Wed, 14 Oct 2020 17:44:48 +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 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2161 FILE *procmapsFile = NULL;
2163 // Open the procfs maps file for the current process
2164 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2165 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2166 char line[PATH_MAX + 100];
2168 // Read line by line from 'file'
2169 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2170 u8 base, top, offset, inode;
2171 char permissions[5];
2172 char device[6];
2173 char name[PATH_MAX + 1];
2175 // Parse fields from line
2176 sscanf(line, UINT64_FORMAT_X "-" UINT64_FORMAT_X " %4s " UINT64_FORMAT_X " %7s " INT64_FORMAT " %s",
2177 &base, &top, permissions, &offset, device, &inode, name);
2179 // Filter by device id '00:00' so that we only get file system mapped files.
2180 if (strcmp(device, "00:00") != 0) {
2182 // Call callback with the fields of interest
2183 if(callback(name, (address)base, (address)top, param)) {
2184 // Oops abort, callback aborted
2185 fclose(procmapsFile);
2186 return 1;
2187 }
2188 }
2189 }
2190 fclose(procmapsFile);
2191 }
2192 return 0;
2193 }
2195 void os::print_os_info_brief(outputStream* st) {
2196 os::Linux::print_distro_info(st);
2198 os::Posix::print_uname_info(st);
2200 os::Linux::print_libversion_info(st);
2202 }
2204 void os::print_os_info(outputStream* st) {
2205 st->print("OS:");
2207 os::Linux::print_distro_info(st);
2209 os::Posix::print_uname_info(st);
2211 // Print warning if unsafe chroot environment detected
2212 if (unsafe_chroot_detected) {
2213 st->print("WARNING!! ");
2214 st->print_cr("%s", unstable_chroot_error);
2215 }
2217 os::Linux::print_libversion_info(st);
2219 os::Posix::print_rlimit_info(st);
2221 os::Posix::print_load_average(st);
2223 os::Linux::print_full_memory_info(st);
2225 os::Linux::print_container_info(st);
2226 }
2228 // Try to identify popular distros.
2229 // Most Linux distributions have a /etc/XXX-release file, which contains
2230 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2231 // file that also contains the OS version string. Some have more than one
2232 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2233 // /etc/redhat-release.), so the order is important.
2234 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2235 // their own specific XXX-release file as well as a redhat-release file.
2236 // Because of this the XXX-release file needs to be searched for before the
2237 // redhat-release file.
2238 // Since Red Hat has a lsb-release file that is not very descriptive the
2239 // search for redhat-release needs to be before lsb-release.
2240 // Since the lsb-release file is the new standard it needs to be searched
2241 // before the older style release files.
2242 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2243 // next to last resort. The os-release file is a new standard that contains
2244 // distribution information and the system-release file seems to be an old
2245 // standard that has been replaced by the lsb-release and os-release files.
2246 // Searching for the debian_version file is the last resort. It contains
2247 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2248 // "Debian " is printed before the contents of the debian_version file.
2249 void os::Linux::print_distro_info(outputStream* st) {
2250 if (!_print_ascii_file("/etc/oracle-release", st) &&
2251 !_print_ascii_file("/etc/mandriva-release", st) &&
2252 !_print_ascii_file("/etc/mandrake-release", st) &&
2253 !_print_ascii_file("/etc/sun-release", st) &&
2254 !_print_ascii_file("/etc/redhat-release", st) &&
2255 !_print_ascii_file("/etc/lsb-release", st) &&
2256 !_print_ascii_file("/etc/SuSE-release", st) &&
2257 !_print_ascii_file("/etc/turbolinux-release", st) &&
2258 !_print_ascii_file("/etc/gentoo-release", st) &&
2259 !_print_ascii_file("/etc/ltib-release", st) &&
2260 !_print_ascii_file("/etc/angstrom-version", st) &&
2261 !_print_ascii_file("/etc/system-release", st) &&
2262 !_print_ascii_file("/etc/os-release", st)) {
2264 if (file_exists("/etc/debian_version")) {
2265 st->print("Debian ");
2266 _print_ascii_file("/etc/debian_version", st);
2267 } else {
2268 st->print("Linux");
2269 }
2270 }
2271 st->cr();
2272 }
2274 void os::Linux::print_libversion_info(outputStream* st) {
2275 // libc, pthread
2276 st->print("libc:");
2277 st->print("%s ", os::Linux::glibc_version());
2278 st->print("%s ", os::Linux::libpthread_version());
2279 if (os::Linux::is_LinuxThreads()) {
2280 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2281 }
2282 st->cr();
2283 }
2285 void os::Linux::print_full_memory_info(outputStream* st) {
2286 st->print("\n/proc/meminfo:\n");
2287 _print_ascii_file("/proc/meminfo", st);
2288 st->cr();
2289 }
2291 void os::Linux::print_container_info(outputStream* st) {
2292 if (!OSContainer::is_containerized()) {
2293 return;
2294 }
2296 st->print("container (cgroup) information:\n");
2298 const char *p_ct = OSContainer::container_type();
2299 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2301 char *p = OSContainer::cpu_cpuset_cpus();
2302 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2303 free(p);
2305 p = OSContainer::cpu_cpuset_memory_nodes();
2306 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2307 free(p);
2309 int i = OSContainer::active_processor_count();
2310 if (i > 0) {
2311 st->print("active_processor_count: %d\n", i);
2312 } else {
2313 st->print("active_processor_count: failed\n");
2314 }
2316 i = OSContainer::cpu_quota();
2317 st->print("cpu_quota: %d\n", i);
2319 i = OSContainer::cpu_period();
2320 st->print("cpu_period: %d\n", i);
2322 i = OSContainer::cpu_shares();
2323 st->print("cpu_shares: %d\n", i);
2325 jlong j = OSContainer::memory_limit_in_bytes();
2326 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2328 j = OSContainer::memory_and_swap_limit_in_bytes();
2329 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2331 j = OSContainer::memory_soft_limit_in_bytes();
2332 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2334 j = OSContainer::OSContainer::memory_usage_in_bytes();
2335 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2337 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2338 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2339 st->cr();
2340 }
2342 void os::print_memory_info(outputStream* st) {
2344 st->print("Memory:");
2345 st->print(" %dk page", os::vm_page_size()>>10);
2347 // values in struct sysinfo are "unsigned long"
2348 struct sysinfo si;
2349 sysinfo(&si);
2351 st->print(", physical " UINT64_FORMAT "k",
2352 os::physical_memory() >> 10);
2353 st->print("(" UINT64_FORMAT "k free)",
2354 os::available_memory() >> 10);
2355 st->print(", swap " UINT64_FORMAT "k",
2356 ((jlong)si.totalswap * si.mem_unit) >> 10);
2357 st->print("(" UINT64_FORMAT "k free)",
2358 ((jlong)si.freeswap * si.mem_unit) >> 10);
2359 st->cr();
2360 }
2362 void os::pd_print_cpu_info(outputStream* st) {
2363 st->print("\n/proc/cpuinfo:\n");
2364 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2365 st->print(" <Not Available>");
2366 }
2367 st->cr();
2368 }
2370 void os::print_siginfo(outputStream* st, void* siginfo) {
2371 const siginfo_t* si = (const siginfo_t*)siginfo;
2373 os::Posix::print_siginfo_brief(st, si);
2374 #if INCLUDE_CDS
2375 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2376 UseSharedSpaces) {
2377 FileMapInfo* mapinfo = FileMapInfo::current_info();
2378 if (mapinfo->is_in_shared_space(si->si_addr)) {
2379 st->print("\n\nError accessing class data sharing archive." \
2380 " Mapped file inaccessible during execution, " \
2381 " possible disk/network problem.");
2382 }
2383 }
2384 #endif
2385 st->cr();
2386 }
2389 static void print_signal_handler(outputStream* st, int sig,
2390 char* buf, size_t buflen);
2392 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2393 st->print_cr("Signal Handlers:");
2394 print_signal_handler(st, SIGSEGV, buf, buflen);
2395 print_signal_handler(st, SIGBUS , buf, buflen);
2396 print_signal_handler(st, SIGFPE , buf, buflen);
2397 print_signal_handler(st, SIGPIPE, buf, buflen);
2398 print_signal_handler(st, SIGXFSZ, buf, buflen);
2399 print_signal_handler(st, SIGILL , buf, buflen);
2400 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2401 print_signal_handler(st, SR_signum, buf, buflen);
2402 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2403 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2404 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2405 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2406 #if defined(PPC64)
2407 print_signal_handler(st, SIGTRAP, buf, buflen);
2408 #endif
2409 }
2411 static char saved_jvm_path[MAXPATHLEN] = {0};
2413 // Find the full path to the current module, libjvm.so
2414 void os::jvm_path(char *buf, jint buflen) {
2415 // Error checking.
2416 if (buflen < MAXPATHLEN) {
2417 assert(false, "must use a large-enough buffer");
2418 buf[0] = '\0';
2419 return;
2420 }
2421 // Lazy resolve the path to current module.
2422 if (saved_jvm_path[0] != 0) {
2423 strcpy(buf, saved_jvm_path);
2424 return;
2425 }
2427 char dli_fname[MAXPATHLEN];
2428 bool ret = dll_address_to_library_name(
2429 CAST_FROM_FN_PTR(address, os::jvm_path),
2430 dli_fname, sizeof(dli_fname), NULL);
2431 assert(ret, "cannot locate libjvm");
2432 char *rp = NULL;
2433 if (ret && dli_fname[0] != '\0') {
2434 rp = realpath(dli_fname, buf);
2435 }
2436 if (rp == NULL)
2437 return;
2439 if (Arguments::created_by_gamma_launcher()) {
2440 // Support for the gamma launcher. Typical value for buf is
2441 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2442 // the right place in the string, then assume we are installed in a JDK and
2443 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2444 // up the path so it looks like libjvm.so is installed there (append a
2445 // fake suffix hotspot/libjvm.so).
2446 const char *p = buf + strlen(buf) - 1;
2447 for (int count = 0; p > buf && count < 5; ++count) {
2448 for (--p; p > buf && *p != '/'; --p)
2449 /* empty */ ;
2450 }
2452 if (strncmp(p, "/jre/lib/", 9) != 0) {
2453 // Look for JAVA_HOME in the environment.
2454 char* java_home_var = ::getenv("JAVA_HOME");
2455 if (java_home_var != NULL && java_home_var[0] != 0) {
2456 char* jrelib_p;
2457 int len;
2459 // Check the current module name "libjvm.so".
2460 p = strrchr(buf, '/');
2461 assert(strstr(p, "/libjvm") == p, "invalid library name");
2463 rp = realpath(java_home_var, buf);
2464 if (rp == NULL)
2465 return;
2467 // determine if this is a legacy image or modules image
2468 // modules image doesn't have "jre" subdirectory
2469 len = strlen(buf);
2470 assert(len < buflen, "Ran out of buffer room");
2471 jrelib_p = buf + len;
2472 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2473 if (0 != access(buf, F_OK)) {
2474 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2475 }
2477 if (0 == access(buf, F_OK)) {
2478 // Use current module name "libjvm.so"
2479 len = strlen(buf);
2480 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2481 } else {
2482 // Go back to path of .so
2483 rp = realpath(dli_fname, buf);
2484 if (rp == NULL)
2485 return;
2486 }
2487 }
2488 }
2489 }
2491 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2492 }
2494 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2495 // no prefix required, not even "_"
2496 }
2498 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2499 // no suffix required
2500 }
2502 ////////////////////////////////////////////////////////////////////////////////
2503 // sun.misc.Signal support
2505 static volatile jint sigint_count = 0;
2507 static void
2508 UserHandler(int sig, void *siginfo, void *context) {
2509 // 4511530 - sem_post is serialized and handled by the manager thread. When
2510 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2511 // don't want to flood the manager thread with sem_post requests.
2512 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2513 return;
2515 // Ctrl-C is pressed during error reporting, likely because the error
2516 // handler fails to abort. Let VM die immediately.
2517 if (sig == SIGINT && is_error_reported()) {
2518 os::die();
2519 }
2521 os::signal_notify(sig);
2522 }
2524 void* os::user_handler() {
2525 return CAST_FROM_FN_PTR(void*, UserHandler);
2526 }
2528 class Semaphore : public StackObj {
2529 public:
2530 Semaphore();
2531 ~Semaphore();
2532 void signal();
2533 void wait();
2534 bool trywait();
2535 bool timedwait(unsigned int sec, int nsec);
2536 private:
2537 sem_t _semaphore;
2538 };
2540 Semaphore::Semaphore() {
2541 sem_init(&_semaphore, 0, 0);
2542 }
2544 Semaphore::~Semaphore() {
2545 sem_destroy(&_semaphore);
2546 }
2548 void Semaphore::signal() {
2549 sem_post(&_semaphore);
2550 }
2552 void Semaphore::wait() {
2553 sem_wait(&_semaphore);
2554 }
2556 bool Semaphore::trywait() {
2557 return sem_trywait(&_semaphore) == 0;
2558 }
2560 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2562 struct timespec ts;
2563 // Semaphore's are always associated with CLOCK_REALTIME
2564 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2565 // see unpackTime for discussion on overflow checking
2566 if (sec >= MAX_SECS) {
2567 ts.tv_sec += MAX_SECS;
2568 ts.tv_nsec = 0;
2569 } else {
2570 ts.tv_sec += sec;
2571 ts.tv_nsec += nsec;
2572 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2573 ts.tv_nsec -= NANOSECS_PER_SEC;
2574 ++ts.tv_sec; // note: this must be <= max_secs
2575 }
2576 }
2578 while (1) {
2579 int result = sem_timedwait(&_semaphore, &ts);
2580 if (result == 0) {
2581 return true;
2582 } else if (errno == EINTR) {
2583 continue;
2584 } else if (errno == ETIMEDOUT) {
2585 return false;
2586 } else {
2587 return false;
2588 }
2589 }
2590 }
2592 extern "C" {
2593 typedef void (*sa_handler_t)(int);
2594 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2595 }
2597 void* os::signal(int signal_number, void* handler) {
2598 struct sigaction sigAct, oldSigAct;
2600 sigfillset(&(sigAct.sa_mask));
2601 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2602 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2604 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2605 // -1 means registration failed
2606 return (void *)-1;
2607 }
2609 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2610 }
2612 void os::signal_raise(int signal_number) {
2613 ::raise(signal_number);
2614 }
2616 /*
2617 * The following code is moved from os.cpp for making this
2618 * code platform specific, which it is by its very nature.
2619 */
2621 // Will be modified when max signal is changed to be dynamic
2622 int os::sigexitnum_pd() {
2623 return NSIG;
2624 }
2626 // a counter for each possible signal value
2627 static volatile jint pending_signals[NSIG+1] = { 0 };
2629 // Linux(POSIX) specific hand shaking semaphore.
2630 static sem_t sig_sem;
2631 static Semaphore sr_semaphore;
2633 void os::signal_init_pd() {
2634 // Initialize signal structures
2635 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2637 // Initialize signal semaphore
2638 ::sem_init(&sig_sem, 0, 0);
2639 }
2641 void os::signal_notify(int sig) {
2642 Atomic::inc(&pending_signals[sig]);
2643 ::sem_post(&sig_sem);
2644 }
2646 static int check_pending_signals(bool wait) {
2647 Atomic::store(0, &sigint_count);
2648 for (;;) {
2649 for (int i = 0; i < NSIG + 1; i++) {
2650 jint n = pending_signals[i];
2651 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2652 return i;
2653 }
2654 }
2655 if (!wait) {
2656 return -1;
2657 }
2658 JavaThread *thread = JavaThread::current();
2659 ThreadBlockInVM tbivm(thread);
2661 bool threadIsSuspended;
2662 do {
2663 thread->set_suspend_equivalent();
2664 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2665 ::sem_wait(&sig_sem);
2667 // were we externally suspended while we were waiting?
2668 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2669 if (threadIsSuspended) {
2670 //
2671 // The semaphore has been incremented, but while we were waiting
2672 // another thread suspended us. We don't want to continue running
2673 // while suspended because that would surprise the thread that
2674 // suspended us.
2675 //
2676 ::sem_post(&sig_sem);
2678 thread->java_suspend_self();
2679 }
2680 } while (threadIsSuspended);
2681 }
2682 }
2684 int os::signal_lookup() {
2685 return check_pending_signals(false);
2686 }
2688 int os::signal_wait() {
2689 return check_pending_signals(true);
2690 }
2692 ////////////////////////////////////////////////////////////////////////////////
2693 // Virtual Memory
2695 int os::vm_page_size() {
2696 // Seems redundant as all get out
2697 assert(os::Linux::page_size() != -1, "must call os::init");
2698 return os::Linux::page_size();
2699 }
2701 // Solaris allocates memory by pages.
