Tue, 02 Apr 2019 21:03:37 +0800
#8567 Set_handle_wrong_method_stub is not declared in zero/mips
Reviewed-by: aoqi
1 /*
2 * Copyright (c) 1999, 2018, 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 Mutex* os::Linux::_createThread_lock = NULL;
140 pthread_t os::Linux::_main_thread;
141 int os::Linux::_page_size = -1;
142 const int os::Linux::_vm_default_page_size = (8 * K);
143 bool os::Linux::_is_floating_stack = false;
144 bool os::Linux::_is_NPTL = false;
145 bool os::Linux::_supports_fast_thread_cpu_time = false;
146 const char * os::Linux::_glibc_version = NULL;
147 const char * os::Linux::_libpthread_version = NULL;
148 pthread_condattr_t os::Linux::_condattr[1];
150 static jlong initial_time_count=0;
152 static int clock_tics_per_sec = 100;
154 // For diagnostics to print a message once. see run_periodic_checks
155 static sigset_t check_signal_done;
156 static bool check_signals = true;
158 static pid_t _initial_pid = 0;
160 /* Signal number used to suspend/resume a thread */
162 /* do not use any signal number less than SIGSEGV, see 4355769 */
163 static int SR_signum = SIGUSR2;
164 sigset_t SR_sigset;
166 /* Used to protect dlsym() calls */
167 static pthread_mutex_t dl_mutex;
169 // Declarations
170 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
172 // utility functions
174 static int SR_initialize();
176 julong os::available_memory() {
177 return Linux::available_memory();
178 }
180 julong os::Linux::available_memory() {
181 // values in struct sysinfo are "unsigned long"
182 struct sysinfo si;
183 julong avail_mem;
185 if (OSContainer::is_containerized()) {
186 jlong mem_limit, mem_usage;
187 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
188 if (PrintContainerInfo) {
189 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
190 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
191 }
192 }
194 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
195 if (PrintContainerInfo) {
196 tty->print_cr("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
197 }
198 }
200 if (mem_limit > 0 && mem_usage > 0 ) {
201 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
202 if (PrintContainerInfo) {
203 tty->print_cr("available container memory: " JULONG_FORMAT, avail_mem);
204 }
205 return avail_mem;
206 }
207 }
209 sysinfo(&si);
210 avail_mem = (julong)si.freeram * si.mem_unit;
211 if (Verbose) {
212 tty->print_cr("available memory: " JULONG_FORMAT, avail_mem);
213 }
214 return avail_mem;
215 }
217 julong os::physical_memory() {
218 jlong phys_mem = 0;
219 if (OSContainer::is_containerized()) {
220 jlong mem_limit;
221 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
222 if (PrintContainerInfo) {
223 tty->print_cr("total container memory: " JLONG_FORMAT, mem_limit);
224 }
225 return mem_limit;
226 }
228 if (PrintContainerInfo) {
229 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
230 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
231 }
232 }
234 phys_mem = Linux::physical_memory();
235 if (Verbose) {
236 tty->print_cr("total system memory: " JLONG_FORMAT, phys_mem);
237 }
238 return phys_mem;
239 }
241 ////////////////////////////////////////////////////////////////////////////////
242 // environment support
244 bool os::getenv(const char* name, char* buf, int len) {
245 const char* val = ::getenv(name);
246 if (val != NULL && strlen(val) < (size_t)len) {
247 strcpy(buf, val);
248 return true;
249 }
250 if (len > 0) buf[0] = 0; // return a null string
251 return false;
252 }
255 // Return true if user is running as root.
257 bool os::have_special_privileges() {
258 static bool init = false;
259 static bool privileges = false;
260 if (!init) {
261 privileges = (getuid() != geteuid()) || (getgid() != getegid());
262 init = true;
263 }
264 return privileges;
265 }
268 #ifndef SYS_gettid
269 // i386: 224, ia64: 1105, amd64: 186, sparc 143
270 #ifdef __ia64__
271 #define SYS_gettid 1105
272 #else
273 #ifdef __i386__
274 #define SYS_gettid 224
275 #else
276 #ifdef __amd64__
277 #define SYS_gettid 186
278 #else
279 #ifdef __sparc__
280 #define SYS_gettid 143
281 #else
282 #error define gettid for the arch
283 #endif
284 #endif
285 #endif
286 #endif
287 #endif
289 // Cpu architecture string
290 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
292 // pid_t gettid()
293 //
294 // Returns the kernel thread id of the currently running thread. Kernel
295 // thread id is used to access /proc.
