status == EINTR ||
6312 status == ETIME || status == ETIMEDOUT,
6313 status, "cond_timedwait");
6315 #ifdef ASSERT
6316 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6317 #endif
6319 _counter = 0 ;
6320 status = pthread_mutex_unlock(_mutex) ;
6321 assert_status(status == 0, status, "invariant") ;
6322 // Paranoia to ensure our locked and lock-free paths interact
6323 // correctly with each other and Java-level accesses.
6324 OrderAccess::fence();
6326 // If externally suspended while waiting, re-suspend
6327 if (jt->handle_special_suspend_equivalent_condition()) {
6328 jt->java_suspend_self();
6329 }
6330 }
6332 void Parker::unpark() {
6333 int s, status ;
6334 status = pthread_mutex_lock(_mutex);
6335 assert (status == 0, "invariant") ;
6336 s = _counter;
6337 _counter = 1;
6338 if (s < 1) {
6339 // thread might be parked
6340 if (_cur_index != -1) {
6341 // thread is definitely parked
6342 if (WorkAroundNPTLTimedWaitHang) {
6343 status = pthread_cond_signal (&_cond[_cur_index]);
6344 assert (status == 0, "invariant");
6345 status = pthread_mutex_unlock(_mutex);
6346 assert (status == 0, "invariant");
6347 } else {
6348 // must capture correct index before unlocking
6349 int index = _cur_index;
6350 status = pthread_mutex_unlock(_mutex);
6351 assert (status == 0, "invariant");
6352 status = pthread_cond_signal (&_cond[index]);
6353 assert (status == 0, "invariant");
6354 }
6355 } else {
6356 pthread_mutex_unlock(_mutex);
6357 assert (status == 0, "invariant") ;
6358 }
6359 } else {
6360 pthread_mutex_unlock(_mutex);
6361 assert (status == 0, "invariant") ;
6362 }
6363 }
6366 extern char** environ;
6368 // Run the specified command in a separate process. Return its exit value,
6369 // or -1 on failure (e.g. can't fork a new process).
6370 // Unlike system(), this function can be called from signal handler. It
6371 // doesn't block SIGINT et al.
6372 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
6373 const char * argv[4] = {"sh", "-c", cmd, NULL};
6375 pid_t pid ;
6377 if (use_vfork_if_available) {
6378 pid = vfork();
6379 } else {
6380 pid = fork();
6381 }
6383 if (pid < 0) {
6384 // fork failed
6385 return -1;
6387 } else if (pid == 0) {
6388 // child process
6390 execve("/bin/sh", (char* const*)argv, environ);
6392 // execve failed
6393 _exit(-1);
6395 } else {
6396 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6397 // care about the actual exit code, for now.
6399 int status;
6401 // Wait for the child process to exit. This returns immediately if
6402 // the child has already exited. */
6403 while (waitpid(pid, &status, 0) < 0) {
6404 switch (errno) {
6405 case ECHILD: return 0;
6406 case EINTR: break;
6407 default: return -1;
6408 }
6409 }
6411 if (WIFEXITED(status)) {
6412 // The child exited normally; get its exit code.
6413 return WEXITSTATUS(status);
6414 } else if (WIFSIGNALED(status)) {
6415 // The child exited because of a signal
6416 // The best value to return is 0x80 + signal number,
6417 // because that is what all Unix shells do, and because
6418 // it allows callers to distinguish between process exit and
6419 // process death by signal.
6420 return 0x80 + WTERMSIG(status);
6421 } else {
6422 // Unknown exit code; pass it through
6423 return status;
6424 }
6425 }
6426 }
6428 // is_headless_jre()
6429 //
6430 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6431 // in order to report if we are running in a headless jre
6432 //
6433 // Since JDK8 xawt/libmawt.so was moved into the same directory
6434 // as libawt.so, and renamed libawt_xawt.so
6435 //
6436 bool os::is_headless_jre() {
6437 struct stat statbuf;
6438 char buf[MAXPATHLEN];
6439 char libmawtpath[MAXPATHLEN];
6440 const char *xawtstr = "/xawt/libmawt.so";
6441 const char *new_xawtstr = "/libawt_xawt.so";
6442 char *p;
6444 // Get path to libjvm.so
6445 os::jvm_path(buf, sizeof(buf));
6447 // Get rid of libjvm.so
6448 p = strrchr(buf, '/');
6449 if (p == NULL) return false;
6450 else *p = '\0';
6452 // Get rid of client or server
6453 p = strrchr(buf, '/');
6454 if (p == NULL) return false;
6455 else *p = '\0';
6457 // check xawt/libmawt.so
6458 strcpy(libmawtpath, buf);
6459 strcat(libmawtpath, xawtstr);
6460 if (::stat(libmawtpath, &statbuf) == 0) return false;
6462 // check libawt_xawt.so
6463 strcpy(libmawtpath, buf);
6464 strcat(libmawtpath, new_xawtstr);
6465 if (::stat(libmawtpath, &statbuf) == 0) return false;
6467 return true;
6468 }
6470 // Get the default path to the core file
6471 // Returns the length of the string
6472 int os::get_core_path(char* buffer, size_t bufferSize) {
6473 const char* p = get_current_directory(buffer, bufferSize);
6475 if (p == NULL) {
6476 assert(p != NULL, "failed to get current directory");
6477 return 0;
6478 }
6480 return strlen(buffer);
6481 }
6483 /////////////// Unit tests ///////////////
6485 #ifndef PRODUCT
6487 #define test_log(...) \
6488 do {\
6489 if (VerboseInternalVMTests) { \
6490 tty->print_cr(__VA_ARGS__); \
6491 tty->flush(); \
6492 }\
6493 } while (false)
6495 class TestReserveMemorySpecial : AllStatic {
6496 public:
6497 static void small_page_write(void* addr, size_t size) {
6498 size_t page_size = os::vm_page_size();
6500 char* end = (char*)addr + size;
6501 for (char* p = (char*)addr; p < end; p += page_size) {
6502 *p = 1;
6503 }
6504 }
6506 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6507 if (!UseHugeTLBFS) {