2702 int os::vm_allocation_granularity() {
2703 assert(os::Linux::page_size() != -1, "must call os::init");
2704 return os::Linux::page_size();
2705 }
2707 // Rationale behind this function:
2708 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2709 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2710 // samples for JITted code. Here we create private executable mapping over the code cache
2711 // and then we can use standard (well, almost, as mapping can change) way to provide
2712 // info for the reporting script by storing timestamp and location of symbol
2713 void linux_wrap_code(char* base, size_t size) {
2714 static volatile jint cnt = 0;
2716 if (!UseOprofile) {
2717 return;
2718 }
2720 char buf[PATH_MAX+1];
2721 int num = Atomic::add(1, &cnt);
2723 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2724 os::get_temp_directory(), os::current_process_id(), num);
2725 unlink(buf);
2727 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2729 if (fd != -1) {
2730 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2731 if (rv != (off_t)-1) {
2732 if (::write(fd, "", 1) == 1) {
2733 mmap(base, size,
2734 PROT_READ|PROT_WRITE|PROT_EXEC,
2735 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2736 }
2737 }
2738 ::close(fd);
2739 unlink(buf);
2740 }
2741 }
2743 static bool recoverable_mmap_error(int err) {
2744 // See if the error is one we can let the caller handle. This
2745 // list of errno values comes from JBS-6843484. I can't find a
2746 // Linux man page that documents this specific set of errno
2747 // values so while this list currently matches Solaris, it may
2748 // change as we gain experience with this failure mode.
2749 switch (err) {
2750 case EBADF:
2751 case EINVAL:
2752 case ENOTSUP:
2753 // let the caller deal with these errors
2754 return true;
2756 default:
2757 // Any remaining errors on this OS can cause our reserved mapping
2758 // to be lost. That can cause confusion where different data
2759 // structures think they have the same memory mapped. The worst
2760 // scenario is if both the VM and a library think they have the
2761 // same memory mapped.
2762 return false;
2763 }
2764 }
2766 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2767 int err) {
2768 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2769 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2770 strerror(err), err);
2771 }
2773 static void warn_fail_commit_memory(char* addr, size_t size,
2774 size_t alignment_hint, bool exec,
2775 int err) {
2776 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2777 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2778 alignment_hint, exec, strerror(err), err);
2779 }
2781 // NOTE: Linux kernel does not really reserve the pages for us.
2782 // All it does is to check if there are enough free pages
2783 // left at the time of mmap(). This could be a potential
2784 // problem.
2785 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2786 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2787 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2788 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2789 if (res != (uintptr_t) MAP_FAILED) {
2790 if (UseNUMAInterleaving) {
2791 numa_make_global(addr, size);
2792 }
2793 return 0;
2794 }
2796 int err = errno; // save errno from mmap() call above
2798 if (!recoverable_mmap_error(err)) {
2799 warn_fail_commit_memory(addr, size, exec, err);
2800 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2801 }
2803 return err;
2804 }
2806 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2807 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2808 }
2810 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2811 const char* mesg) {
2812 assert(mesg != NULL, "mesg must be specified");
2813 int err = os::Linux::commit_memory_impl(addr, size, exec);
2814 if (err != 0) {
2815 // the caller wants all commit errors to exit with the specified mesg:
2816 warn_fail_commit_memory(addr, size, exec, err);
2817 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2818 }
2819 }
2821 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2822 #ifndef MAP_HUGETLB
2823 #define MAP_HUGETLB 0x40000
2824 #endif
2826 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2827 #ifndef MADV_HUGEPAGE
2828 #define MADV_HUGEPAGE 14
2829 #endif
2831 int os::Linux::commit_memory_impl(char* addr, size_t size,
2832 size_t alignment_hint, bool exec) {
2833 int err = os::Linux::commit_memory_impl(addr, size, exec);
2834 if (err == 0) {
2835 realign_memory(addr, size, alignment_hint);
2836 }
2837 return err;
2838 }
2840 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2841 bool exec) {
2842 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2843 }
2845 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2846 size_t alignment_hint, bool exec,
2847 const char* mesg) {
2848 assert(mesg != NULL, "mesg must be specified");
2849 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2850 if (err != 0) {
2851 // the caller wants all commit errors to exit with the specified mesg:
2852 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2853 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2854 }
2855 }
2857 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2858 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2859 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2860 // be supported or the memory may already be backed by huge pages.
2861 ::madvise(addr, bytes, MADV_HUGEPAGE);
2862 }
2863 }
2865 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2866 // This method works by doing an mmap over an existing mmaping and effectively discarding
2867 // the existing pages. However it won't work for SHM-based large pages that cannot be
2868 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2869 // small pages on top of the SHM segment. This method always works for small pages, so we
2870 // allow that in any case.
2871 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2872 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2873 }
2874 }
2876 void os::numa_make_global(char *addr, size_t bytes) {
2877 Linux::numa_interleave_memory(addr, bytes);
2878 }
2880 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2881 // bind policy to MPOL_PREFERRED for the current thread.
2882 #define USE_MPOL_PREFERRED 0
2884 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2885 // To make NUMA and large pages more robust when both enabled, we need to ease
2886 // the requirements on where the memory should be allocated. MPOL_BIND is the
2887 // default policy and it will force memory to be allocated on the specified
2888 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2889 // the specified node, but will not force it. Using this policy will prevent
2890 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2891 // free large pages.
2892 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2893 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2894 }
2896 bool os::numa_topology_changed() { return false; }
2898 size_t os::numa_get_groups_num() {
2899 // Return just the number of nodes in which it's possible to allocate memory
2900 // (in numa terminology, configured nodes).
2901 return Linux::numa_num_configured_nodes();
2902 }
2904 int os::numa_get_group_id() {
2905 int cpu_id = Linux::sched_getcpu();
2906 if (cpu_id != -1) {
2907 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2908 if (lgrp_id != -1) {
2909 return lgrp_id;
2910 }
2911 }
2912 return 0;
2913 }
2915 int os::Linux::get_existing_num_nodes() {
2916 size_t node;
2917 size_t highest_node_number = Linux::numa_max_node();
2918 int num_nodes = 0;
2920 // Get the total number of nodes in the system including nodes without memory.
2921 for (node = 0; node <= highest_node_number; node++) {
2922 if (isnode_in_existing_nodes(node)) {
2923 num_nodes++;
2924 }
2925 }
2926 return num_nodes;
2927 }
2929 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2930 size_t highest_node_number = Linux::numa_max_node();
2931 size_t i = 0;
2933 // Map all node ids in which is possible to allocate memory. Also nodes are
2934 // not always consecutively available, i.e. available from 0 to the highest
2935 // node number.
2936 for (size_t node = 0; node <= highest_node_number; node++) {
2937 if (Linux::isnode_in_configured_nodes(node)) {
2938 ids[i++] = node;
2939 }
2940 }
2941 return i;
2942 }
2944 bool os::get_page_info(char *start, page_info* info) {
2945 return false;
2946 }
2948 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2949 return end;
2950 }
2953 int os::Linux::sched_getcpu_syscall(void) {
2954 unsigned int cpu = 0;
2955 int retval = -1;
2957 #if defined(IA32)
2958 # ifndef SYS_getcpu
2959 # define SYS_getcpu 318
2960 # endif
2961 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2962 #elif defined(AMD64)
2963 // Unfortunately we have to bring all these macros here from vsyscall.h
2964 // to be able to compile on old linuxes.
2965 # define __NR_vgetcpu 2
2966 # define VSYSCALL_START (-10UL << 20)
2967 # define VSYSCALL_SIZE 1024
2968 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2969 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2970 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2971 retval = vgetcpu(&cpu, NULL, NULL);
2972 #endif
2974 return (retval == -1) ? retval : cpu;
2975 }
2977 // Something to do with the numa-aware allocator needs these symbols
2978 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2979 extern "C" JNIEXPORT void numa_error(char *where) { }
2980 extern "C" JNIEXPORT int fork1() { return fork(); }
2982 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2983 // load symbol from base version instead.
2984 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2985 void *f = dlvsym(handle, name, "libnuma_1.1");
2986 if (f == NULL) {
2987 f = dlsym(handle, name);
2988 }
2989 return f;
2990 }
2992 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2993 // Return NULL if the symbol is not defined in this particular version.
2994 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2995 return dlvsym(handle, name, "libnuma_1.2");
2996 }
2998 bool os::Linux::libnuma_init() {
2999 // sched_getcpu() should be in libc.
3000 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
3001 dlsym(RTLD_DEFAULT, "sched_getcpu")));
3003 // If it's not, try a direct syscall.
3004 if (sched_getcpu() == -1)
3005 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
3007 if (sched_getcpu() != -1) { // Does it work?
3008 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
3009 if (handle != NULL) {
3010 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
3011 libnuma_dlsym(handle, "numa_node_to_cpus")));
3012 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
3013 libnuma_dlsym(handle, "numa_max_node")));
3014 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
3015 libnuma_dlsym(handle, "numa_num_configured_nodes")));
3016 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
3017 libnuma_dlsym(handle, "numa_available")));
3018 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
3019 libnuma_dlsym(handle, "numa_tonode_memory")));
3020 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
3021 libnuma_dlsym(handle, "numa_interleave_memory")));
3022 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
3023 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
3024 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
3025 libnuma_dlsym(handle, "numa_set_bind_policy")));
3026 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
3027 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
3028 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
3029 libnuma_dlsym(handle, "numa_distance")));
3031 if (numa_available() != -1) {
3032 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
3033 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
3034 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
3035 // Create an index -> node mapping, since nodes are not always consecutive
3036 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3037 rebuild_nindex_to_node_map();
3038 // Create a cpu -> node mapping
3039 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3040 rebuild_cpu_to_node_map();
3041 return true;
3042 }
3043 }
3044 }
3045 return false;
3046 }
3048 void os::Linux::rebuild_nindex_to_node_map() {
3049 int highest_node_number = Linux::numa_max_node();
3051 nindex_to_node()->clear();
3052 for (int node = 0; node <= highest_node_number; node++) {
3053 if (Linux::isnode_in_existing_nodes(node)) {
3054 nindex_to_node()->append(node);
3055 }
3056 }
3057 }
3059 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3060 // The table is later used in get_node_by_cpu().
3061 void os::Linux::rebuild_cpu_to_node_map() {
3062 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3063 // in libnuma (possible values are starting from 16,
3064 // and continuing up with every other power of 2, but less
3065 // than the maximum number of CPUs supported by kernel), and
3066 // is a subject to change (in libnuma version 2 the requirements
3067 // are more reasonable) we'll just hardcode the number they use
3068 // in the library.
3069 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3071 size_t cpu_num = processor_count();
3072 size_t cpu_map_size = NCPUS / BitsPerCLong;
3073 size_t cpu_map_valid_size =
3074 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3076 cpu_to_node()->clear();
3077 cpu_to_node()->at_grow(cpu_num - 1);
3079 size_t node_num = get_existing_num_nodes();
3081 int distance = 0;
3082 int closest_distance = INT_MAX;
3083 int closest_node = 0;
3084 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3085 for (size_t i = 0; i < node_num; i++) {
3086 // Check if node is configured (not a memory-less node). If it is not, find
3087 // the closest configured node.
3088 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3089 closest_distance = INT_MAX;
3090 // Check distance from all remaining nodes in the system. Ignore distance
3091 // from itself and from another non-configured node.
3092 for (size_t m = 0; m < node_num; m++) {
3093 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3094 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3095 // If a closest node is found, update. There is always at least one
3096 // configured node in the system so there is always at least one node
3097 // close.
3098 if (distance != 0 && distance < closest_distance) {
3099 closest_distance = distance;
3100 closest_node = nindex_to_node()->at(m);
3101 }
3102 }
3103 }
3104 } else {
3105 // Current node is already a configured node.
3106 closest_node = nindex_to_node()->at(i);
3107 }
3109 // Get cpus from the original node and map them to the closest node. If node
3110 // is a configured node (not a memory-less node), then original node and
3111 // closest node are the same.
3112 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3113 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3114 if (cpu_map[j] != 0) {
3115 for (size_t k = 0; k < BitsPerCLong; k++) {
3116 if (cpu_map[j] & (1UL << k)) {
3117 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3118 }
3119 }
3120 }
3121 }
3122 }
3123 }
3124 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
3125 }
3127 int os::Linux::get_node_by_cpu(int cpu_id) {
3128 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3129 return cpu_to_node()->at(cpu_id);
3130 }
3131 return -1;
3132 }
3134 GrowableArray<int>* os::Linux::_cpu_to_node;
3135 GrowableArray<int>* os::Linux::_nindex_to_node;
3136 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3137 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3138 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3139 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3140 os::Linux::numa_available_func_t os::Linux::_numa_available;
3141 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3142 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3143 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3144 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3145 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3146 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3147 unsigned long* os::Linux::_numa_all_nodes;
3148 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3149 struct bitmask* os::Linux::_numa_nodes_ptr;
3151 bool os::pd_uncommit_memory(char* addr, size_t size) {
3152 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3153 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3154 return res != (uintptr_t) MAP_FAILED;
3155 }
3157 static
3158 address get_stack_commited_bottom(address bottom, size_t size) {
3159 address nbot = bottom;
3160 address ntop = bottom + size;
3162 size_t page_sz = os::vm_page_size();
3163 unsigned pages = size / page_sz;
3165 unsigned char vec[1];
3166 unsigned imin = 1, imax = pages + 1, imid;
3167 int mincore_return_value = 0;
3169 assert(imin <= imax, "Unexpected page size");
3171 while (imin < imax) {
3172 imid = (imax + imin) / 2;
3173 nbot = ntop - (imid * page_sz);
3175 // Use a trick with mincore to check whether the page is mapped or not.
3176 // mincore sets vec to 1 if page resides in memory and to 0 if page
3177 // is swapped output but if page we are asking for is unmapped
3178 // it returns -1,ENOMEM
3179 mincore_return_value = mincore(nbot, page_sz, vec);
3181 if (mincore_return_value == -1) {
3182 // Page is not mapped go up
3183 // to find first mapped page
3184 if (errno != EAGAIN) {
3185 assert(errno == ENOMEM, "Unexpected mincore errno");
3186 imax = imid;
3187 }
3188 } else {
3189 // Page is mapped go down
3190 // to find first not mapped page
3191 imin = imid + 1;
3192 }
3193 }
3195 nbot = nbot + page_sz;
3197 // Adjust stack bottom one page up if last checked page is not mapped
3198 if (mincore_return_value == -1) {
3199 nbot = nbot + page_sz;
3200 }
3202 return nbot;
3203 }
3206 // Linux uses a growable mapping for the stack, and if the mapping for
3207 // the stack guard pages is not removed when we detach a thread the
3208 // stack cannot grow beyond the pages where the stack guard was
3209 // mapped. If at some point later in the process the stack expands to
3210 // that point, the Linux kernel cannot expand the stack any further
3211 // because the guard pages are in the way, and a segfault occurs.
3212 //
3213 // However, it's essential not to split the stack region by unmapping
3214 // a region (leaving a hole) that's already part of the stack mapping,
3215 // so if the stack mapping has already grown beyond the guard pages at
3216 // the time we create them, we have to truncate the stack mapping.
3217 // So, we need to know the extent of the stack mapping when
3218 // create_stack_guard_pages() is called.
3220 // We only need this for stacks that are growable: at the time of
3221 // writing thread stacks don't use growable mappings (i.e. those
3222 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3223 // only applies to the main thread.
3225 // If the (growable) stack mapping already extends beyond the point
3226 // where we're going to put our guard pages, truncate the mapping at
3227 // that point by munmap()ping it. This ensures that when we later
3228 // munmap() the guard pages we don't leave a hole in the stack
3229 // mapping. This only affects the main/primordial thread
3231 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3233 if (os::is_primordial_thread()) {
3234 // As we manually grow stack up to bottom inside create_attached_thread(),
3235 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3236 // we don't need to do anything special.
3237 // Check it first, before calling heavy function.
3238 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3239 unsigned char vec[1];
3241 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3242 // Fallback to slow path on all errors, including EAGAIN
3243 stack_extent = (uintptr_t) get_stack_commited_bottom(
3244 os::Linux::initial_thread_stack_bottom(),
3245 (size_t)addr - stack_extent);
3246 }
3248 if (stack_extent < (uintptr_t)addr) {
3249 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3250 }
3251 }
3253 return os::commit_memory(addr, size, !ExecMem);
3254 }
3256 // If this is a growable mapping, remove the guard pages entirely by
3257 // munmap()ping them. If not, just call uncommit_memory(). This only
3258 // affects the main/primordial thread, but guard against future OS changes.