296 //
297 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
298 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
299 //
300 pid_t os::Linux::gettid() {
301 int rslt = syscall(SYS_gettid);
302 if (rslt == -1) {
303 // old kernel, no NPTL support
304 return getpid();
305 } else {
306 return (pid_t)rslt;
307 }
308 }
310 // Most versions of linux have a bug where the number of processors are
311 // determined by looking at the /proc file system. In a chroot environment,
312 // the system call returns 1. This causes the VM to act as if it is
313 // a single processor and elide locking (see is_MP() call).
314 static bool unsafe_chroot_detected = false;
315 static const char *unstable_chroot_error = "/proc file system not found.\n"
316 "Java may be unstable running multithreaded in a chroot "
317 "environment on Linux when /proc filesystem is not mounted.";
319 void os::Linux::initialize_system_info() {
320 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
321 if (processor_count() == 1) {
322 pid_t pid = os::Linux::gettid();
323 char fname[32];
324 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
325 FILE *fp = fopen(fname, "r");
326 if (fp == NULL) {
327 unsafe_chroot_detected = true;
328 } else {
329 fclose(fp);
330 }
331 }
332 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
333 assert(processor_count() > 0, "linux error");
334 }
336 void os::init_system_properties_values() {
337 // The next steps are taken in the product version:
338 //
339 // Obtain the JAVA_HOME value from the location of libjvm.so.
340 // This library should be located at:
341 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
342 //
343 // If "/jre/lib/" appears at the right place in the path, then we
344 // assume libjvm.so is installed in a JDK and we use this path.
345 //
346 // Otherwise exit with message: "Could not create the Java virtual machine."
347 //
348 // The following extra steps are taken in the debugging version:
349 //
350 // If "/jre/lib/" does NOT appear at the right place in the path
351 // instead of exit check for $JAVA_HOME environment variable.
352 //
353 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
354 // then we append a fake suffix "hotspot/libjvm.so" to this path so
355 // it looks like libjvm.so is installed there
356 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
357 //
358 // Otherwise exit.
359 //
360 // Important note: if the location of libjvm.so changes this
361 // code needs to be changed accordingly.
363 // See ld(1):
364 // The linker uses the following search paths to locate required
365 // shared libraries:
366 // 1: ...
367 // ...
368 // 7: The default directories, normally /lib and /usr/lib.
369 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
370 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
371 #else
372 #define DEFAULT_LIBPATH "/lib:/usr/lib"
373 #endif
375 // Base path of extensions installed on the system.
376 #define SYS_EXT_DIR "/usr/java/packages"
377 #define EXTENSIONS_DIR "/lib/ext"
378 #define ENDORSED_DIR "/lib/endorsed"
380 // Buffer that fits several sprintfs.
381 // Note that the space for the colon and the trailing null are provided
382 // by the nulls included by the sizeof operator.
383 const size_t bufsize =
384 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
385 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
386 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
387 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
389 // sysclasspath, java_home, dll_dir
390 {
391 char *pslash;
392 os::jvm_path(buf, bufsize);
394 // Found the full path to libjvm.so.
395 // Now cut the path to <java_home>/jre if we can.
396 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
397 pslash = strrchr(buf, '/');
398 if (pslash != NULL) {
399 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
400 }
401 Arguments::set_dll_dir(buf);
403 if (pslash != NULL) {
404 pslash = strrchr(buf, '/');
405 if (pslash != NULL) {
406 *pslash = '\0'; // Get rid of /<arch>.
407 pslash = strrchr(buf, '/');
408 if (pslash != NULL) {
409 *pslash = '\0'; // Get rid of /lib.
410 }
411 }
412 }
413 Arguments::set_java_home(buf);
414 set_boot_path('/', ':');
415 }
417 // Where to look for native libraries.
418 //
419 // Note: Due to a legacy implementation, most of the library path
420 // is set in the launcher. This was to accomodate linking restrictions
421 // on legacy Linux implementations (which are no longer supported).
422 // Eventually, all the library path setting will be done here.
423 //
424 // However, to prevent the proliferation of improperly built native
425 // libraries, the new path component /usr/java/packages is added here.
426 // Eventually, all the library path setting will be done here.
427 {
428 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
429 // should always exist (until the legacy problem cited above is
430 // addressed).
431 const char *v = ::getenv("LD_LIBRARY_PATH");
432 const char *v_colon = ":";
433 if (v == NULL) { v = ""; v_colon = ""; }
434 // That's +1 for the colon and +1 for the trailing '\0'.
435 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
436 strlen(v) + 1 +
437 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
438 mtInternal);
439 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
440 Arguments::set_library_path(ld_library_path);
441 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
442 }
444 // Extensions directories.
445 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
446 Arguments::set_ext_dirs(buf);
448 // Endorsed standards default directory.