6508 return;
6509 }
6511 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6513 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6515 if (addr != NULL) {
6516 small_page_write(addr, size);
6518 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6519 }
6520 }
6522 static void test_reserve_memory_special_huge_tlbfs_only() {
6523 if (!UseHugeTLBFS) {
6524 return;
6525 }
6527 size_t lp = os::large_page_size();
6529 for (size_t size = lp; size <= lp * 10; size += lp) {
6530 test_reserve_memory_special_huge_tlbfs_only(size);
6531 }
6532 }
6534 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6535 size_t lp = os::large_page_size();
6536 size_t ag = os::vm_allocation_granularity();
6538 // sizes to test
6539 const size_t sizes[] = {
6540 lp, lp + ag, lp + lp / 2, lp * 2,
6541 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6542 lp * 10, lp * 10 + lp / 2
6543 };
6544 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6546 // For each size/alignment combination, we test three scenarios:
6547 // 1) with req_addr == NULL
6548 // 2) with a non-null req_addr at which we expect to successfully allocate
6549 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6550 // expect the allocation to either fail or to ignore req_addr
6552 // Pre-allocate two areas; they shall be as large as the largest allocation
6553 // and aligned to the largest alignment we will be testing.
6554 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6555 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6556 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6557 -1, 0);
6558 assert(mapping1 != MAP_FAILED, "should work");
6560 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6561 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6562 -1, 0);
6563 assert(mapping2 != MAP_FAILED, "should work");
6565 // Unmap the first mapping, but leave the second mapping intact: the first
6566 // mapping will serve as a value for a "good" req_addr (case 2). The second
6567 // mapping, still intact, as "bad" req_addr (case 3).
6568 ::munmap(mapping1, mapping_size);
6570 // Case 1
6571 test_log("%s, req_addr NULL:", __FUNCTION__);
6572 test_log("size align result");
6574 for (int i = 0; i < num_sizes; i++) {
6575 const size_t size = sizes[i];
6576 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6577 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6578 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6579 size, alignment, p, (p != NULL ? "" : "(failed)"));
6580 if (p != NULL) {
6581 assert(is_ptr_aligned(p, alignment), "must be");
6582 small_page_write(p, size);
6583 os::Linux::release_memory_special_huge_tlbfs(p, size);
6584 }
6585 }
6586 }
6588 // Case 2
6589 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6590 test_log("size align req_addr result");
6592 for (int i = 0; i < num_sizes; i++) {
6593 const size_t size = sizes[i];
6594 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6595 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6596 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6597 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6598 size, alignment, req_addr, p,
6599 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6600 if (p != NULL) {
6601 assert(p == req_addr, "must be");
6602 small_page_write(p, size);
6603 os::Linux::release_memory_special_huge_tlbfs(p, size);
6604 }
6605 }
6606 }
6608 // Case 3
6609 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6610 test_log("size align req_addr result");
6612 for (int i = 0; i < num_sizes; i++) {
6613 const size_t size = sizes[i];
6614 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6615 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6616 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6617 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6618 size, alignment, req_addr, p,
6619 ((p != NULL ? "" : "(failed)")));
6620 // as the area around req_addr contains already existing mappings, the API should always
6621 // return NULL (as per contract, it cannot return another address)
6622 assert(p == NULL, "must be");
6623 }
6624 }
6626 ::munmap(mapping2, mapping_size);
6628 }
6630 static void test_reserve_memory_special_huge_tlbfs() {
6631 if (!UseHugeTLBFS) {
6632 return;
6633 }
6635 test_reserve_memory_special_huge_tlbfs_only();
6636 test_reserve_memory_special_huge_tlbfs_mixed();
6637 }
6639 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6640 if (!UseSHM) {
6641 return;
6642 }
6644 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6646 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6648 if (addr != NULL) {
6649 assert(is_ptr_aligned(addr, alignment), "Check");
6650 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6652 small_page_write(addr, size);
6654 os::Linux::release_memory_special_shm(addr, size);
6655 }
6656 }
6658 static void test_reserve_memory_special_shm() {
6659 size_t lp = os::large_page_size();
6660 size_t ag = os::vm_allocation_granularity();
6662 for (size_t size = ag; size < lp * 3; size += ag) {
6663 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6664 test_reserve_memory_special_shm(size, alignment);
6665 }
6666 }
6667 }
6669 static void test() {
6670 test_reserve_memory_special_huge_tlbfs();
6671 test_reserve_memory_special_shm();
6672 }
6673 };
6675 void TestReserveMemorySpecial_test() {
6676 TestReserveMemorySpecial::test();
6677 }
6679 #endif