3259 // It's safe to always unmap guard pages for primordial thread because we
3260 // always place it right after end of the mapped region.
3262 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3263 uintptr_t stack_extent, stack_base;
3265 if (os::is_primordial_thread()) {
3266 return ::munmap(addr, size) == 0;
3267 }
3269 return os::uncommit_memory(addr, size);
3270 }
3272 static address _highest_vm_reserved_address = NULL;
3274 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3275 // at 'requested_addr'. If there are existing memory mappings at the same
3276 // location, however, they will be overwritten. If 'fixed' is false,
3277 // 'requested_addr' is only treated as a hint, the return value may or
3278 // may not start from the requested address. Unlike Linux mmap(), this
3279 // function returns NULL to indicate failure.
3280 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3281 char * addr;
3282 int flags;
3284 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3285 if (fixed) {
3286 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3287 flags |= MAP_FIXED;
3288 }
3290 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3291 // touch an uncommitted page. Otherwise, the read/write might
3292 // succeed if we have enough swap space to back the physical page.
3293 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3294 flags, -1, 0);
3296 if (addr != MAP_FAILED) {
3297 // anon_mmap() should only get called during VM initialization,
3298 // don't need lock (actually we can skip locking even it can be called
3299 // from multiple threads, because _highest_vm_reserved_address is just a
3300 // hint about the upper limit of non-stack memory regions.)
3301 if ((address)addr + bytes > _highest_vm_reserved_address) {
3302 _highest_vm_reserved_address = (address)addr + bytes;
3303 }
3304 }
3306 return addr == MAP_FAILED ? NULL : addr;
3307 }
3309 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3310 // (req_addr != NULL) or with a given alignment.
3311 // - bytes shall be a multiple of alignment.
3312 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3313 // - alignment sets the alignment at which memory shall be allocated.
3314 // It must be a multiple of allocation granularity.
3315 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3316 // req_addr or NULL.
3317 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3319 size_t extra_size = bytes;
3320 if (req_addr == NULL && alignment > 0) {
3321 extra_size += alignment;
3322 }
3324 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3325 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3326 -1, 0);
3327 if (start == MAP_FAILED) {
3328 start = NULL;
3329 } else {
3330 if (req_addr != NULL) {
3331 if (start != req_addr) {
3332 ::munmap(start, extra_size);
3333 start = NULL;
3334 }
3335 } else {
3336 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3337 char* const end_aligned = start_aligned + bytes;
3338 char* const end = start + extra_size;
3339 if (start_aligned > start) {
3340 ::munmap(start, start_aligned - start);
3341 }
3342 if (end_aligned < end) {
3343 ::munmap(end_aligned, end - end_aligned);
3344 }
3345 start = start_aligned;
3346 }
3347 }
3348 return start;
3349 }
3351 // Don't update _highest_vm_reserved_address, because there might be memory
3352 // regions above addr + size. If so, releasing a memory region only creates
3353 // a hole in the address space, it doesn't help prevent heap-stack collision.
3354 //
3355 static int anon_munmap(char * addr, size_t size) {
3356 return ::munmap(addr, size) == 0;
3357 }
3359 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3360 size_t alignment_hint) {
3361 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3362 }
3364 bool os::pd_release_memory(char* addr, size_t size) {
3365 return anon_munmap(addr, size);
3366 }
3368 static address highest_vm_reserved_address() {
3369 return _highest_vm_reserved_address;
3370 }
3372 static bool linux_mprotect(char* addr, size_t size, int prot) {
3373 // Linux wants the mprotect address argument to be page aligned.
3374 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3376 // According to SUSv3, mprotect() should only be used with mappings
3377 // established by mmap(), and mmap() always maps whole pages. Unaligned
3378 // 'addr' likely indicates problem in the VM (e.g. trying to change
3379 // protection of malloc'ed or statically allocated memory). Check the
3380 // caller if you hit this assert.
3381 assert(addr == bottom, "sanity check");
3383 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3384 return ::mprotect(bottom, size, prot) == 0;
3385 }
3387 // Set protections specified
3388 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3389 bool is_committed) {
3390 unsigned int p = 0;
3391 switch (prot) {
3392 case MEM_PROT_NONE: p = PROT_NONE; break;
3393 case MEM_PROT_READ: p = PROT_READ; break;
3394 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3395 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3396 default:
3397 ShouldNotReachHere();
3398 }
3399 // is_committed is unused.
3400 return linux_mprotect(addr, bytes, p);
3401 }
3403 bool os::guard_memory(char* addr, size_t size) {
3404 return linux_mprotect(addr, size, PROT_NONE);
3405 }
3407 bool os::unguard_memory(char* addr, size_t size) {
3408 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3409 }
3411 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3412 bool result = false;
3413 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3414 MAP_ANONYMOUS|MAP_PRIVATE,
3415 -1, 0);
3416 if (p != MAP_FAILED) {
3417 void *aligned_p = align_ptr_up(p, page_size);
3419 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3421 munmap(p, page_size * 2);
3422 }
3424 if (warn && !result) {
3425 warning("TransparentHugePages is not supported by the operating system.");
3426 }
3428 return result;
3429 }
3431 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3432 bool result = false;
3433 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3434 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3435 -1, 0);
3437 if (p != MAP_FAILED) {
3438 // We don't know if this really is a huge page or not.
3439 FILE *fp = fopen("/proc/self/maps", "r");
3440 if (fp) {
3441 while (!feof(fp)) {
3442 char chars[257];
3443 long x = 0;
3444 if (fgets(chars, sizeof(chars), fp)) {
3445 if (sscanf(chars, "%lx-%*x", &x) == 1
3446 && x == (long)p) {
3447 if (strstr (chars, "hugepage")) {
3448 result = true;
3449 break;
3450 }
3451 }
3452 }
3453 }
3454 fclose(fp);
3455 }
3456 munmap(p, page_size);
3457 }
3459 if (warn && !result) {
3460 warning("HugeTLBFS is not supported by the operating system.");
3461 }
3463 return result;
3464 }
3466 /*
3467 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3468 *
3469 * From the coredump_filter documentation:
3470 *
3471 * - (bit 0) anonymous private memory
3472 * - (bit 1) anonymous shared memory
3473 * - (bit 2) file-backed private memory
3474 * - (bit 3) file-backed shared memory
3475 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3476 * effective only if the bit 2 is cleared)
3477 * - (bit 5) hugetlb private memory
3478 * - (bit 6) hugetlb shared memory
3479 */
3480 static void set_coredump_filter(void) {
3481 FILE *f;
3482 long cdm;
3484 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3485 return;
3486 }
3488 if (fscanf(f, "%lx", &cdm) != 1) {
3489 fclose(f);
3490 return;
3491 }
3493 rewind(f);
3495 if ((cdm & LARGEPAGES_BIT) == 0) {
3496 cdm |= LARGEPAGES_BIT;
3497 fprintf(f, "%#lx", cdm);
3498 }
3500 fclose(f);
3501 }
3503 // Large page support
3505 static size_t _large_page_size = 0;
3507 size_t os::Linux::find_large_page_size() {
3508 size_t large_page_size = 0;
3510 // large_page_size on Linux is used to round up heap size. x86 uses either
3511 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3512 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3513 // page as large as 256M.
3514 //
3515 // Here we try to figure out page size by parsing /proc/meminfo and looking
3516 // for a line with the following format:
3517 // Hugepagesize: 2048 kB
3518 //
3519 // If we can't determine the value (e.g. /proc is not mounted, or the text
3520 // format has been changed), we'll use the largest page size supported by
3521 // the processor.
3523 #ifndef ZERO
3524 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3525 ARM_ONLY(2 * M) PPC_ONLY(4 * M) MIPS64_ONLY(4 * M); //In MIPS _large_page_size is seted 4*M.
3526 #endif // ZERO
3528 FILE *fp = fopen("/proc/meminfo", "r");
3529 if (fp) {
3530 while (!feof(fp)) {
3531 int x = 0;
3532 char buf[16];
3533 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3534 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3535 large_page_size = x * K;
3536 break;
3537 }
3538 } else {
3539 // skip to next line
3540 for (;;) {
3541 int ch = fgetc(fp);
3542 if (ch == EOF || ch == (int)'\n') break;
3543 }
3544 }
3545 }
3546 fclose(fp);
3547 }
3549 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3550 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3551 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3552 proper_unit_for_byte_size(large_page_size));
3553 }
3555 return large_page_size;
3556 }
3558 size_t os::Linux::setup_large_page_size() {
3559 _large_page_size = Linux::find_large_page_size();
3560 const size_t default_page_size = (size_t)Linux::page_size();
3561 if (_large_page_size > default_page_size) {
3562 _page_sizes[0] = _large_page_size;
3563 _page_sizes[1] = default_page_size;
3564 _page_sizes[2] = 0;
3565 }
3567 return _large_page_size;
3568 }
3570 bool os::Linux::setup_large_page_type(size_t page_size) {
3571 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3572 FLAG_IS_DEFAULT(UseSHM) &&
3573 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3575 // The type of large pages has not been specified by the user.
3577 // Try UseHugeTLBFS and then UseSHM.
3578 UseHugeTLBFS = UseSHM = true;
3580 // Don't try UseTransparentHugePages since there are known
3581 // performance issues with it turned on. This might change in the future.
3582 UseTransparentHugePages = false;
3583 }
3585 if (UseTransparentHugePages) {
3586 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3587 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3588 UseHugeTLBFS = false;
3589 UseSHM = false;
3590 return true;
3591 }
3592 UseTransparentHugePages = false;
3593 }
3595 if (UseHugeTLBFS) {
3596 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3597 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3598 UseSHM = false;
3599 return true;
3600 }
3601 UseHugeTLBFS = false;
3602 }
3604 return UseSHM;
3605 }
3607 void os::large_page_init() {
3608 if (!UseLargePages &&
3609 !UseTransparentHugePages &&
3610 !UseHugeTLBFS &&
3611 !UseSHM) {
3612 // Not using large pages.
3613 return;
3614 }
3616 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3617 // The user explicitly turned off large pages.
3618 // Ignore the rest of the large pages flags.
3619 UseTransparentHugePages = false;
3620 UseHugeTLBFS = false;
3621 UseSHM = false;
3622 return;
3623 }
3625 size_t large_page_size = Linux::setup_large_page_size();
3626 UseLargePages = Linux::setup_large_page_type(large_page_size);
3628 set_coredump_filter();
3629 }
3631 #ifndef SHM_HUGETLB
3632 #define SHM_HUGETLB 04000
3633 #endif
3635 #define shm_warning_format(format, ...) \
3636 do { \
3637 if (UseLargePages && \
3638 (!FLAG_IS_DEFAULT(UseLargePages) || \
3639 !FLAG_IS_DEFAULT(UseSHM) || \
3640 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3641 warning(format, __VA_ARGS__); \
3642 } \
3643 } while (0)
3645 #define shm_warning(str) shm_warning_format("%s", str)
3647 #define shm_warning_with_errno(str) \
3648 do { \
3649 int err = errno; \
3650 shm_warning_format(str " (error = %d)", err); \
3651 } while (0)
3653 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3654 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3656 if (!is_size_aligned(alignment, SHMLBA)) {
3657 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3658 return NULL;
3659 }
3661 // To ensure that we get 'alignment' aligned memory from shmat,
3662 // we pre-reserve aligned virtual memory and then attach to that.
3664 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3665 if (pre_reserved_addr == NULL) {
3666 // Couldn't pre-reserve aligned memory.
3667 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3668 return NULL;
3669 }
3671 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3672 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3674 if ((intptr_t)addr == -1) {
3675 int err = errno;
3676 shm_warning_with_errno("Failed to attach shared memory.");
3678 assert(err != EACCES, "Unexpected error");
3679 assert(err != EIDRM, "Unexpected error");
3680 assert(err != EINVAL, "Unexpected error");
3682 // Since we don't know if the kernel unmapped the pre-reserved memory area
3683 // we can't unmap it, since that would potentially unmap memory that was
3684 // mapped from other threads.
3685 return NULL;
3686 }
3688 return addr;
3689 }
3691 static char* shmat_at_address(int shmid, char* req_addr) {
3692 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3693 assert(false, "Requested address needs to be SHMLBA aligned");
3694 return NULL;
3695 }
3697 char* addr = (char*)shmat(shmid, req_addr, 0);
3699 if ((intptr_t)addr == -1) {
3700 shm_warning_with_errno("Failed to attach shared memory.");
3701 return NULL;
3702 }
3704 return addr;
3705 }
3707 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3708 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3709 if (req_addr != NULL) {
3710 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3711 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3712 return shmat_at_address(shmid, req_addr);
3713 }
3715 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3716 // return large page size aligned memory addresses when req_addr == NULL.
3717 // However, if the alignment is larger than the large page size, we have
3718 // to manually ensure that the memory returned is 'alignment' aligned.
3719 if (alignment > os::large_page_size()) {
3720 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3721 return shmat_with_alignment(shmid, bytes, alignment);
3722 } else {
3723 return shmat_at_address(shmid, NULL);
3724 }
3725 }
3727 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3728 // "exec" is passed in but not used. Creating the shared image for
3729 // the code cache doesn't have an SHM_X executable permission to check.
3730 assert(UseLargePages && UseSHM, "only for SHM large pages");
3731 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3732 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3734 if (!is_size_aligned(bytes, os::large_page_size())) {
3735 return NULL; // Fallback to small pages.
3736 }
3738 // Create a large shared memory region to attach to based on size.
3739 // Currently, size is the total size of the heap.
3740 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3741 if (shmid == -1) {
3742 // Possible reasons for shmget failure:
3743 // 1. shmmax is too small for Java heap.
3744 // > check shmmax value: cat /proc/sys/kernel/shmmax
3745 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3746 // 2. not enough large page memory.
3747 // > check available large pages: cat /proc/meminfo
3748 // > increase amount of large pages:
3749 // echo new_value > /proc/sys/vm/nr_hugepages
3750 // Note 1: different Linux may use different name for this property,
3751 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3752 // Note 2: it's possible there's enough physical memory available but
3753 // they are so fragmented after a long run that they can't
3754 // coalesce into large pages. Try to reserve large pages when
3755 // the system is still "fresh".
3756 shm_warning_with_errno("Failed to reserve shared memory.");
3757 return NULL;
3758 }
3760 // Attach to the region.
3761 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3763 // Remove shmid. If shmat() is successful, the actual shared memory segment
3764 // will be deleted when it's detached by shmdt() or when the process
3765 // terminates. If shmat() is not successful this will remove the shared
3766 // segment immediately.
3767 shmctl(shmid, IPC_RMID, NULL);
3769 return addr;
3770 }
3772 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3773 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3775 bool warn_on_failure = UseLargePages &&
3776 (!FLAG_IS_DEFAULT(UseLargePages) ||
3777 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3778 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3780 if (warn_on_failure) {
3781 char msg[128];
3782 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3783 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3784 warning("%s", msg);
3785 }
3786 }
3788 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3789 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3790 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3791 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3793 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3794 char* addr = (char*)::mmap(req_addr, bytes, prot,
3795 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3796 -1, 0);
3798 if (addr == MAP_FAILED) {
3799 warn_on_large_pages_failure(req_addr, bytes, errno);
3800 return NULL;
3801 }
3803 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3805 return addr;
3806 }
3808 // Reserve memory using mmap(MAP_HUGETLB).
3809 // - bytes shall be a multiple of alignment.
3810 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3811 // - alignment sets the alignment at which memory shall be allocated.
3812 // It must be a multiple of allocation granularity.
3813 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3814 // req_addr or NULL.
3815 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3816 size_t large_page_size = os::large_page_size();
3817 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3819 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3820 assert(is_size_aligned(bytes, alignment), "Must be");
3822 // First reserve - but not commit - the address range in small pages.
3823 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3825 if (start == NULL) {
3826 return NULL;
3827 }
3829 assert(is_ptr_aligned(start, alignment), "Must be");
3831 char* end = start + bytes;
3833 // Find the regions of the allocated chunk that can be promoted to large pages.
3834 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3835 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3837 size_t lp_bytes = lp_end - lp_start;
3839 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3841 if (lp_bytes == 0) {
3842 // The mapped region doesn't even span the start and the end of a large page.
3843 // Fall back to allocate a non-special area.
3844 ::munmap(start, end - start);
3845 return NULL;
3846 }
3848 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3850 void* result;
3852 // Commit small-paged leading area.
3853 if (start != lp_start) {
3854 result = ::mmap(start, lp_start - start, prot,
3855 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3856 -1, 0);
3857 if (result == MAP_FAILED) {
3858 ::munmap(lp_start, end - lp_start);
3859 return NULL;
3860 }
3861 }
3863 // Commit large-paged area.