449 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
450 Arguments::set_endorsed_dirs(buf);
452 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
454 #undef DEFAULT_LIBPATH
455 #undef SYS_EXT_DIR
456 #undef EXTENSIONS_DIR
457 #undef ENDORSED_DIR
458 }
460 ////////////////////////////////////////////////////////////////////////////////
461 // breakpoint support
463 void os::breakpoint() {
464 BREAKPOINT;
465 }
467 extern "C" void breakpoint() {
468 // use debugger to set breakpoint here
469 }
471 ////////////////////////////////////////////////////////////////////////////////
472 // signal support
474 debug_only(static bool signal_sets_initialized = false);
475 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
477 bool os::Linux::is_sig_ignored(int sig) {
478 struct sigaction oact;
479 sigaction(sig, (struct sigaction*)NULL, &oact);
480 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
481 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
482 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
483 return true;
484 else
485 return false;
486 }
488 void os::Linux::signal_sets_init() {
489 // Should also have an assertion stating we are still single-threaded.
490 assert(!signal_sets_initialized, "Already initialized");
491 // Fill in signals that are necessarily unblocked for all threads in
492 // the VM. Currently, we unblock the following signals:
493 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
494 // by -Xrs (=ReduceSignalUsage));
495 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
496 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
497 // the dispositions or masks wrt these signals.
498 // Programs embedding the VM that want to use the above signals for their
499 // own purposes must, at this time, use the "-Xrs" option to prevent
500 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
501 // (See bug 4345157, and other related bugs).
502 // In reality, though, unblocking these signals is really a nop, since
503 // these signals are not blocked by default.
504 sigemptyset(&unblocked_sigs);
505 sigemptyset(&allowdebug_blocked_sigs);
506 sigaddset(&unblocked_sigs, SIGILL);
507 sigaddset(&unblocked_sigs, SIGSEGV);
508 sigaddset(&unblocked_sigs, SIGBUS);
509 sigaddset(&unblocked_sigs, SIGFPE);
510 #if defined(PPC64)
511 sigaddset(&unblocked_sigs, SIGTRAP);
512 #endif
513 sigaddset(&unblocked_sigs, SR_signum);
515 if (!ReduceSignalUsage) {
516 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
517 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
518 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
519 }
520 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
521 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
522 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
523 }
524 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
525 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
526 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
527 }
528 }
529 // Fill in signals that are blocked by all but the VM thread.
530 sigemptyset(&vm_sigs);
531 if (!ReduceSignalUsage)
532 sigaddset(&vm_sigs, BREAK_SIGNAL);
533 debug_only(signal_sets_initialized = true);
535 }
537 // These are signals that are unblocked while a thread is running Java.
538 // (For some reason, they get blocked by default.)
539 sigset_t* os::Linux::unblocked_signals() {
540 assert(signal_sets_initialized, "Not initialized");
541 return &unblocked_sigs;
542 }
544 // These are the signals that are blocked while a (non-VM) thread is
545 // running Java. Only the VM thread handles these signals.
546 sigset_t* os::Linux::vm_signals() {
547 assert(signal_sets_initialized, "Not initialized");
548 return &vm_sigs;
549 }
551 // These are signals that are blocked during cond_wait to allow debugger in
552 sigset_t* os::Linux::allowdebug_blocked_signals() {
553 assert(signal_sets_initialized, "Not initialized");
554 return &allowdebug_blocked_sigs;
555 }
557 void os::Linux::hotspot_sigmask(Thread* thread) {
559 //Save caller's signal mask before setting VM signal mask
560 sigset_t caller_sigmask;
561 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
563 OSThread* osthread = thread->osthread();
564 osthread->set_caller_sigmask(caller_sigmask);
566 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
568 if (!ReduceSignalUsage) {
569 if (thread->is_VM_thread()) {
570 // Only the VM thread handles BREAK_SIGNAL ...
571 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
572 } else {
573 // ... all other threads block BREAK_SIGNAL
574 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
575 }
576 }
577 }
579 //////////////////////////////////////////////////////////////////////////////
580 // detecting pthread library
582 void os::Linux::libpthread_init() {
583 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
584 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
585 // generic name for earlier versions.
586 // Define macros here so we can build HotSpot on old systems.
587 # ifndef _CS_GNU_LIBC_VERSION
588 # define _CS_GNU_LIBC_VERSION 2
589 # endif
590 # ifndef _CS_GNU_LIBPTHREAD_VERSION
591 # define _CS_GNU_LIBPTHREAD_VERSION 3
592 # endif
594 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
595 if (n > 0) {
596 char *str = (char *)malloc(n, mtInternal);
597 confstr(_CS_GNU_LIBC_VERSION, str, n);
598 os::Linux::set_glibc_version(str);
599 } else {
600 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
601 static char _gnu_libc_version[32];
602 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
603 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
604 os::Linux::set_glibc_version(_gnu_libc_version);
605 }
607 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
608 if (n > 0) {
609 char *str = (char *)malloc(n, mtInternal);
610 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
611 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
612 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
613 // is the case. LinuxThreads has a hard limit on max number of threads.