3864 result = ::mmap(lp_start, lp_bytes, prot,
3865 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3866 -1, 0);
3867 if (result == MAP_FAILED) {
3868 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3869 // If the mmap above fails, the large pages region will be unmapped and we
3870 // have regions before and after with small pages. Release these regions.
3871 //
3872 // | mapped | unmapped | mapped |
3873 // ^ ^ ^ ^
3874 // start lp_start lp_end end
3875 //
3876 ::munmap(start, lp_start - start);
3877 ::munmap(lp_end, end - lp_end);
3878 return NULL;
3879 }
3881 // Commit small-paged trailing area.
3882 if (lp_end != end) {
3883 result = ::mmap(lp_end, end - lp_end, prot,
3884 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3885 -1, 0);
3886 if (result == MAP_FAILED) {
3887 ::munmap(start, lp_end - start);
3888 return NULL;
3889 }
3890 }
3892 return start;
3893 }
3895 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3896 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3897 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3898 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3899 assert(is_power_of_2(os::large_page_size()), "Must be");
3900 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3902 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3903 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3904 } else {
3905 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3906 }
3907 }
3909 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3910 assert(UseLargePages, "only for large pages");
3912 char* addr;
3913 if (UseSHM) {
3914 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3915 } else {
3916 assert(UseHugeTLBFS, "must be");
3917 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3918 }
3920 if (addr != NULL) {
3921 if (UseNUMAInterleaving) {
3922 numa_make_global(addr, bytes);
3923 }
3925 // The memory is committed
3926 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3927 }
3929 return addr;
3930 }
3932 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3933 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3934 return shmdt(base) == 0;
3935 }
3937 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3938 return pd_release_memory(base, bytes);
3939 }
3941 bool os::release_memory_special(char* base, size_t bytes) {
3942 bool res;
3943 if (MemTracker::tracking_level() > NMT_minimal) {
3944 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3945 res = os::Linux::release_memory_special_impl(base, bytes);
3946 if (res) {
3947 tkr.record((address)base, bytes);
3948 }
3950 } else {
3951 res = os::Linux::release_memory_special_impl(base, bytes);
3952 }
3953 return res;
3954 }
3956 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3957 assert(UseLargePages, "only for large pages");
3958 bool res;
3960 if (UseSHM) {
3961 res = os::Linux::release_memory_special_shm(base, bytes);
3962 } else {
3963 assert(UseHugeTLBFS, "must be");
3964 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3965 }
3966 return res;
3967 }
3969 size_t os::large_page_size() {
3970 return _large_page_size;
3971 }
3973 // With SysV SHM the entire memory region must be allocated as shared
3974 // memory.
3975 // HugeTLBFS allows application to commit large page memory on demand.
3976 // However, when committing memory with HugeTLBFS fails, the region
3977 // that was supposed to be committed will lose the old reservation
3978 // and allow other threads to steal that memory region. Because of this
3979 // behavior we can't commit HugeTLBFS memory.
3980 bool os::can_commit_large_page_memory() {
3981 return UseTransparentHugePages;
3982 }
3984 bool os::can_execute_large_page_memory() {
3985 return UseTransparentHugePages || UseHugeTLBFS;
3986 }
3988 // Reserve memory at an arbitrary address, only if that area is
3989 // available (and not reserved for something else).
3991 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3992 const int max_tries = 10;
3993 char* base[max_tries];
3994 size_t size[max_tries];
3995 const size_t gap = 0x000000;
3997 // Assert only that the size is a multiple of the page size, since
3998 // that's all that mmap requires, and since that's all we really know
3999 // about at this low abstraction level. If we need higher alignment,
4000 // we can either pass an alignment to this method or verify alignment
4001 // in one of the methods further up the call chain. See bug 5044738.
4002 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
4004 // Repeatedly allocate blocks until the block is allocated at the
4005 // right spot. Give up after max_tries. Note that reserve_memory() will
4006 // automatically update _highest_vm_reserved_address if the call is
4007 // successful. The variable tracks the highest memory address every reserved
4008 // by JVM. It is used to detect heap-stack collision if running with
4009 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
4010 // space than needed, it could confuse the collision detecting code. To
4011 // solve the problem, save current _highest_vm_reserved_address and
4012 // calculate the correct value before return.
4013 address old_highest = _highest_vm_reserved_address;
4015 // Linux mmap allows caller to pass an address as hint; give it a try first,
4016 // if kernel honors the hint then we can return immediately.
4017 char * addr = anon_mmap(requested_addr, bytes, false);
4018 if (addr == requested_addr) {
4019 return requested_addr;
4020 }
4022 if (addr != NULL) {
4023 // mmap() is successful but it fails to reserve at the requested address
4024 anon_munmap(addr, bytes);
4025 }
4027 int i;
4028 for (i = 0; i < max_tries; ++i) {
4029 base[i] = reserve_memory(bytes);
4031 if (base[i] != NULL) {
4032 // Is this the block we wanted?
4033 if (base[i] == requested_addr) {
4034 size[i] = bytes;
4035 break;
4036 }
4038 // Does this overlap the block we wanted? Give back the overlapped
4039 // parts and try again.
4041 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
4042 if (top_overlap >= 0 && top_overlap < bytes) {
4043 unmap_memory(base[i], top_overlap);
4044 base[i] += top_overlap;
4045 size[i] = bytes - top_overlap;
4046 } else {
4047 size_t bottom_overlap = base[i] + bytes - requested_addr;
4048 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
4049 unmap_memory(requested_addr, bottom_overlap);
4050 size[i] = bytes - bottom_overlap;
4051 } else {
4052 size[i] = bytes;
4053 }
4054 }
4055 }
4056 }
4058 // Give back the unused reserved pieces.
4060 for (int j = 0; j < i; ++j) {
4061 if (base[j] != NULL) {
4062 unmap_memory(base[j], size[j]);
4063 }
4064 }
4066 if (i < max_tries) {
4067 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
4068 return requested_addr;
4069 } else {
4070 _highest_vm_reserved_address = old_highest;
4071 return NULL;
4072 }
4073 }
4075 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4076 return ::read(fd, buf, nBytes);
4077 }
4079 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
4080 return ::pread(fd, buf, nBytes, offset);
4081 }
4083 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
4084 // Solaris uses poll(), linux uses park().
4085 // Poll() is likely a better choice, assuming that Thread.interrupt()
4086 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
4087 // SIGSEGV, see 4355769.
4089 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
4090 assert(thread == Thread::current(), "thread consistency check");
4092 ParkEvent * const slp = thread->_SleepEvent ;
4093 slp->reset() ;
4094 OrderAccess::fence() ;
4096 if (interruptible) {
4097 jlong prevtime = javaTimeNanos();
4099 for (;;) {
4100 if (os::is_interrupted(thread, true)) {
4101 return OS_INTRPT;
4102 }
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) {
4115 return OS_OK;
4116 }
4118 prevtime = newtime;
4120 {
4121 assert(thread->is_Java_thread(), "sanity check");
4122 JavaThread *jt = (JavaThread *) thread;
4123 ThreadBlockInVM tbivm(jt);
4124 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
4126 jt->set_suspend_equivalent();
4127 // cleared by handle_special_suspend_equivalent_condition() or
4128 // java_suspend_self() via check_and_wait_while_suspended()
4130 slp->park(millis);
4132 // were we externally suspended while we were waiting?
4133 jt->check_and_wait_while_suspended();
4134 }
4135 }
4136 } else {
4137 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4138 jlong prevtime = javaTimeNanos();
4140 for (;;) {
4141 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
4142 // the 1st iteration ...
4143 jlong newtime = javaTimeNanos();
4145 if (newtime - prevtime < 0) {
4146 // time moving backwards, should only happen if no monotonic clock
4147 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4148 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4149 } else {
4150 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4151 }
4153 if(millis <= 0) break ;
4155 prevtime = newtime;
4156 slp->park(millis);
4157 }
4158 return OS_OK ;
4159 }
4160 }
4162 //
4163 // Short sleep, direct OS call.
4164 //
4165 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4166 // sched_yield(2) will actually give up the CPU:
4167 //
4168 // * Alone on this pariticular CPU, keeps running.
4169 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4170 // (pre 2.6.39).
4171 //
4172 // So calling this with 0 is an alternative.
4173 //
4174 void os::naked_short_sleep(jlong ms) {
4175 struct timespec req;
4177 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4178 req.tv_sec = 0;
4179 if (ms > 0) {
4180 req.tv_nsec = (ms % 1000) * 1000000;
4181 }
4182 else {
4183 req.tv_nsec = 1;
4184 }
4186 nanosleep(&req, NULL);
4188 return;
4189 }
4191 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4192 void os::infinite_sleep() {
4193 while (true) { // sleep forever ...
4194 ::sleep(100); // ... 100 seconds at a time
4195 }
4196 }
4198 // Used to convert frequent JVM_Yield() to nops
4199 bool os::dont_yield() {
4200 return DontYieldALot;
4201 }
4203 void os::yield() {
4204 sched_yield();
4205 }
4207 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
4209 void os::yield_all(int attempts) {
4210 // Yields to all threads, including threads with lower priorities
4211 // Threads on Linux are all with same priority. The Solaris style
4212 // os::yield_all() with nanosleep(1ms) is not necessary.
4213 sched_yield();
4214 }
4216 // Called from the tight loops to possibly influence time-sharing heuristics
4217 void os::loop_breaker(int attempts) {
4218 os::yield_all(attempts);
4219 }
4221 ////////////////////////////////////////////////////////////////////////////////
4222 // thread priority support
4224 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4225 // only supports dynamic priority, static priority must be zero. For real-time
4226 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4227 // However, for large multi-threaded applications, SCHED_RR is not only slower
4228 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4229 // of 5 runs - Sep 2005).
4230 //
4231 // The following code actually changes the niceness of kernel-thread/LWP. It
4232 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4233 // not the entire user process, and user level threads are 1:1 mapped to kernel
4234 // threads. It has always been the case, but could change in the future. For
4235 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4236 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4238 int os::java_to_os_priority[CriticalPriority + 1] = {
4239 19, // 0 Entry should never be used
4241 4, // 1 MinPriority
4242 3, // 2
4243 2, // 3
4245 1, // 4
4246 0, // 5 NormPriority
4247 -1, // 6
4249 -2, // 7
4250 -3, // 8
4251 -4, // 9 NearMaxPriority
4253 -5, // 10 MaxPriority
4255 -5 // 11 CriticalPriority
4256 };
4258 static int prio_init() {
4259 if (ThreadPriorityPolicy == 1) {
4260 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4261 // if effective uid is not root. Perhaps, a more elegant way of doing
4262 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4263 if (geteuid() != 0) {
4264 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4265 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4266 }
4267 ThreadPriorityPolicy = 0;
4268 }
4269 }
4270 if (UseCriticalJavaThreadPriority) {
4271 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4272 }
4273 return 0;
4274 }
4276 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4277 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4279 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4280 return (ret == 0) ? OS_OK : OS_ERR;
4281 }
4283 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4284 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4285 *priority_ptr = java_to_os_priority[NormPriority];
4286 return OS_OK;
4287 }
4289 errno = 0;
4290 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4291 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4292 }
4294 // Hint to the underlying OS that a task switch would not be good.
4295 // Void return because it's a hint and can fail.
4296 void os::hint_no_preempt() {}
4298 ////////////////////////////////////////////////////////////////////////////////
4299 // suspend/resume support
4301 // the low-level signal-based suspend/resume support is a remnant from the
4302 // old VM-suspension that used to be for java-suspension, safepoints etc,
4303 // within hotspot. Now there is a single use-case for this:
4304 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4305 // that runs in the watcher thread.
4306 // The remaining code is greatly simplified from the more general suspension
4307 // code that used to be used.
4308 //
4309 // The protocol is quite simple:
4310 // - suspend:
4311 // - sends a signal to the target thread
4312 // - polls the suspend state of the osthread using a yield loop
4313 // - target thread signal handler (SR_handler) sets suspend state
4314 // and blocks in sigsuspend until continued
4315 // - resume:
4316 // - sets target osthread state to continue
4317 // - sends signal to end the sigsuspend loop in the SR_handler
4318 //
4319 // Note that the SR_lock plays no role in this suspend/resume protocol.
4320 //
4322 static void resume_clear_context(OSThread *osthread) {
4323 osthread->set_ucontext(NULL);
4324 osthread->set_siginfo(NULL);
4325 }
4327 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4328 osthread->set_ucontext(context);
4329 osthread->set_siginfo(siginfo);
4330 }
4332 //
4333 // Handler function invoked when a thread's execution is suspended or
4334 // resumed. We have to be careful that only async-safe functions are
4335 // called here (Note: most pthread functions are not async safe and
4336 // should be avoided.)
4337 //
4338 // Note: sigwait() is a more natural fit than sigsuspend() from an
4339 // interface point of view, but sigwait() prevents the signal hander
4340 // from being run. libpthread would get very confused by not having
4341 // its signal handlers run and prevents sigwait()'s use with the
4342 // mutex granting granting signal.
4343 //
4344 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4345 //
4346 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4347 // Save and restore errno to avoid confusing native code with EINTR
4348 // after sigsuspend.
4349 int old_errno = errno;
4351 Thread* thread = Thread::current();
4352 OSThread* osthread = thread->osthread();
4353 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4355 os::SuspendResume::State current = osthread->sr.state();
4356 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4357 suspend_save_context(osthread, siginfo, context);
4359 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4360 os::SuspendResume::State state = osthread->sr.suspended();
4361 if (state == os::SuspendResume::SR_SUSPENDED) {
4362 sigset_t suspend_set; // signals for sigsuspend()
4364 // get current set of blocked signals and unblock resume signal
4365 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4366 sigdelset(&suspend_set, SR_signum);
4368 sr_semaphore.signal();
4369 // wait here until we are resumed
4370 while (1) {
4371 sigsuspend(&suspend_set);
4373 os::SuspendResume::State result = osthread->sr.running();
4374 if (result == os::SuspendResume::SR_RUNNING) {
4375 sr_semaphore.signal();
4376 break;
4377 }
4378 }
4380 } else if (state == os::SuspendResume::SR_RUNNING) {
4381 // request was cancelled, continue
4382 } else {
4383 ShouldNotReachHere();
4384 }
4386 resume_clear_context(osthread);
4387 } else if (current == os::SuspendResume::SR_RUNNING) {
4388 // request was cancelled, continue
4389 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4390 // ignore
4391 } else {
4392 // ignore
4393 }
4395 errno = old_errno;
4396 }
4399 static int SR_initialize() {
4400 struct sigaction act;
4401 char *s;
4402 /* Get signal number to use for suspend/resume */
4403 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4404 int sig = ::strtol(s, 0, 10);
4405 if (sig > 0 || sig < _NSIG) {
4406 SR_signum = sig;
4407 }
4408 }
4410 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4411 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4413 sigemptyset(&SR_sigset);
4414 sigaddset(&SR_sigset, SR_signum);
4416 /* Set up signal handler for suspend/resume */
4417 act.sa_flags = SA_RESTART|SA_SIGINFO;
4418 act.sa_handler = (void (*)(int)) SR_handler;
4420 // SR_signum is blocked by default.
4421 // 4528190 - We also need to block pthread restart signal (32 on all
4422 // supported Linux platforms). Note that LinuxThreads need to block
4423 // this signal for all threads to work properly. So we don't have
4424 // to use hard-coded signal number when setting up the mask.
4425 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4427 if (sigaction(SR_signum, &act, 0) == -1) {
4428 return -1;
4429 }
4431 // Save signal flag
4432 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4433 return 0;
4434 }
4436 static int sr_notify(OSThread* osthread) {
4437 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4438 assert_status(status == 0, status, "pthread_kill");
4439 return status;
4440 }
4442 // "Randomly" selected value for how long we want to spin
4443 // before bailing out on suspending a thread, also how often
4444 // we send a signal to a thread we want to resume
4445 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4446 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4448 // returns true on success and false on error - really an error is fatal
4449 // but this seems the normal response to library errors
4450 static bool do_suspend(OSThread* osthread) {
4451 assert(osthread->sr.is_running(), "thread should be running");
4452 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4454 // mark as suspended and send signal
4455 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4456 // failed to switch, state wasn't running?