614 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
615 // On the other hand, NPTL does not have such a limit, sysconf()
616 // will return -1 and errno is not changed. Check if it is really NPTL.
617 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
618 strstr(str, "NPTL") &&
619 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
620 free(str);
621 os::Linux::set_libpthread_version("linuxthreads");
622 } else {
623 os::Linux::set_libpthread_version(str);
624 }
625 } else {
626 // glibc before 2.3.2 only has LinuxThreads.
627 os::Linux::set_libpthread_version("linuxthreads");
628 }
630 if (strstr(libpthread_version(), "NPTL")) {
631 os::Linux::set_is_NPTL();
632 } else {
633 os::Linux::set_is_LinuxThreads();
634 }
636 // LinuxThreads have two flavors: floating-stack mode, which allows variable
637 // stack size; and fixed-stack mode. NPTL is always floating-stack.
638 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
639 os::Linux::set_is_floating_stack();
640 }
641 }
643 /////////////////////////////////////////////////////////////////////////////
644 // thread stack
646 // Force Linux kernel to expand current thread stack. If "bottom" is close
647 // to the stack guard, caller should block all signals.
648 //
649 // MAP_GROWSDOWN:
650 // A special mmap() flag that is used to implement thread stacks. It tells
651 // kernel that the memory region should extend downwards when needed. This
652 // allows early versions of LinuxThreads to only mmap the first few pages
653 // when creating a new thread. Linux kernel will automatically expand thread
654 // stack as needed (on page faults).
655 //
656 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
657 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
658 // region, it's hard to tell if the fault is due to a legitimate stack
659 // access or because of reading/writing non-exist memory (e.g. buffer
660 // overrun). As a rule, if the fault happens below current stack pointer,
661 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
662 // application (see Linux kernel fault.c).
663 //
664 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
665 // stack overflow detection.
666 //
667 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
668 // not use this flag. However, the stack of initial thread is not created
669 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
670 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
671 // and then attach the thread to JVM.
672 //
673 // To get around the problem and allow stack banging on Linux, we need to
674 // manually expand thread stack after receiving the SIGSEGV.
675 //
676 // There are two ways to expand thread stack to address "bottom", we used
677 // both of them in JVM before 1.5:
678 // 1. adjust stack pointer first so that it is below "bottom", and then
679 // touch "bottom"
680 // 2. mmap() the page in question
681 //
682 // Now alternate signal stack is gone, it's harder to use 2. For instance,
683 // if current sp is already near the lower end of page 101, and we need to
684 // call mmap() to map page 100, it is possible that part of the mmap() frame
685 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
686 // That will destroy the mmap() frame and cause VM to crash.
687 //
688 // The following code works by adjusting sp first, then accessing the "bottom"
689 // page to force a page fault. Linux kernel will then automatically expand the
690 // stack mapping.
691 //
692 // _expand_stack_to() assumes its frame size is less than page size, which
693 // should always be true if the function is not inlined.
695 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
696 #define NOINLINE
697 #else
698 #define NOINLINE __attribute__ ((noinline))
699 #endif
701 static void _expand_stack_to(address bottom) NOINLINE;
703 static void _expand_stack_to(address bottom) {
704 address sp;
705 size_t size;
706 volatile char *p;
708 // Adjust bottom to point to the largest address within the same page, it
709 // gives us a one-page buffer if alloca() allocates slightly more memory.
710 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
711 bottom += os::Linux::page_size() - 1;
713 // sp might be slightly above current stack pointer; if that's the case, we
714 // will alloca() a little more space than necessary, which is OK. Don't use
715 // os::current_stack_pointer(), as its result can be slightly below current
716 // stack pointer, causing us to not alloca enough to reach "bottom".
717 sp = (address)&sp;
719 if (sp > bottom) {
720 size = sp - bottom;
721 p = (volatile char *)alloca(size);
722 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
723 p[0] = '\0';
724 }
725 }
727 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
728 assert(t!=NULL, "just checking");
729 assert(t->osthread()->expanding_stack(), "expand should be set");
730 assert(t->stack_base() != NULL, "stack_base was not initialized");
732 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
733 sigset_t mask_all, old_sigset;
734 sigfillset(&mask_all);
735 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
736 _expand_stack_to(addr);
737 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
738 return true;
739 }
740 return false;
741 }
743 //////////////////////////////////////////////////////////////////////////////
744 // create new thread
746 static address highest_vm_reserved_address();
748 // check if it's safe to start a new thread
749 static bool _thread_safety_check(Thread* thread) {
750 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
751 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
752 // Heap is mmap'ed at lower end of memory space. Thread stacks are
753 // allocated (MAP_FIXED) from high address space. Every thread stack
754 // occupies a fixed size slot (usually 2Mbytes, but user can change
755 // it to other values if they rebuild LinuxThreads).
756 //
757 // Problem with MAP_FIXED is that mmap() can still succeed even part of
758 // the memory region has already been mmap'ed. That means if we have too
759 // many threads and/or very large heap, eventually thread stack will
760 // collide with heap.
761 //
762 // Here we try to prevent heap/stack collision by comparing current
763 // stack bottom with the highest address that has been mmap'ed by JVM
764 // plus a safety margin for memory maps created by native code.
765 //
766 // This feature can be disabled by setting ThreadSafetyMargin to 0
767 //
768 if (ThreadSafetyMargin > 0) {
769 address stack_bottom = os::current_stack_base() - os::current_stack_size();
771 // not safe if our stack extends below the safety margin
772 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
773 } else {
774 return true;
775 }
776 } else {
777 // Floating stack LinuxThreads or NPTL:
778 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
779 // there's not enough space left, pthread_create() will fail. If we come
780 // here, that means enough space has been reserved for stack.
781 return true;
782 }
783 }
785 // Thread start routine for all newly created threads
786 static void *java_start(Thread *thread) {
787 // Try to randomize the cache line index of hot stack frames.
788 // This helps when threads of the same stack traces evict each other's
789 // cache lines. The threads can be either from the same JVM instance, or
790 // from different JVM instances. The benefit is especially true for
791 // processors with hyperthreading technology.
792 static int counter = 0;
793 int pid = os::current_process_id();
794 alloca(((pid ^ counter++) & 7) * 128);
796 ThreadLocalStorage::set_thread(thread);
798 OSThread* osthread = thread->osthread();
799 Monitor* sync = osthread->startThread_lock();
801 // non floating stack LinuxThreads needs extra check, see above
802 if (!_thread_safety_check(thread)) {
803 // notify parent thread
804 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
805 osthread->set_state(ZOMBIE);
806 sync->notify_all();
807 return NULL;
808 }
810 // thread_id is kernel thread id (similar to Solaris LWP id)
811 osthread->set_thread_id(os::Linux::gettid());
813 if (UseNUMA) {
814 int lgrp_id = os::numa_get_group_id();
815 if (lgrp_id != -1) {
816 thread->set_lgrp_id(lgrp_id);
817 }
818 }
819 // initialize signal mask for this thread
820 os::Linux::hotspot_sigmask(thread);
822 // initialize floating point control register
823 os::Linux::init_thread_fpu_state();
825 // handshaking with parent thread
826 {
827 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
829 // notify parent thread
830 osthread->set_state(INITIALIZED);
831 sync->notify_all();
833 // wait until os::start_thread()
834 while (osthread->get_state() == INITIALIZED) {
835 sync->wait(Mutex::_no_safepoint_check_flag);
836 }
837 }
839 // call one more level start routine
840 thread->run();
842 return 0;
843 }
845 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
846 assert(thread->osthread() == NULL, "caller responsible");
848 // Allocate the OSThread object
849 OSThread* osthread = new OSThread(NULL, NULL);
850 if (osthread == NULL) {
851 return false;
852 }
854 // set the correct thread state
855 osthread->set_thread_type(thr_type);
857 // Initial state is ALLOCATED but not INITIALIZED
858 osthread->set_state(ALLOCATED);
860 thread->set_osthread(osthread);
862 // init thread attributes
863 pthread_attr_t attr;
864 pthread_attr_init(&attr);
865 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
867 // stack size
868 if (os::Linux::supports_variable_stack_size()) {
869 // calculate stack size if it's not specified by caller
870 if (stack_size == 0) {
871 stack_size = os::Linux::default_stack_size(thr_type);
873 switch (thr_type) {
874 case os::java_thread:
875 // Java threads use ThreadStackSize which default value can be
876 // changed with the flag -Xss
877 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
878 stack_size = JavaThread::stack_size_at_create();
879 break;
880 case os::compiler_thread:
881 if (CompilerThreadStackSize > 0) {
882 stack_size = (size_t)(CompilerThreadStackSize * K);
883 break;
884 } // else fall through:
885 // use VMThreadStackSize if CompilerThreadStackSize is not defined
886 case os::vm_thread:
887 case os::pgc_thread:
888 case os::cgc_thread:
889 case os::watcher_thread:
890 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
891 break;
892 }
893 }
895 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
896 pthread_attr_setstacksize(&attr, stack_size);
897 } else {
898 // let pthread_create() pick the default value.