4457 ShouldNotReachHere();
4458 return false;
4459 }
4461 if (sr_notify(osthread) != 0) {
4462 ShouldNotReachHere();
4463 }
4465 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4466 while (true) {
4467 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4468 break;
4469 } else {
4470 // timeout
4471 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4472 if (cancelled == os::SuspendResume::SR_RUNNING) {
4473 return false;
4474 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4475 // make sure that we consume the signal on the semaphore as well
4476 sr_semaphore.wait();
4477 break;
4478 } else {
4479 ShouldNotReachHere();
4480 return false;
4481 }
4482 }
4483 }
4485 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4486 return true;
4487 }
4489 static void do_resume(OSThread* osthread) {
4490 assert(osthread->sr.is_suspended(), "thread should be suspended");
4491 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4493 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4494 // failed to switch to WAKEUP_REQUEST
4495 ShouldNotReachHere();
4496 return;
4497 }
4499 while (true) {
4500 if (sr_notify(osthread) == 0) {
4501 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4502 if (osthread->sr.is_running()) {
4503 return;
4504 }
4505 }
4506 } else {
4507 ShouldNotReachHere();
4508 }
4509 }
4511 guarantee(osthread->sr.is_running(), "Must be running!");
4512 }
4514 ////////////////////////////////////////////////////////////////////////////////
4515 // interrupt support
4517 void os::interrupt(Thread* thread) {
4518 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4519 "possibility of dangling Thread pointer");
4521 OSThread* osthread = thread->osthread();
4523 if (!osthread->interrupted()) {
4524 osthread->set_interrupted(true);
4525 // More than one thread can get here with the same value of osthread,
4526 // resulting in multiple notifications. We do, however, want the store
4527 // to interrupted() to be visible to other threads before we execute unpark().
4528 OrderAccess::fence();
4529 ParkEvent * const slp = thread->_SleepEvent ;
4530 if (slp != NULL) slp->unpark() ;
4531 }
4533 // For JSR166. Unpark even if interrupt status already was set
4534 if (thread->is_Java_thread())
4535 ((JavaThread*)thread)->parker()->unpark();
4537 ParkEvent * ev = thread->_ParkEvent ;
4538 if (ev != NULL) ev->unpark() ;
4540 }
4542 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4543 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4544 "possibility of dangling Thread pointer");
4546 OSThread* osthread = thread->osthread();
4548 bool interrupted = osthread->interrupted();
4550 if (interrupted && clear_interrupted) {
4551 osthread->set_interrupted(false);
4552 // consider thread->_SleepEvent->reset() ... optional optimization
4553 }
4555 return interrupted;
4556 }
4558 ///////////////////////////////////////////////////////////////////////////////////
4559 // signal handling (except suspend/resume)
4561 // This routine may be used by user applications as a "hook" to catch signals.
4562 // The user-defined signal handler must pass unrecognized signals to this
4563 // routine, and if it returns true (non-zero), then the signal handler must
4564 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4565 // routine will never retun false (zero), but instead will execute a VM panic
4566 // routine kill the process.
4567 //
4568 // If this routine returns false, it is OK to call it again. This allows
4569 // the user-defined signal handler to perform checks either before or after
4570 // the VM performs its own checks. Naturally, the user code would be making
4571 // a serious error if it tried to handle an exception (such as a null check
4572 // or breakpoint) that the VM was generating for its own correct operation.
4573 //
4574 // This routine may recognize any of the following kinds of signals:
4575 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4576 // It should be consulted by handlers for any of those signals.
4577 //
4578 // The caller of this routine must pass in the three arguments supplied
4579 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4580 // field of the structure passed to sigaction(). This routine assumes that
4581 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4582 //
4583 // Note that the VM will print warnings if it detects conflicting signal
4584 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4585 //
4586 extern "C" JNIEXPORT int
4587 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4588 void* ucontext, int abort_if_unrecognized);
4590 void signalHandler(int sig, siginfo_t* info, void* uc) {
4591 assert(info != NULL && uc != NULL, "it must be old kernel");
4592 int orig_errno = errno; // Preserve errno value over signal handler.
4593 JVM_handle_linux_signal(sig, info, uc, true);
4594 errno = orig_errno;
4595 }
4598 // This boolean allows users to forward their own non-matching signals
4599 // to JVM_handle_linux_signal, harmlessly.
4600 bool os::Linux::signal_handlers_are_installed = false;
4602 // For signal-chaining
4603 struct sigaction os::Linux::sigact[MAXSIGNUM];
4604 unsigned int os::Linux::sigs = 0;
4605 bool os::Linux::libjsig_is_loaded = false;
4606 typedef struct sigaction *(*get_signal_t)(int);
4607 get_signal_t os::Linux::get_signal_action = NULL;
4609 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4610 struct sigaction *actp = NULL;
4612 if (libjsig_is_loaded) {
4613 // Retrieve the old signal handler from libjsig
4614 actp = (*get_signal_action)(sig);
4615 }
4616 if (actp == NULL) {
4617 // Retrieve the preinstalled signal handler from jvm
4618 actp = get_preinstalled_handler(sig);
4619 }
4621 return actp;
4622 }
4624 static bool call_chained_handler(struct sigaction *actp, int sig,
4625 siginfo_t *siginfo, void *context) {
4626 // Call the old signal handler
4627 if (actp->sa_handler == SIG_DFL) {
4628 // It's more reasonable to let jvm treat it as an unexpected exception
4629 // instead of taking the default action.
4630 return false;
4631 } else if (actp->sa_handler != SIG_IGN) {
4632 if ((actp->sa_flags & SA_NODEFER) == 0) {
4633 // automaticlly block the signal
4634 sigaddset(&(actp->sa_mask), sig);
4635 }
4637 sa_handler_t hand = NULL;
4638 sa_sigaction_t sa = NULL;
4639 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4640 // retrieve the chained handler
4641 if (siginfo_flag_set) {
4642 sa = actp->sa_sigaction;
4643 } else {
4644 hand = actp->sa_handler;
4645 }
4647 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4648 actp->sa_handler = SIG_DFL;
4649 }
4651 // try to honor the signal mask
4652 sigset_t oset;
4653 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4655 // call into the chained handler
4656 if (siginfo_flag_set) {
4657 (*sa)(sig, siginfo, context);
4658 } else {
4659 (*hand)(sig);
4660 }
4662 // restore the signal mask
4663 pthread_sigmask(SIG_SETMASK, &oset, 0);
4664 }
4665 // Tell jvm's signal handler the signal is taken care of.
4666 return true;
4667 }
4669 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4670 bool chained = false;
4671 // signal-chaining
4672 if (UseSignalChaining) {
4673 struct sigaction *actp = get_chained_signal_action(sig);
4674 if (actp != NULL) {
4675 chained = call_chained_handler(actp, sig, siginfo, context);
4676 }
4677 }
4678 return chained;
4679 }
4681 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4682 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4683 return &sigact[sig];
4684 }
4685 return NULL;
4686 }
4688 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4689 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4690 sigact[sig] = oldAct;
4691 sigs |= (unsigned int)1 << sig;
4692 }
4694 // for diagnostic
4695 int os::Linux::sigflags[MAXSIGNUM];
4697 int os::Linux::get_our_sigflags(int sig) {
4698 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4699 return sigflags[sig];
4700 }
4702 void os::Linux::set_our_sigflags(int sig, int flags) {
4703 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4704 sigflags[sig] = flags;
4705 }
4707 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4708 // Check for overwrite.
4709 struct sigaction oldAct;
4710 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4712 void* oldhand = oldAct.sa_sigaction
4713 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4714 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4715 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4716 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4717 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4718 if (AllowUserSignalHandlers || !set_installed) {
4719 // Do not overwrite; user takes responsibility to forward to us.
4720 return;
4721 } else if (UseSignalChaining) {
4722 // save the old handler in jvm
4723 save_preinstalled_handler(sig, oldAct);
4724 // libjsig also interposes the sigaction() call below and saves the
4725 // old sigaction on it own.
4726 } else {
4727 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4728 "%#lx for signal %d.", (long)oldhand, sig));
4729 }
4730 }
4732 struct sigaction sigAct;
4733 sigfillset(&(sigAct.sa_mask));
4734 sigAct.sa_handler = SIG_DFL;
4735 if (!set_installed) {
4736 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4737 } else {
4738 sigAct.sa_sigaction = signalHandler;
4739 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4740 }
4741 // Save flags, which are set by ours
4742 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4743 sigflags[sig] = sigAct.sa_flags;
4745 int ret = sigaction(sig, &sigAct, &oldAct);
4746 assert(ret == 0, "check");
4748 void* oldhand2 = oldAct.sa_sigaction
4749 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4750 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4751 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4752 }
4754 // install signal handlers for signals that HotSpot needs to
4755 // handle in order to support Java-level exception handling.
4757 void os::Linux::install_signal_handlers() {
4758 if (!signal_handlers_are_installed) {
4759 signal_handlers_are_installed = true;
4761 // signal-chaining
4762 typedef void (*signal_setting_t)();
4763 signal_setting_t begin_signal_setting = NULL;
4764 signal_setting_t end_signal_setting = NULL;
4765 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4766 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4767 if (begin_signal_setting != NULL) {
4768 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4769 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4770 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4771 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4772 libjsig_is_loaded = true;
4773 assert(UseSignalChaining, "should enable signal-chaining");
4774 }
4775 if (libjsig_is_loaded) {
4776 // Tell libjsig jvm is setting signal handlers
4777 (*begin_signal_setting)();
4778 }
4780 set_signal_handler(SIGSEGV, true);
4781 set_signal_handler(SIGPIPE, true);
4782 set_signal_handler(SIGBUS, true);
4783 set_signal_handler(SIGILL, true);
4784 set_signal_handler(SIGFPE, true);
4785 #if defined(PPC64)
4786 set_signal_handler(SIGTRAP, true);
4787 #endif
4788 set_signal_handler(SIGXFSZ, true);
4790 if (libjsig_is_loaded) {
4791 // Tell libjsig jvm finishes setting signal handlers
4792 (*end_signal_setting)();
4793 }
4795 // We don't activate signal checker if libjsig is in place, we trust ourselves
4796 // and if UserSignalHandler is installed all bets are off.
4797 // Log that signal checking is off only if -verbose:jni is specified.
4798 if (CheckJNICalls) {
4799 if (libjsig_is_loaded) {
4800 if (PrintJNIResolving) {
4801 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4802 }
4803 check_signals = false;
4804 }
4805 if (AllowUserSignalHandlers) {
4806 if (PrintJNIResolving) {
4807 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4808 }
4809 check_signals = false;
4810 }
4811 }
4812 }
4813 }
4815 // This is the fastest way to get thread cpu time on Linux.
4816 // Returns cpu time (user+sys) for any thread, not only for current.
4817 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4818 // It might work on 2.6.10+ with a special kernel/glibc patch.
4819 // For reference, please, see IEEE Std 1003.1-2004:
4820 // http://www.unix.org/single_unix_specification
4822 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4823 struct timespec tp;
4824 int rc = os::Linux::clock_gettime(clockid, &tp);
4825 assert(rc == 0, "clock_gettime is expected to return 0 code");
4827 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4828 }
4830 /////
4831 // glibc on Linux platform uses non-documented flag
4832 // to indicate, that some special sort of signal
4833 // trampoline is used.
4834 // We will never set this flag, and we should
4835 // ignore this flag in our diagnostic
4836 #ifdef SIGNIFICANT_SIGNAL_MASK
4837 #undef SIGNIFICANT_SIGNAL_MASK
4838 #endif
4839 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4841 static const char* get_signal_handler_name(address handler,
4842 char* buf, int buflen) {
4843 int offset = 0;
4844 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4845 if (found) {
4846 // skip directory names
4847 const char *p1, *p2;
4848 p1 = buf;
4849 size_t len = strlen(os::file_separator());
4850 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4851 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4852 } else {
4853 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4854 }
4855 return buf;
4856 }
4858 static void print_signal_handler(outputStream* st, int sig,
4859 char* buf, size_t buflen) {
4860 struct sigaction sa;
4862 sigaction(sig, NULL, &sa);
4864 // See comment for SIGNIFICANT_SIGNAL_MASK define
4865 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4867 st->print("%s: ", os::exception_name(sig, buf, buflen));
4869 address handler = (sa.sa_flags & SA_SIGINFO)
4870 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4871 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4873 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4874 st->print("SIG_DFL");
4875 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4876 st->print("SIG_IGN");
4877 } else {
4878 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4879 }
4881 st->print(", sa_mask[0]=");
4882 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4884 address rh = VMError::get_resetted_sighandler(sig);
4885 // May be, handler was resetted by VMError?
4886 if(rh != NULL) {
4887 handler = rh;
4888 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4889 }
4891 st->print(", sa_flags=");
4892 os::Posix::print_sa_flags(st, sa.sa_flags);
4894 // Check: is it our handler?
4895 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4896 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4897 // It is our signal handler
4898 // check for flags, reset system-used one!
4899 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4900 st->print(
4901 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4902 os::Linux::get_our_sigflags(sig));
4903 }
4904 }
4905 st->cr();
4906 }
4909 #define DO_SIGNAL_CHECK(sig) \
4910 if (!sigismember(&check_signal_done, sig)) \
4911 os::Linux::check_signal_handler(sig)
4913 // This method is a periodic task to check for misbehaving JNI applications
4914 // under CheckJNI, we can add any periodic checks here
4916 void os::run_periodic_checks() {
4918 if (check_signals == false) return;
4920 // SEGV and BUS if overridden could potentially prevent
4921 // generation of hs*.log in the event of a crash, debugging
4922 // such a case can be very challenging, so we absolutely
4923 // check the following for a good measure:
4924 DO_SIGNAL_CHECK(SIGSEGV);
4925 DO_SIGNAL_CHECK(SIGILL);
4926 DO_SIGNAL_CHECK(SIGFPE);
4927 DO_SIGNAL_CHECK(SIGBUS);
4928 DO_SIGNAL_CHECK(SIGPIPE);
4929 DO_SIGNAL_CHECK(SIGXFSZ);
4930 #if defined(PPC64)
4931 DO_SIGNAL_CHECK(SIGTRAP);
4932 #endif
4934 // ReduceSignalUsage allows the user to override these handlers
4935 // see comments at the very top and jvm_solaris.h
4936 if (!ReduceSignalUsage) {
4937 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4938 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4939 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4940 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4941 }
4943 DO_SIGNAL_CHECK(SR_signum);
4944 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4945 }
4947 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4949 static os_sigaction_t os_sigaction = NULL;
4951 void os::Linux::check_signal_handler(int sig) {
4952 char buf[O_BUFLEN];
4953 address jvmHandler = NULL;
4956 struct sigaction act;
4957 if (os_sigaction == NULL) {
4958 // only trust the default sigaction, in case it has been interposed
4959 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4960 if (os_sigaction == NULL) return;
4961 }
4963 os_sigaction(sig, (struct sigaction*)NULL, &act);
4966 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4968 address thisHandler = (act.sa_flags & SA_SIGINFO)
4969 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4970 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4973 switch(sig) {
4974 case SIGSEGV:
4975 case SIGBUS:
4976 case SIGFPE:
4977 case SIGPIPE:
4978 case SIGILL:
4979 case SIGXFSZ:
4980 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4981 break;
4983 case SHUTDOWN1_SIGNAL:
4984 case SHUTDOWN2_SIGNAL:
4985 case SHUTDOWN3_SIGNAL:
4986 case BREAK_SIGNAL:
4987 jvmHandler = (address)user_handler();
4988 break;
4990 case INTERRUPT_SIGNAL:
4991 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4992 break;
4994 default:
4995 if (sig == SR_signum) {
4996 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4997 } else {
4998 return;
4999 }
5000 break;
5001 }
5003 if (thisHandler != jvmHandler) {
5004 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
5005 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
5006 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
5007 // No need to check this sig any longer
5008 sigaddset(&check_signal_done, sig);
5009 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
5010 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
5011 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
5012 exception_name(sig, buf, O_BUFLEN));
5013 }
5014 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
5015 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
5016 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
5017 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
5018 // No need to check this sig any longer
5019 sigaddset(&check_signal_done, sig);
5020 }
5022 // Dump all the signal
5023 if (sigismember(&check_signal_done, sig)) {
5024 print_signal_handlers(tty, buf, O_BUFLEN);
5025 }
5026 }
5028 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
5030 extern bool signal_name(int signo, char* buf, size_t len);
5032 const char* os::exception_name(int exception_code, char* buf, size_t size) {
5033 if (0 < exception_code && exception_code <= SIGRTMAX) {
5034 // signal
5035 if (!signal_name(exception_code, buf, size)) {
5036 jio_snprintf(buf, size, "SIG%d", exception_code);
5037 }
5038 return buf;
5039 } else {
5040 return NULL;
5041 }
5042 }
5044 // this is called _before_ most of the global arguments have been parsed
5045 void os::init(void) {
5046 char dummy; /* used to get a guess on initial stack address */
5048 // With LinuxThreads the JavaMain thread pid (primordial thread)
5049 // is different than the pid of the java launcher thread.