899 }
901 // glibc guard page
902 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
904 ThreadState state;
906 {
907 // Serialize thread creation if we are running with fixed stack LinuxThreads
908 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
909 if (lock) {
910 os::Linux::createThread_lock()->lock_without_safepoint_check();
911 }
913 pthread_t tid;
914 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
916 pthread_attr_destroy(&attr);
918 if (ret != 0) {
919 if (PrintMiscellaneous && (Verbose || WizardMode)) {
920 perror("pthread_create()");
921 }
922 // Need to clean up stuff we've allocated so far
923 thread->set_osthread(NULL);
924 delete osthread;
925 if (lock) os::Linux::createThread_lock()->unlock();
926 return false;
927 }
929 // Store pthread info into the OSThread
930 osthread->set_pthread_id(tid);
932 // Wait until child thread is either initialized or aborted
933 {
934 Monitor* sync_with_child = osthread->startThread_lock();
935 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
936 while ((state = osthread->get_state()) == ALLOCATED) {
937 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
938 }
939 }
941 if (lock) {
942 os::Linux::createThread_lock()->unlock();
943 }
944 }
946 // Aborted due to thread limit being reached
947 if (state == ZOMBIE) {
948 thread->set_osthread(NULL);
949 delete osthread;
950 return false;
951 }
953 // The thread is returned suspended (in state INITIALIZED),
954 // and is started higher up in the call chain
955 assert(state == INITIALIZED, "race condition");
956 return true;
957 }
959 /////////////////////////////////////////////////////////////////////////////
960 // attach existing thread
962 // bootstrap the main thread
963 bool os::create_main_thread(JavaThread* thread) {
964 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
965 return create_attached_thread(thread);
966 }
968 bool os::create_attached_thread(JavaThread* thread) {
969 #ifdef ASSERT
970 thread->verify_not_published();
971 #endif
973 // Allocate the OSThread object
974 OSThread* osthread = new OSThread(NULL, NULL);
976 if (osthread == NULL) {
977 return false;
978 }
980 // Store pthread info into the OSThread
981 osthread->set_thread_id(os::Linux::gettid());
982 osthread->set_pthread_id(::pthread_self());
984 // initialize floating point control register
985 os::Linux::init_thread_fpu_state();
987 // Initial thread state is RUNNABLE
988 osthread->set_state(RUNNABLE);
990 thread->set_osthread(osthread);
992 if (UseNUMA) {
993 int lgrp_id = os::numa_get_group_id();
994 if (lgrp_id != -1) {
995 thread->set_lgrp_id(lgrp_id);
996 }
997 }
999 if (os::is_primordial_thread()) {
1000 // If current thread is primordial thread, its stack is mapped on demand,
1001 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1002 // the entire stack region to avoid SEGV in stack banging.
1003 // It is also useful to get around the heap-stack-gap problem on SuSE
1004 // kernel (see 4821821 for details). We first expand stack to the top
1005 // of yellow zone, then enable stack yellow zone (order is significant,
1006 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1007 // is no gap between the last two virtual memory regions.
1009 JavaThread *jt = (JavaThread *)thread;
1010 address addr = jt->stack_yellow_zone_base();
1011 assert(addr != NULL, "initialization problem?");
1012 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1014 osthread->set_expanding_stack();
1015 os::Linux::manually_expand_stack(jt, addr);
1016 osthread->clear_expanding_stack();
1017 }
1019 // initialize signal mask for this thread
1020 // and save the caller's signal mask
1021 os::Linux::hotspot_sigmask(thread);
1023 return true;
1024 }
1026 void os::pd_start_thread(Thread* thread) {
1027 OSThread * osthread = thread->osthread();
1028 assert(osthread->get_state() != INITIALIZED, "just checking");
1029 Monitor* sync_with_child = osthread->startThread_lock();
1030 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1031 sync_with_child->notify();
1033 #if defined MIPS && !defined ZERO
1034 //To be accessed in NativeGeneralJump::patch_verified_entry()
1035 if (thread->is_Java_thread())
1036 {
1037 ((JavaThread*)thread)->set_handle_wrong_method_stub(SharedRuntime::get_handle_wrong_method_stub());
1038 }
1039 #endif
1040 }
1042 // Free Linux resources related to the OSThread
1043 void os::free_thread(OSThread* osthread) {
1044 assert(osthread != NULL, "osthread not set");
1046 if (Thread::current()->osthread() == osthread) {
1047 // Restore caller's signal mask
1048 sigset_t sigmask = osthread->caller_sigmask();
1049 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1050 }
1052 delete osthread;
1053 }
1055 //////////////////////////////////////////////////////////////////////////////
1056 // thread local storage
1058 // Restore the thread pointer if the destructor is called. This is in case
1059 // someone from JNI code sets up a destructor with pthread_key_create to run
1060 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1061 // will hang or crash. When detachCurrentThread is called the key will be set
1062 // to null and we will not be called again. If detachCurrentThread is never
1063 // called we could loop forever depending on the pthread implementation.
1064 static void restore_thread_pointer(void* p) {
1065 Thread* thread = (Thread*) p;
1066 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1067 }
1069 int os::allocate_thread_local_storage() {
1070 pthread_key_t key;
1071 int rslt = pthread_key_create(&key, restore_thread_pointer);
1072 assert(rslt == 0, "cannot allocate thread local storage");
1073 return (int)key;
1074 }
1076 // Note: This is currently not used by VM, as we don't destroy TLS key
1077 // on VM exit.
1078 void os::free_thread_local_storage(int index) {
1079 int rslt = pthread_key_delete((pthread_key_t)index);
1080 assert(rslt == 0, "invalid index");
1081 }
1083 void os::thread_local_storage_at_put(int index, void* value) {
1084 int rslt = pthread_setspecific((pthread_key_t)index, value);
1085 assert(rslt == 0, "pthread_setspecific failed");
1086 }
1088 extern "C" Thread* get_thread() {
1089 return ThreadLocalStorage::thread();
1090 }
1092 //////////////////////////////////////////////////////////////////////////////
1093 // primordial thread
1095 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1096 bool os::is_primordial_thread(void) {
1097 char dummy;
1098 // If called before init complete, thread stack bottom will be null.
1099 // Can be called if fatal error occurs before initialization.
1100 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1101 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1102 os::Linux::initial_thread_stack_size() != 0,
1103 "os::init did not locate primordial thread's stack region");
1104 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1105 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1106 os::Linux::initial_thread_stack_size()) {
1107 return true;
1108 } else {
1109 return false;
1110 }
1111 }
1113 // Find the virtual memory area that contains addr
1114 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1115 FILE *fp = fopen("/proc/self/maps", "r");
1116 if (fp) {
1117 address low, high;
1118 while (!feof(fp)) {
1119 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1120 if (low <= addr && addr < high) {
1121 if (vma_low) *vma_low = low;
1122 if (vma_high) *vma_high = high;
1123 fclose (fp);
1124 return true;
1125 }
1126 }
1127 for (;;) {
1128 int ch = fgetc(fp);
1129 if (ch == EOF || ch == (int)'\n') break;
1130 }
1131 }
1132 fclose(fp);
1133 }
1134 return false;
1135 }
1137 // Locate primordial thread stack. This special handling of primordial thread stack
1138 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1139 // bogus value for the primordial process thread. While the launcher has created
1140 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1141 // JNI invocation API from a primordial thread.
1142 void os::Linux::capture_initial_stack(size_t max_size) {
1144 // max_size is either 0 (which means accept OS default for thread stacks) or
1145 // a user-specified value known to be at least the minimum needed. If we
1146 // are actually on the primordial thread we can make it appear that we have a
1147 // smaller max_size stack by inserting the guard pages at that location. But we
1148 // cannot do anything to emulate a larger stack than what has been provided by
1149 // the OS or threading library. In fact if we try to use a stack greater than
1150 // what is set by rlimit then we will crash the hosting process.
1152 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1153 // If this is "unlimited" then it will be a huge value.
1154 struct rlimit rlim;
1155 getrlimit(RLIMIT_STACK, &rlim);
1156 size_t stack_size = rlim.rlim_cur;
1158 // 6308388: a bug in ld.so will relocate its own .data section to the
1159 // lower end of primordial stack; reduce ulimit -s value a little bit
1160 // so we won't install guard page on ld.so's data section.
1161 // But ensure we don't underflow the stack size - allow 1 page spare
1162 if (stack_size >= (size_t)(3 * page_size())) {
1163 stack_size -= 2 * page_size();
1164 }
1166 // Try to figure out where the stack base (top) is. This is harder.
1167 //
1168 // When an application is started, glibc saves the initial stack pointer in
1169 // a global variable "__libc_stack_end", which is then used by system
1170 // libraries. __libc_stack_end should be pretty close to stack top. The
1171 // variable is available since the very early days. However, because it is
1172 // a private interface, it could disappear in the future.
1173 //
1174 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1175 // to __libc_stack_end, it is very close to stack top, but isn't the real
1176 // stack top. Note that /proc may not exist if VM is running as a chroot
1177 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1178 // /proc/<pid>/stat could change in the future (though unlikely).