5050 // So, on Linux, the launcher thread pid is passed to the VM
5051 // via the sun.java.launcher.pid property.
5052 // Use this property instead of getpid() if it was correctly passed.
5053 // See bug 6351349.
5054 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
5056 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
5058 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5060 init_random(1234567);
5062 ThreadCritical::initialize();
5064 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5065 if (Linux::page_size() == -1) {
5066 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
5067 strerror(errno)));
5068 }
5069 init_page_sizes((size_t) Linux::page_size());
5071 Linux::initialize_system_info();
5073 // _main_thread points to the thread that created/loaded the JVM.
5074 Linux::_main_thread = pthread_self();
5076 Linux::clock_init();
5077 initial_time_count = javaTimeNanos();
5079 // pthread_condattr initialization for monotonic clock
5080 int status;
5081 pthread_condattr_t* _condattr = os::Linux::condAttr();
5082 if ((status = pthread_condattr_init(_condattr)) != 0) {
5083 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
5084 }
5085 // Only set the clock if CLOCK_MONOTONIC is available
5086 if (Linux::supports_monotonic_clock()) {
5087 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
5088 if (status == EINVAL) {
5089 warning("Unable to use monotonic clock with relative timed-waits" \
5090 " - changes to the time-of-day clock may have adverse affects");
5091 } else {
5092 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
5093 }
5094 }
5095 }
5096 // else it defaults to CLOCK_REALTIME
5098 pthread_mutex_init(&dl_mutex, NULL);
5100 // If the pagesize of the VM is greater than 8K determine the appropriate
5101 // number of initial guard pages. The user can change this with the
5102 // command line arguments, if needed.
5103 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
5104 StackYellowPages = 1;
5105 StackRedPages = 1;
5106 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
5107 }
5109 // retrieve entry point for pthread_setname_np
5110 Linux::_pthread_setname_np =
5111 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
5113 }
5115 // To install functions for atexit system call
5116 extern "C" {
5117 static void perfMemory_exit_helper() {
5118 perfMemory_exit();
5119 }
5120 }
5122 void os::pd_init_container_support() {
5123 OSContainer::init();
5124 }
5126 // this is called _after_ the global arguments have been parsed
5127 jint os::init_2(void)
5128 {
5129 Linux::fast_thread_clock_init();
5131 // Allocate a single page and mark it as readable for safepoint polling
5132 #ifdef OPT_SAFEPOINT
5133 void * p = (void *)(0x10000);
5134 address polling_page = (address) ::mmap(p, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5135 #else
5136 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5137 #endif
5138 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
5140 os::set_polling_page( polling_page );
5142 #ifndef PRODUCT
5143 if(Verbose && PrintMiscellaneous)
5144 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5145 #endif
5147 if (!UseMembar) {
5148 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5149 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
5150 os::set_memory_serialize_page( mem_serialize_page );
5152 #ifndef PRODUCT
5153 if(Verbose && PrintMiscellaneous)
5154 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5155 #endif
5156 }
5158 // initialize suspend/resume support - must do this before signal_sets_init()
5159 if (SR_initialize() != 0) {
5160 perror("SR_initialize failed");
5161 return JNI_ERR;
5162 }
5164 Linux::signal_sets_init();
5165 Linux::install_signal_handlers();
5167 // Check minimum allowable stack size for thread creation and to initialize
5168 // the java system classes, including StackOverflowError - depends on page
5169 // size. Add a page for compiler2 recursion in main thread.
5170 // Add in 2*BytesPerWord times page size to account for VM stack during
5171 // class initialization depending on 32 or 64 bit VM.
5173 /* 2014/1/2 Liao: JDK8 requires larger -Xss option.
5174 * TongWeb cannot run with -Xss192K.
5175 * We are not sure whether this causes errors, so simply print a warning. */
5176 size_t min_stack_allowed_jdk6 = os::Linux::min_stack_allowed;
5177 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
5178 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
5179 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
5181 size_t threadStackSizeInBytes = ThreadStackSize * K;
5182 if (threadStackSizeInBytes != 0 &&
5183 threadStackSizeInBytes < min_stack_allowed_jdk6) {
5184 tty->print_cr("\nThe stack size specified is too small, "
5185 "Specify at least %dk",
5186 os::Linux::min_stack_allowed/ K);
5187 return JNI_ERR;
5188 }
5190 // Make the stack size a multiple of the page size so that
5191 // the yellow/red zones can be guarded.
5192 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5193 vm_page_size()));
5195 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5197 #if defined(IA32)
5198 workaround_expand_exec_shield_cs_limit();
5199 #endif
5201 Linux::libpthread_init();
5202 if (PrintMiscellaneous && (Verbose || WizardMode)) {
5203 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
5204 Linux::glibc_version(), Linux::libpthread_version(),
5205 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
5206 }
5208 if (UseNUMA) {
5209 if (!Linux::libnuma_init()) {
5210 UseNUMA = false;
5211 } else {
5212 if ((Linux::numa_max_node() < 1)) {
5213 // There's only one node(they start from 0), disable NUMA.
5214 UseNUMA = false;
5215 }
5216 }
5217 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5218 // we can make the adaptive lgrp chunk resizing work. If the user specified
5219 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
5220 // disable adaptive resizing.
5221 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5222 if (FLAG_IS_DEFAULT(UseNUMA)) {
5223 UseNUMA = false;
5224 } else {
5225 if (FLAG_IS_DEFAULT(UseLargePages) &&
5226 FLAG_IS_DEFAULT(UseSHM) &&
5227 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
5228 UseLargePages = false;
5229 } else {
5230 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
5231 UseAdaptiveSizePolicy = false;
5232 UseAdaptiveNUMAChunkSizing = false;
5233 }
5234 }
5235 }
5236 if (!UseNUMA && ForceNUMA) {
5237 UseNUMA = true;
5238 }
5239 }
5241 if (MaxFDLimit) {
5242 // set the number of file descriptors to max. print out error
5243 // if getrlimit/setrlimit fails but continue regardless.
5244 struct rlimit nbr_files;
5245 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5246 if (status != 0) {
5247 if (PrintMiscellaneous && (Verbose || WizardMode))
5248 perror("os::init_2 getrlimit failed");
5249 } else {
5250 nbr_files.rlim_cur = nbr_files.rlim_max;
5251 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5252 if (status != 0) {
5253 if (PrintMiscellaneous && (Verbose || WizardMode))
5254 perror("os::init_2 setrlimit failed");
5255 }
5256 }
5257 }
5259 // Initialize lock used to serialize thread creation (see os::create_thread)
5260 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5262 // at-exit methods are called in the reverse order of their registration.
5263 // atexit functions are called on return from main or as a result of a
5264 // call to exit(3C). There can be only 32 of these functions registered
5265 // and atexit() does not set errno.
5267 if (PerfAllowAtExitRegistration) {
5268 // only register atexit functions if PerfAllowAtExitRegistration is set.
5269 // atexit functions can be delayed until process exit time, which
5270 // can be problematic for embedded VM situations. Embedded VMs should
5271 // call DestroyJavaVM() to assure that VM resources are released.
5273 // note: perfMemory_exit_helper atexit function may be removed in
5274 // the future if the appropriate cleanup code can be added to the
5275 // VM_Exit VMOperation's doit method.
5276 if (atexit(perfMemory_exit_helper) != 0) {
5277 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5278 }
5279 }
5281 // initialize thread priority policy
5282 prio_init();
5284 return JNI_OK;
5285 }
5287 // Mark the polling page as unreadable
5288 void os::make_polling_page_unreadable(void) {
5289 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5290 fatal("Could not disable polling page");
5291 };
5293 // Mark the polling page as readable
5294 void os::make_polling_page_readable(void) {
5295 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5296 fatal("Could not enable polling page");
5297 }
5298 };
5300 static int os_cpu_count(const cpu_set_t* cpus) {
5301 int count = 0;
5302 // only look up to the number of configured processors
5303 for (int i = 0; i < os::processor_count(); i++) {
5304 if (CPU_ISSET(i, cpus)) {
5305 count++;
5306 }
5307 }
5308 return count;
5309 }
5311 // Get the current number of available processors for this process.
5312 // This value can change at any time during a process's lifetime.
5313 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5314 // If anything goes wrong we fallback to returning the number of online
5315 // processors - which can be greater than the number available to the process.
5316 int os::Linux::active_processor_count() {
5317 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5318 int cpus_size = sizeof(cpu_set_t);
5319 int cpu_count = 0;
5321 // pid 0 means the current thread - which we have to assume represents the process
5322 if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
5323 cpu_count = os_cpu_count(&cpus);
5324 if (PrintActiveCpus) {
5325 tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5326 }
5327 }
5328 else {
5329 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5330 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5331 "which may exceed available processors", strerror(errno), cpu_count);
5332 }
5334 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5335 return cpu_count;
5336 }
5338 // Determine the active processor count from one of
5339 // three different sources:
5340 //
5341 // 1. User option -XX:ActiveProcessorCount
5342 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5343 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5344 //
5345 // Option 1, if specified, will always override.
5346 // If the cgroup subsystem is active and configured, we
5347 // will return the min of the cgroup and option 2 results.
5348 // This is required since tools, such as numactl, that
5349 // alter cpu affinity do not update cgroup subsystem
5350 // cpuset configuration files.
5351 int os::active_processor_count() {
5352 // User has overridden the number of active processors
5353 if (ActiveProcessorCount > 0) {
5354 if (PrintActiveCpus) {
5355 tty->print_cr("active_processor_count: "
5356 "active processor count set by user : %d",
5357 ActiveProcessorCount);
5358 }
5359 return ActiveProcessorCount;
5360 }
5362 int active_cpus;
5363 if (OSContainer::is_containerized()) {
5364 active_cpus = OSContainer::active_processor_count();
5365 if (PrintActiveCpus) {
5366 tty->print_cr("active_processor_count: determined by OSContainer: %d",
5367 active_cpus);
5368 }
5369 } else {
5370 active_cpus = os::Linux::active_processor_count();
5371 }
5373 return active_cpus;
5374 }
5376 void os::set_native_thread_name(const char *name) {
5377 if (Linux::_pthread_setname_np) {
5378 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
5379 snprintf(buf, sizeof(buf), "%s", name);
5380 buf[sizeof(buf) - 1] = '\0';
5381 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
5382 // ERANGE should not happen; all other errors should just be ignored.
5383 assert(rc != ERANGE, "pthread_setname_np failed");
5384 }
5385 }
5387 bool os::distribute_processes(uint length, uint* distribution) {
5388 // Not yet implemented.
5389 return false;
5390 }
5392 bool os::bind_to_processor(uint processor_id) {
5393 // Not yet implemented.
5394 return false;
5395 }
5397 ///
5399 void os::SuspendedThreadTask::internal_do_task() {
5400 if (do_suspend(_thread->osthread())) {
5401 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5402 do_task(context);
5403 do_resume(_thread->osthread());
5404 }
5405 }
5407 class PcFetcher : public os::SuspendedThreadTask {
5408 public:
5409 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5410 ExtendedPC result();
5411 protected:
5412 void do_task(const os::SuspendedThreadTaskContext& context);
5413 private:
5414 ExtendedPC _epc;
5415 };
5417 ExtendedPC PcFetcher::result() {
5418 guarantee(is_done(), "task is not done yet.");
5419 return _epc;
5420 }
5422 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5423 Thread* thread = context.thread();
5424 OSThread* osthread = thread->osthread();
5425 if (osthread->ucontext() != NULL) {
5426 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5427 } else {
5428 // NULL context is unexpected, double-check this is the VMThread
5429 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5430 }
5431 }
5433 // Suspends the target using the signal mechanism and then grabs the PC before
5434 // resuming the target. Used by the flat-profiler only
5435 ExtendedPC os::get_thread_pc(Thread* thread) {
5436 // Make sure that it is called by the watcher for the VMThread
5437 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5438 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5440 PcFetcher fetcher(thread);
5441 fetcher.run();
5442 return fetcher.result();
5443 }
5445 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5446 {
5447 if (is_NPTL()) {
5448 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5449 } else {
5450 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5451 // word back to default 64bit precision if condvar is signaled. Java
5452 // wants 53bit precision. Save and restore current value.
5453 int fpu = get_fpu_control_word();
5454 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5455 set_fpu_control_word(fpu);
5456 return status;
5457 }
5458 }
5460 ////////////////////////////////////////////////////////////////////////////////
5461 // debug support
5463 bool os::find(address addr, outputStream* st) {
5464 Dl_info dlinfo;
5465 memset(&dlinfo, 0, sizeof(dlinfo));
5466 if (dladdr(addr, &dlinfo) != 0) {
5467 st->print(PTR_FORMAT ": ", addr);
5468 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5469 st->print("%s+%#x", dlinfo.dli_sname,
5470 addr - (intptr_t)dlinfo.dli_saddr);
5471 } else if (dlinfo.dli_fbase != NULL) {
5472 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5473 } else {
5474 st->print("<absolute address>");
5475 }
5476 if (dlinfo.dli_fname != NULL) {
5477 st->print(" in %s", dlinfo.dli_fname);
5478 }
5479 if (dlinfo.dli_fbase != NULL) {
5480 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5481 }
5482 st->cr();
5484 if (Verbose) {
5485 // decode some bytes around the PC
5486 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5487 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5488 address lowest = (address) dlinfo.dli_sname;
5489 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5490 if (begin < lowest) begin = lowest;
5491 Dl_info dlinfo2;
5492 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5493 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5494 end = (address) dlinfo2.dli_saddr;
5495 Disassembler::decode(begin, end, st);
5496 }
5497 return true;
5498 }
5499 return false;
5500 }
5502 ////////////////////////////////////////////////////////////////////////////////
5503 // misc
5505 // This does not do anything on Linux. This is basically a hook for being
5506 // able to use structured exception handling (thread-local exception filters)
5507 // on, e.g., Win32.
5508 void
5509 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5510 JavaCallArguments* args, Thread* thread) {
5511 f(value, method, args, thread);
5512 }
5514 void os::print_statistics() {
5515 }
5517 int os::message_box(const char* title, const char* message) {
5518 int i;
5519 fdStream err(defaultStream::error_fd());
5520 for (i = 0; i < 78; i++) err.print_raw("=");
5521 err.cr();
5522 err.print_raw_cr(title);
5523 for (i = 0; i < 78; i++) err.print_raw("-");
5524 err.cr();
5525 err.print_raw_cr(message);
5526 for (i = 0; i < 78; i++) err.print_raw("=");
5527 err.cr();
5529 char buf[16];
5530 // Prevent process from exiting upon "read error" without consuming all CPU
5531 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5533 return buf[0] == 'y' || buf[0] == 'Y';
5534 }
5536 int os::stat(const char *path, struct stat *sbuf) {
5537 char pathbuf[MAX_PATH];
5538 if (strlen(path) > MAX_PATH - 1) {
5539 errno = ENAMETOOLONG;
5540 return -1;
5541 }
5542 os::native_path(strcpy(pathbuf, path));
5543 return ::stat(pathbuf, sbuf);
5544 }
5546 bool os::check_heap(bool force) {
5547 return true;
5548 }
5550 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5551 return ::vsnprintf(buf, count, format, args);
5552 }
5554 // Is a (classpath) directory empty?
5555 bool os::dir_is_empty(const char* path) {
5556 DIR *dir = NULL;
5557 struct dirent *ptr;
5559 dir = opendir(path);
5560 if (dir == NULL) return true;
5562 /* Scan the directory */
5563 bool result = true;
5564 while (result && (ptr = readdir(dir)) != NULL) {
5565 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5566 result = false;
5567 }
5568 }
5569 closedir(dir);
5570 return result;
5571 }
5573 // This code originates from JDK's sysOpen and open64_w
5574 // from src/solaris/hpi/src/system_md.c
5576 #ifndef O_DELETE
5577 #define O_DELETE 0x10000
5578 #endif
5580 // Open a file. Unlink the file immediately after open returns
5581 // if the specified oflag has the O_DELETE flag set.
5582 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5584 int os::open(const char *path, int oflag, int mode) {
5586 if (strlen(path) > MAX_PATH - 1) {
5587 errno = ENAMETOOLONG;
5588 return -1;
5589 }
5590 int fd;
5591 int o_delete = (oflag & O_DELETE);
5592 oflag = oflag & ~O_DELETE;
5594 fd = ::open64(path, oflag, mode);
5595 if (fd == -1) return -1;
5597 //If the open succeeded, the file might still be a directory
5598 {
5599 struct stat64 buf64;
5600 int ret = ::fstat64(fd, &buf64);
5601 int st_mode = buf64.st_mode;
5603 if (ret != -1) {
5604 if ((st_mode & S_IFMT) == S_IFDIR) {
5605 errno = EISDIR;
5606 ::close(fd);
5607 return -1;
5608 }
5609 } else {
5610 ::close(fd);
5611 return -1;
5612 }
5613 }
5615 /*
5616 * All file descriptors that are opened in the JVM and not
5617 * specifically destined for a subprocess should have the
5618 * close-on-exec flag set. If we don't set it, then careless 3rd
5619 * party native code might fork and exec without closing all
5620 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5621 * UNIXProcess.c), and this in turn might:
5622 *
5623 * - cause end-of-file to fail to be detected on some file
5624 * descriptors, resulting in mysterious hangs, or
5625 *
5626 * - might cause an fopen in the subprocess to fail on a system
5627 * suffering from bug 1085341.
5628 *
5629 * (Yes, the default setting of the close-on-exec flag is a Unix
5630 * design flaw)
5631 *
5632 * See:
5633 * 1085341: 32-bit stdio routines should support file descriptors >255
5634 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5635 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5636 */
5637 #ifdef FD_CLOEXEC
5638 {
5639 int flags = ::fcntl(fd, F_GETFD);
5640 if (flags != -1)
5641 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5642 }
5643 #endif
5645 if (o_delete != 0) {
5646 ::unlink(path);
5647 }
5648 return fd;
5649 }
5652 // create binary file, rewriting existing file if required
5653 int os::create_binary_file(const char* path, bool rewrite_existing) {
5654 int oflags = O_WRONLY | O_CREAT;
5655 if (!rewrite_existing) {
5656 oflags |= O_EXCL;
5657 }
5658 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5659 }
5661 // return current position of file pointer
5662 jlong os::current_file_offset(int fd) {
5663 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5664 }
5666 // move file pointer to the specified offset
5667 jlong os::seek_to_file_offset(int fd, jlong offset) {
5668 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5669 }
5671 // This code originates from JDK's sysAvailable
5672 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5674 int os::available(int fd, jlong *bytes) {
5675 jlong cur, end;
5676 int mode;
5677 struct stat64 buf64;
5679 if (::fstat64(fd, &buf64) >= 0) {
5680 mode = buf64.st_mode;
5681 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5682 /*
5683 * XXX: is the following call interruptible? If so, this might
5684 * need to go through the INTERRUPT_IO() wrapper as for other
5685 * blocking, interruptible calls in this file.
5686 */
5687 int n;
5688 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5689 *bytes = n;
5690 return 1;
5691 }
5692 }
5693 }
5694 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5695 return 0;
5696 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5697 return 0;
5698 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5699 return 0;
5700 }
5701 *bytes = end - cur;
5702 return 1;
5703 }
5705 int os::socket_available(int fd, jint *pbytes) {
5706 // Linux doc says EINTR not returned, unlike Solaris
5707 int ret = ::ioctl(fd, FIONREAD, pbytes);
5709 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5710 // is expected to return 0 on failure and 1 on success to the jdk.
5711 return (ret < 0) ? 0 : 1;
5712 }
5714 // Map a block of memory.
5715 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5716 char *addr, size_t bytes, bool read_only,
5717 bool allow_exec) {
5718 int prot;
5719 int flags = MAP_PRIVATE;
5721 if (read_only) {
5722 prot = PROT_READ;
5723 } else {
5724 prot = PROT_READ | PROT_WRITE;
5725 }
5727 if (allow_exec) {
5728 prot |= PROT_EXEC;
5729 }
5731 if (addr != NULL) {
5732 flags |= MAP_FIXED;
5733 }
5735 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5736 fd, file_offset);
5737 if (mapped_address == MAP_FAILED) {
5738 return NULL;
5739 }
5740 return mapped_address;
5741 }
5744 // Remap a block of memory.
5745 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5746 char *addr, size_t bytes, bool read_only,
5747 bool allow_exec) {
5748 // same as map_memory() on this OS
5749 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5750 allow_exec);
5751 }
5754 // Unmap a block of memory.
5755 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5756 return munmap(addr, bytes) == 0;
5757 }
5759 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5761 static clockid_t thread_cpu_clockid(Thread* thread) {
5762 pthread_t tid = thread->osthread()->pthread_id();
5763 clockid_t clockid;
5765 // Get thread clockid
5766 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5767 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5768 return clockid;
5769 }
5771 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5772 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5773 // of a thread.
5774 //
5775 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5776 // the fast estimate available on the platform.
5778 jlong os::current_thread_cpu_time() {
5779 if (os::Linux::supports_fast_thread_cpu_time()) {
5780 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5781 } else {
5782 // return user + sys since the cost is the same
5783 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5784 }
5785 }
5787 jlong os::thread_cpu_time(Thread* thread) {
5788 // consistent with what current_thread_cpu_time() returns
5789 if (os::Linux::supports_fast_thread_cpu_time()) {
5790 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5791 } else {
5792 return slow_thread_cpu_time(thread, true /* user + sys */);
5793 }
5794 }
5796 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5797 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5798 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5799 } else {
5800 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5801 }
5802 }
5804 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5805 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5806 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5807 } else {
5808 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5809 }
5810 }
5812 //
5813 // -1 on error.
5814 //
5816 PRAGMA_DIAG_PUSH
5817 PRAGMA_FORMAT_NONLITERAL_IGNORED
5818 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5819 static bool proc_task_unchecked = true;
5820 static const char *proc_stat_path = "/proc/%d/stat";
5821 pid_t tid = thread->osthread()->thread_id();
5822 char *s;
5823 char stat[2048];
5824 int statlen;
5825 char proc_name[64];
5826 int count;
5827 long sys_time, user_time;
5828 char cdummy;
5829 int idummy;
5830 long ldummy;
5831 FILE *fp;
5833 // The /proc/<tid>/stat aggregates per-process usage on
5834 // new Linux kernels 2.6+ where NPTL is supported.
5835 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5836 // See bug 6328462.
5837 // There possibly can be cases where there is no directory
5838 // /proc/self/task, so we check its availability.
5839 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5840 // This is executed only once
5841 proc_task_unchecked = false;
5842 fp = fopen("/proc/self/task", "r");
5843 if (fp != NULL) {
5844 proc_stat_path = "/proc/self/task/%d/stat";
5845 fclose(fp);
5846 }
5847 }
5849 sprintf(proc_name, proc_stat_path, tid);
5850 fp = fopen(proc_name, "r");
5851 if ( fp == NULL ) return -1;
5852 statlen = fread(stat, 1, 2047, fp);
5853 stat[statlen] = '\0';
5854 fclose(fp);
5856 // Skip pid and the command string. Note that we could be dealing with
5857 // weird command names, e.g. user could decide to rename java launcher
5858 // to "java 1.4.2 :)", then the stat file would look like
5859 // 1234 (java 1.4.2 :)) R ... ...
5860 // We don't really need to know the command string, just find the last
5861 // occurrence of ")" and then start parsing from there. See bug 4726580.
5862 s = strrchr(stat, ')');
5863 if (s == NULL ) return -1;
5865 // Skip blank chars
5866 do s++; while (isspace(*s));
5868 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5869 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5870 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5871 &user_time, &sys_time);
5872 if ( count != 13 ) return -1;
5873 if (user_sys_cpu_time) {
5874 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5875 } else {
5876 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5877 }
5878 }
5879 PRAGMA_DIAG_POP
5881 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5882 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5883 info_ptr->may_skip_backward = false; // elapsed time not wall time
5884 info_ptr->may_skip_forward = false; // elapsed time not wall time
5885 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5886 }
5888 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5889 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5890 info_ptr->may_skip_backward = false; // elapsed time not wall time
5891 info_ptr->may_skip_forward = false; // elapsed time not wall time
5892 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5893 }
5895 bool os::is_thread_cpu_time_supported() {
5896 return true;
5897 }
5899 // System loadavg support. Returns -1 if load average cannot be obtained.
5900 // Linux doesn't yet have a (official) notion of processor sets,
5901 // so just return the system wide load average.
5902 int os::loadavg(double loadavg[], int nelem) {
5903 return ::getloadavg(loadavg, nelem);
5904 }
5906 void os::pause() {
5907 char filename[MAX_PATH];
5908 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5909 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5910 } else {
5911 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5912 }
5914 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5915 if (fd != -1) {
5916 struct stat buf;
5917 ::close(fd);
5918 while (::stat(filename, &buf) == 0) {
5919 (void)::poll(NULL, 0, 100);
5920 }
5921 } else {
5922 jio_fprintf(stderr,
5923 "Could not open pause file '%s', continuing immediately.\n", filename);
5924 }
5925 }
5928 // Refer to the comments in os_solaris.cpp park-unpark.
5929 //
5930 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5931 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5932 // For specifics regarding the bug see GLIBC BUGID 261237 :
5933 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5934 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5935 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5936 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5937 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5938 // and monitorenter when we're using 1-0 locking. All those operations may result in
5939 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5940 // of libpthread avoids the problem, but isn't practical.
5941 //
5942 // Possible remedies:
5943 //
5944 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5945 // This is palliative and probabilistic, however. If the thread is preempted
5946 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5947 // than the minimum period may have passed, and the abstime may be stale (in the
5948 // past) resultin in a hang. Using this technique reduces the odds of a hang
5949 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5950 //
5951 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5952 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5953 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5954 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5955 // thread.
5956 //
5957 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5958 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5959 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5960 // This also works well. In fact it avoids kernel-level scalability impediments
5961 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5962 // timers in a graceful fashion.
5963 //
5964 // 4. When the abstime value is in the past it appears that control returns
5965 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5966 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5967 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5968 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5969 // It may be possible to avoid reinitialization by checking the return
5970 // value from pthread_cond_timedwait(). In addition to reinitializing the
5971 // condvar we must establish the invariant that cond_signal() is only called
5972 // within critical sections protected by the adjunct mutex. This prevents
5973 // cond_signal() from "seeing" a condvar that's in the midst of being
5974 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5975 // desirable signal-after-unlock optimization that avoids futile context switching.
5976 //
5977 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5978 // structure when a condvar is used or initialized. cond_destroy() would
5979 // release the helper structure. Our reinitialize-after-timedwait fix
5980 // put excessive stress on malloc/free and locks protecting the c-heap.
5981 //
5982 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5983 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5984 // and only enabling the work-around for vulnerable environments.
5986 // utility to compute the abstime argument to timedwait:
5987 // millis is the relative timeout time
5988 // abstime will be the absolute timeout time
5989 // TODO: replace compute_abstime() with unpackTime()
5991 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5992 if (millis < 0) millis = 0;
5994 jlong seconds = millis / 1000;
5995 millis %= 1000;
5996 if (seconds > 50000000) { // see man cond_timedwait(3T)
5997 seconds = 50000000;
5998 }
6000 if (os::Linux::supports_monotonic_clock()) {
6001 struct timespec now;
6002 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6003 assert_status(status == 0, status, "clock_gettime");
6004 abstime->tv_sec = now.tv_sec + seconds;
6005 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
6006 if (nanos >= NANOSECS_PER_SEC) {
6007 abstime->tv_sec += 1;
6008 nanos -= NANOSECS_PER_SEC;
6009 }
6010 abstime->tv_nsec = nanos;
6011 } else {
6012 struct timeval now;
6013 int status = gettimeofday(&now, NULL);
6014 assert(status == 0, "gettimeofday");
6015 abstime->tv_sec = now.tv_sec + seconds;
6016 long usec = now.tv_usec + millis * 1000;
6017 if (usec >= 1000000) {
6018 abstime->tv_sec += 1;
6019 usec -= 1000000;
6020 }
6021 abstime->tv_nsec = usec * 1000;
6022 }
6023 return abstime;
6024 }
6027 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
6028 // Conceptually TryPark() should be equivalent to park(0).
6030 int os::PlatformEvent::TryPark() {
6031 for (;;) {
6032 const int v = _Event ;
6033 guarantee ((v == 0) || (v == 1), "invariant") ;
6034 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
6035 }
6036 }
6038 void os::PlatformEvent::park() { // AKA "down()"
6039 // Invariant: Only the thread associated with the Event/PlatformEvent
6040 // may call park().
6041 // TODO: assert that _Assoc != NULL or _Assoc == Self
6042 int v ;
6043 for (;;) {
6044 v = _Event ;
6045 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6046 }
6047 guarantee (v >= 0, "invariant") ;
6048 if (v == 0) {
6049 // Do this the hard way by blocking ...
6050 int status = pthread_mutex_lock(_mutex);
6051 assert_status(status == 0, status, "mutex_lock");
6052 guarantee (_nParked == 0, "invariant") ;
6053 ++ _nParked ;
6054 while (_Event < 0) {
6055 status = pthread_cond_wait(_cond, _mutex);
6056 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6057 // Treat this the same as if the wait was interrupted
6058 if (status == ETIME) { status = EINTR; }
6059 assert_status(status == 0 || status == EINTR, status, "cond_wait");
6060 }
6061 -- _nParked ;
6063 _Event = 0 ;
6064 status = pthread_mutex_unlock(_mutex);
6065 assert_status(status == 0, status, "mutex_unlock");
6066 // Paranoia to ensure our locked and lock-free paths interact
6067 // correctly with each other.
6068 OrderAccess::fence();
6069 }
6070 guarantee (_Event >= 0, "invariant") ;
6071 }
6073 int os::PlatformEvent::park(jlong millis) {
6074 guarantee (_nParked == 0, "invariant") ;
6076 int v ;
6077 for (;;) {
6078 v = _Event ;
6079 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6080 }
6081 guarantee (v >= 0, "invariant") ;
6082 if (v != 0) return OS_OK ;
6084 // We do this the hard way, by blocking the thread.
6085 // Consider enforcing a minimum timeout value.
6086 struct timespec abst;
6087 compute_abstime(&abst, millis);
6089 int ret = OS_TIMEOUT;
6090 int status = pthread_mutex_lock(_mutex);
6091 assert_status(status == 0, status, "mutex_lock");
6092 guarantee (_nParked == 0, "invariant") ;
6093 ++_nParked ;
6095 // Object.wait(timo) will return because of
6096 // (a) notification
6097 // (b) timeout
6098 // (c) thread.interrupt
6099 //
6100 // Thread.interrupt and object.notify{All} both call Event::set.
6101 // That is, we treat thread.interrupt as a special case of notification.
6102 // The underlying Solaris implementation, cond_timedwait, admits
6103 // spurious/premature wakeups, but the JLS/JVM spec prevents the
6104 // JVM from making those visible to Java code. As such, we must
6105 // filter out spurious wakeups. We assume all ETIME returns are valid.
6106 //
6107 // TODO: properly differentiate simultaneous notify+interrupt.
6108 // In that case, we should propagate the notify to another waiter.
6110 while (_Event < 0) {
6111 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
6112 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6113 pthread_cond_destroy (_cond);
6114 pthread_cond_init (_cond, os::Linux::condAttr()) ;
6115 }
6116 assert_status(status == 0 || status == EINTR ||
6117 status == ETIME || status == ETIMEDOUT,
6118 status, "cond_timedwait");
6119 if (!FilterSpuriousWakeups) break ; // previous semantics
6120 if (status == ETIME || status == ETIMEDOUT) break ;
6121 // We consume and ignore EINTR and spurious wakeups.
6122 }
6123 --_nParked ;
6124 if (_Event >= 0) {
6125 ret = OS_OK;
6126 }
6127 _Event = 0 ;
6128 status = pthread_mutex_unlock(_mutex);
6129 assert_status(status == 0, status, "mutex_unlock");
6130 assert (_nParked == 0, "invariant") ;
6131 // Paranoia to ensure our locked and lock-free paths interact
6132 // correctly with each other.
6133 OrderAccess::fence();
6134 return ret;
6135 }
6137 void os::PlatformEvent::unpark() {
6138 // Transitions for _Event:
6139 // 0 :=> 1
6140 // 1 :=> 1
6141 // -1 :=> either 0 or 1; must signal target thread
6142 // That is, we can safely transition _Event from -1 to either
6143 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
6144 // unpark() calls.
6145 // See also: "Semaphores in Plan 9" by Mullender & Cox
6146 //
6147 // Note: Forcing a transition from "-1" to "1" on an unpark() means
6148 // that it will take two back-to-back park() calls for the owning
6149 // thread to block. This has the benefit of forcing a spurious return
6150 // from the first park() call after an unpark() call which will help
6151 // shake out uses of park() and unpark() without condition variables.
6153 if (Atomic::xchg(1, &_Event) >= 0) return;
6155 // Wait for the thread associated with the event to vacate
6156 int status = pthread_mutex_lock(_mutex);
6157 assert_status(status == 0, status, "mutex_lock");
6158 int AnyWaiters = _nParked;
6159 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6160 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
6161 AnyWaiters = 0;
6162 pthread_cond_signal(_cond);
6163 }
6164 status = pthread_mutex_unlock(_mutex);
6165 assert_status(status == 0, status, "mutex_unlock");
6166 if (AnyWaiters != 0) {
6167 status = pthread_cond_signal(_cond);
6168 assert_status(status == 0, status, "cond_signal");
6169 }
6171 // Note that we signal() _after dropping the lock for "immortal" Events.
6172 // This is safe and avoids a common class of futile wakeups. In rare
6173 // circumstances this can cause a thread to return prematurely from
6174 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
6175 // simply re-test the condition and re-park itself.
6176 }
6179 // JSR166
6180 // -------------------------------------------------------
6182 /*
6183 * The solaris and linux implementations of park/unpark are fairly
6184 * conservative for now, but can be improved. They currently use a
6185 * mutex/condvar pair, plus a a count.
6186 * Park decrements count if > 0, else does a condvar wait. Unpark
6187 * sets count to 1 and signals condvar. Only one thread ever waits
6188 * on the condvar. Contention seen when trying to park implies that someone
6189 * is unparking you, so don't wait. And spurious returns are fine, so there
6190 * is no need to track notifications.
6191 */
6193 /*
6194 * This code is common to linux and solaris and will be moved to a
6195 * common place in dolphin.
6196 *
6197 * The passed in time value is either a relative time in nanoseconds
6198 * or an absolute time in milliseconds. Either way it has to be unpacked
6199 * into suitable seconds and nanoseconds components and stored in the
6200 * given timespec structure.
6201 * Given time is a 64-bit value and the time_t used in the timespec is only
6202 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6203 * overflow if times way in the future are given. Further on Solaris versions
6204 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6205 * number of seconds, in abstime, is less than current_time + 100,000,000.
6206 * As it will be 28 years before "now + 100000000" will overflow we can
6207 * ignore overflow and just impose a hard-limit on seconds using the value
6208 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6209 * years from "now".
6210 */
6212 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6213 assert (time > 0, "convertTime");
6214 time_t max_secs = 0;
6216 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
6217 struct timeval now;
6218 int status = gettimeofday(&now, NULL);
6219 assert(status == 0, "gettimeofday");
6221 max_secs = now.tv_sec + MAX_SECS;
6223 if (isAbsolute) {
6224 jlong secs = time / 1000;
6225 if (secs > max_secs) {
6226 absTime->tv_sec = max_secs;
6227 } else {
6228 absTime->tv_sec = secs;
6229 }
6230 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6231 } else {
6232 jlong secs = time / NANOSECS_PER_SEC;
6233 if (secs >= MAX_SECS) {
6234 absTime->tv_sec = max_secs;
6235 absTime->tv_nsec = 0;
6236 } else {
6237 absTime->tv_sec = now.tv_sec + secs;
6238 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6239 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6240 absTime->tv_nsec -= NANOSECS_PER_SEC;
6241 ++absTime->tv_sec; // note: this must be <= max_secs
6242 }
6243 }
6244 }
6245 } else {
6246 // must be relative using monotonic clock
6247 struct timespec now;
6248 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6249 assert_status(status == 0, status, "clock_gettime");
6250 max_secs = now.tv_sec + MAX_SECS;
6251 jlong secs = time / NANOSECS_PER_SEC;
6252 if (secs >= MAX_SECS) {
6253 absTime->tv_sec = max_secs;
6254 absTime->tv_nsec = 0;
6255 } else {
6256 absTime->tv_sec = now.tv_sec + secs;
6257 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
6258 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6259 absTime->tv_nsec -= NANOSECS_PER_SEC;
6260 ++absTime->tv_sec; // note: this must be <= max_secs
6261 }
6262 }
6263 }
6264 assert(absTime->tv_sec >= 0, "tv_sec < 0");
6265 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6266 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6267 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6268 }
6270 void Parker::park(bool isAbsolute, jlong time) {
6271 // Ideally we'd do something useful while spinning, such
6272 // as calling unpackTime().
6274 // Optional fast-path check:
6275 // Return immediately if a permit is available.
6276 // We depend on Atomic::xchg() having full barrier semantics
6277 // since we are doing a lock-free update to _counter.
6278 if (Atomic::xchg(0, &_counter) > 0) return;
6280 Thread* thread = Thread::current();
6281 assert(thread->is_Java_thread(), "Must be JavaThread");
6282 JavaThread *jt = (JavaThread *)thread;
6284 // Optional optimization -- avoid state transitions if there's an interrupt pending.
6285 // Check interrupt before trying to wait
6286 if (Thread::is_interrupted(thread, false)) {
6287 return;
6288 }
6290 // Next, demultiplex/decode time arguments
6291 timespec absTime;
6292 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6293 return;
6294 }
6295 if (time > 0) {
6296 unpackTime(&absTime, isAbsolute, time);
6297 }
6300 // Enter safepoint region
6301 // Beware of deadlocks such as 6317397.
6302 // The per-thread Parker:: mutex is a classic leaf-lock.
6303 // In particular a thread must never block on the Threads_lock while
6304 // holding the Parker:: mutex. If safepoints are pending both the
6305 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6306 ThreadBlockInVM tbivm(jt);
6308 // Don't wait if cannot get lock since interference arises from
6309 // unblocking. Also. check interrupt before trying wait
6310 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6311 return;
6312 }
6314 int status ;
6315 if (_counter > 0) { // no wait needed
6316 _counter = 0;
6317 status = pthread_mutex_unlock(_mutex);
6318 assert (status == 0, "invariant") ;
6319 // Paranoia to ensure our locked and lock-free paths interact
6320 // correctly with each other and Java-level accesses.
6321 OrderAccess::fence();
6322 return;
6323 }
6325 #ifdef ASSERT
6326 // Don't catch signals while blocked; let the running threads have the signals.
6327 // (This allows a debugger to break into the running thread.)
6328 sigset_t oldsigs;
6329 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6330 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6331 #endif
6333 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6334 jt->set_suspend_equivalent();
6335 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6337 assert(_cur_index == -1, "invariant");
6338 if (time == 0) {
6339 _cur_index = REL_INDEX; // arbitrary choice when not timed
6340 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6341 } else {
6342 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6343 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6344 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6345 pthread_cond_destroy (&_cond[_cur_index]) ;
6346 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6347 }
6348 }
6349 _cur_index = -1;
6350 assert_status(status == 0 || status == EINTR ||
6351 status == ETIME || status == ETIMEDOUT,
6352 status, "cond_timedwait");
6354 #ifdef ASSERT
6355 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6356 #endif
6358 _counter = 0 ;
6359 status = pthread_mutex_unlock(_mutex) ;
6360 assert_status(status == 0, status, "invariant") ;
6361 // Paranoia to ensure our locked and lock-free paths interact
6362 // correctly with each other and Java-level accesses.
6363 OrderAccess::fence();
6365 // If externally suspended while waiting, re-suspend
6366 if (jt->handle_special_suspend_equivalent_condition()) {
6367 jt->java_suspend_self();
6368 }
6369 }
6371 void Parker::unpark() {
6372 int s, status ;
6373 status = pthread_mutex_lock(_mutex);
6374 assert (status == 0, "invariant") ;
6375 s = _counter;
6376 _counter = 1;
6377 if (s < 1) {
6378 // thread might be parked
6379 if (_cur_index != -1) {
6380 // thread is definitely parked
6381 if (WorkAroundNPTLTimedWaitHang) {
6382 status = pthread_cond_signal (&_cond[_cur_index]);
6383 assert (status == 0, "invariant");
6384 status = pthread_mutex_unlock(_mutex);
6385 assert (status == 0, "invariant");
6386 } else {
6387 // must capture correct index before unlocking
6388 int index = _cur_index;
6389 status = pthread_mutex_unlock(_mutex);
6390 assert (status == 0, "invariant");
6391 status = pthread_cond_signal (&_cond[index]);
6392 assert (status == 0, "invariant");
6393 }
6394 } else {
6395 pthread_mutex_unlock(_mutex);
6396 assert (status == 0, "invariant") ;
6397 }
6398 } else {
6399 pthread_mutex_unlock(_mutex);
6400 assert (status == 0, "invariant") ;
6401 }
6402 }
6405 extern char** environ;
6407 // Run the specified command in a separate process. Return its exit value,
6408 // or -1 on failure (e.g. can't fork a new process).
6409 // Unlike system(), this function can be called from signal handler. It
6410 // doesn't block SIGINT et al.
6411 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
6412 const char * argv[4] = {"sh", "-c", cmd, NULL};
6414 pid_t pid ;
6416 if (use_vfork_if_available) {
6417 pid = vfork();
6418 } else {
6419 pid = fork();
6420 }
6422 if (pid < 0) {
6423 // fork failed
6424 return -1;
6426 } else if (pid == 0) {
6427 // child process
6429 execve("/bin/sh", (char* const*)argv, environ);
6431 // execve failed
6432 _exit(-1);
6434 } else {
6435 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6436 // care about the actual exit code, for now.
6438 int status;
6440 // Wait for the child process to exit. This returns immediately if
6441 // the child has already exited. */
6442 while (waitpid(pid, &status, 0) < 0) {
6443 switch (errno) {
6444 case ECHILD: return 0;
6445 case EINTR: break;
6446 default: return -1;
6447 }
6448 }
6450 if (WIFEXITED(status)) {
6451 // The child exited normally; get its exit code.
6452 return WEXITSTATUS(status);
6453 } else if (WIFSIGNALED(status)) {
6454 // The child exited because of a signal
6455 // The best value to return is 0x80 + signal number,
6456 // because that is what all Unix shells do, and because
6457 // it allows callers to distinguish between process exit and
6458 // process death by signal.
6459 return 0x80 + WTERMSIG(status);
6460 } else {
6461 // Unknown exit code; pass it through
6462 return status;
6463 }
6464 }
6465 }
6467 // is_headless_jre()
6468 //
6469 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6470 // in order to report if we are running in a headless jre
6471 //
6472 // Since JDK8 xawt/libmawt.so was moved into the same directory
6473 // as libawt.so, and renamed libawt_xawt.so
6474 //
6475 bool os::is_headless_jre() {
6476 struct stat statbuf;
6477 char buf[MAXPATHLEN];
6478 char libmawtpath[MAXPATHLEN];
6479 const char *xawtstr = "/xawt/libmawt.so";
6480 const char *new_xawtstr = "/libawt_xawt.so";
6481 char *p;
6483 // Get path to libjvm.so
6484 os::jvm_path(buf, sizeof(buf));
6486 // Get rid of libjvm.so
6487 p = strrchr(buf, '/');
6488 if (p == NULL) return false;
6489 else *p = '\0';
6491 // Get rid of client or server
6492 p = strrchr(buf, '/');
6493 if (p == NULL) return false;
6494 else *p = '\0';
6496 // check xawt/libmawt.so
6497 strcpy(libmawtpath, buf);
6498 strcat(libmawtpath, xawtstr);
6499 if (::stat(libmawtpath, &statbuf) == 0) return false;
6501 // check libawt_xawt.so
6502 strcpy(libmawtpath, buf);
6503 strcat(libmawtpath, new_xawtstr);
6504 if (::stat(libmawtpath, &statbuf) == 0) return false;
6506 return true;
6507 }
6509 // Get the default path to the core file
6510 // Returns the length of the string
6511 int os::get_core_path(char* buffer, size_t bufferSize) {
6512 const char* p = get_current_directory(buffer, bufferSize);
6514 if (p == NULL) {
6515 assert(p != NULL, "failed to get current directory");
6516 return 0;
6517 }
6519 return strlen(buffer);
6520 }
6522 /////////////// Unit tests ///////////////
6524 #ifndef PRODUCT
6526 #define test_log(...) \
6527 do {\
6528 if (VerboseInternalVMTests) { \
6529 tty->print_cr(__VA_ARGS__); \
6530 tty->flush(); \
6531 }\
6532 } while (false)
6534 class TestReserveMemorySpecial : AllStatic {
6535 public:
6536 static void small_page_write(void* addr, size_t size) {
6537 size_t page_size = os::vm_page_size();
6539 char* end = (char*)addr + size;
6540 for (char* p = (char*)addr; p < end; p += page_size) {
6541 *p = 1;
6542 }
6543 }
6545 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6546 if (!UseHugeTLBFS) {
6547 return;
6548 }
6550 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6552 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6554 if (addr != NULL) {
6555 small_page_write(addr, size);
6557 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6558 }
6559 }
6561 static void test_reserve_memory_special_huge_tlbfs_only() {
6562 if (!UseHugeTLBFS) {
6563 return;
6564 }
6566 size_t lp = os::large_page_size();
6568 for (size_t size = lp; size <= lp * 10; size += lp) {
6569 test_reserve_memory_special_huge_tlbfs_only(size);
6570 }
6571 }
6573 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6574 size_t lp = os::large_page_size();
6575 size_t ag = os::vm_allocation_granularity();
6577 // sizes to test
6578 const size_t sizes[] = {
6579 lp, lp + ag, lp + lp / 2, lp * 2,
6580 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6581 lp * 10, lp * 10 + lp / 2
6582 };
6583 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6585 // For each size/alignment combination, we test three scenarios:
6586 // 1) with req_addr == NULL
6587 // 2) with a non-null req_addr at which we expect to successfully allocate
6588 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6589 // expect the allocation to either fail or to ignore req_addr
6591 // Pre-allocate two areas; they shall be as large as the largest allocation
6592 // and aligned to the largest alignment we will be testing.
6593 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6594 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6595 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6596 -1, 0);
6597 assert(mapping1 != MAP_FAILED, "should work");
6599 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6600 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6601 -1, 0);
6602 assert(mapping2 != MAP_FAILED, "should work");
6604 // Unmap the first mapping, but leave the second mapping intact: the first
6605 // mapping will serve as a value for a "good" req_addr (case 2). The second
6606 // mapping, still intact, as "bad" req_addr (case 3).
6607 ::munmap(mapping1, mapping_size);
6609 // Case 1
6610 test_log("%s, req_addr NULL:", __FUNCTION__);
6611 test_log("size align result");
6613 for (int i = 0; i < num_sizes; i++) {
6614 const size_t size = sizes[i];
6615 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6616 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6617 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6618 size, alignment, p, (p != NULL ? "" : "(failed)"));
6619 if (p != NULL) {
6620 assert(is_ptr_aligned(p, alignment), "must be");
6621 small_page_write(p, size);
6622 os::Linux::release_memory_special_huge_tlbfs(p, size);
6623 }
6624 }
6625 }
6627 // Case 2
6628 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6629 test_log("size align req_addr result");
6631 for (int i = 0; i < num_sizes; i++) {
6632 const size_t size = sizes[i];
6633 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6634 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6635 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6636 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6637 size, alignment, req_addr, p,
6638 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6639 if (p != NULL) {
6640 assert(p == req_addr, "must be");
6641 small_page_write(p, size);
6642 os::Linux::release_memory_special_huge_tlbfs(p, size);
6643 }
6644 }
6645 }
6647 // Case 3
6648 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6649 test_log("size align req_addr result");
6651 for (int i = 0; i < num_sizes; i++) {
6652 const size_t size = sizes[i];
6653 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6654 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6655 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6656 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6657 size, alignment, req_addr, p,
6658 ((p != NULL ? "" : "(failed)")));
6659 // as the area around req_addr contains already existing mappings, the API should always
6660 // return NULL (as per contract, it cannot return another address)
6661 assert(p == NULL, "must be");
6662 }
6663 }
6665 ::munmap(mapping2, mapping_size);
6667 }
6669 static void test_reserve_memory_special_huge_tlbfs() {
6670 if (!UseHugeTLBFS) {
6671 return;
6672 }
6674 test_reserve_memory_special_huge_tlbfs_only();
6675 test_reserve_memory_special_huge_tlbfs_mixed();
6676 }
6678 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6679 if (!UseSHM) {
6680 return;
6681 }
6683 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6685 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6687 if (addr != NULL) {
6688 assert(is_ptr_aligned(addr, alignment), "Check");
6689 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6691 small_page_write(addr, size);
6693 os::Linux::release_memory_special_shm(addr, size);
6694 }
6695 }
6697 static void test_reserve_memory_special_shm() {
6698 size_t lp = os::large_page_size();
6699 size_t ag = os::vm_allocation_granularity();
6701 for (size_t size = ag; size < lp * 3; size += ag) {
6702 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6703 test_reserve_memory_special_shm(size, alignment);
6704 }
6705 }
6706 }
6708 static void test() {
6709 test_reserve_memory_special_huge_tlbfs();
6710 test_reserve_memory_special_shm();
6711 }
6712 };
6714 void TestReserveMemorySpecial_test() {
6715 TestReserveMemorySpecial::test();
6716 }
6718 #endif