1179 //
1180 // We try __libc_stack_end first. If that doesn't work, look for
1181 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1182 // as a hint, which should work well in most cases.
1184 uintptr_t stack_start;
1186 // try __libc_stack_end first
1187 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1188 if (p && *p) {
1189 stack_start = *p;
1190 } else {
1191 // see if we can get the start_stack field from /proc/self/stat
1192 FILE *fp;
1193 int pid;
1194 char state;
1195 int ppid;
1196 int pgrp;
1197 int session;
1198 int nr;
1199 int tpgrp;
1200 unsigned long flags;
1201 unsigned long minflt;
1202 unsigned long cminflt;
1203 unsigned long majflt;
1204 unsigned long cmajflt;
1205 unsigned long utime;
1206 unsigned long stime;
1207 long cutime;
1208 long cstime;
1209 long prio;
1210 long nice;
1211 long junk;
1212 long it_real;
1213 uintptr_t start;
1214 uintptr_t vsize;
1215 intptr_t rss;
1216 uintptr_t rsslim;
1217 uintptr_t scodes;
1218 uintptr_t ecode;
1219 int i;
1221 // Figure what the primordial thread stack base is. Code is inspired
1222 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1223 // followed by command name surrounded by parentheses, state, etc.
1224 char stat[2048];
1225 int statlen;
1227 fp = fopen("/proc/self/stat", "r");
1228 if (fp) {
1229 statlen = fread(stat, 1, 2047, fp);
1230 stat[statlen] = '\0';
1231 fclose(fp);
1233 // Skip pid and the command string. Note that we could be dealing with
1234 // weird command names, e.g. user could decide to rename java launcher
1235 // to "java 1.4.2 :)", then the stat file would look like
1236 // 1234 (java 1.4.2 :)) R ... ...
1237 // We don't really need to know the command string, just find the last
1238 // occurrence of ")" and then start parsing from there. See bug 4726580.
1239 char * s = strrchr(stat, ')');
1241 i = 0;
1242 if (s) {
1243 // Skip blank chars
1244 do s++; while (isspace(*s));
1246 #define _UFM UINTX_FORMAT
1247 #define _DFM INTX_FORMAT
1249 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1250 /* 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 */
1251 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,
1252 &state, /* 3 %c */
1253 &ppid, /* 4 %d */
1254 &pgrp, /* 5 %d */
1255 &session, /* 6 %d */
1256 &nr, /* 7 %d */
1257 &tpgrp, /* 8 %d */
1258 &flags, /* 9 %lu */
1259 &minflt, /* 10 %lu */
1260 &cminflt, /* 11 %lu */
1261 &majflt, /* 12 %lu */
1262 &cmajflt, /* 13 %lu */
1263 &utime, /* 14 %lu */
1264 &stime, /* 15 %lu */
1265 &cutime, /* 16 %ld */
1266 &cstime, /* 17 %ld */
1267 &prio, /* 18 %ld */
1268 &nice, /* 19 %ld */
1269 &junk, /* 20 %ld */
1270 &it_real, /* 21 %ld */
1271 &start, /* 22 UINTX_FORMAT */
1272 &vsize, /* 23 UINTX_FORMAT */
1273 &rss, /* 24 INTX_FORMAT */
1274 &rsslim, /* 25 UINTX_FORMAT */
1275 &scodes, /* 26 UINTX_FORMAT */
1276 &ecode, /* 27 UINTX_FORMAT */
1277 &stack_start); /* 28 UINTX_FORMAT */
1278 }
1280 #undef _UFM
1281 #undef _DFM
1283 if (i != 28 - 2) {
1284 assert(false, "Bad conversion from /proc/self/stat");
1285 // product mode - assume we are the primordial thread, good luck in the
1286 // embedded case.
1287 warning("Can't detect primordial thread stack location - bad conversion");
1288 stack_start = (uintptr_t) &rlim;
1289 }
1290 } else {
1291 // For some reason we can't open /proc/self/stat (for example, running on
1292 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1293 // most cases, so don't abort:
1294 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1295 stack_start = (uintptr_t) &rlim;
1296 }
1297 }
1299 // Now we have a pointer (stack_start) very close to the stack top, the
1300 // next thing to do is to figure out the exact location of stack top. We
1301 // can find out the virtual memory area that contains stack_start by
1302 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1303 // and its upper limit is the real stack top. (again, this would fail if
1304 // running inside chroot, because /proc may not exist.)
1306 uintptr_t stack_top;
1307 address low, high;
1308 if (find_vma((address)stack_start, &low, &high)) {
1309 // success, "high" is the true stack top. (ignore "low", because initial
1310 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1311 stack_top = (uintptr_t)high;
1312 } else {
1313 // failed, likely because /proc/self/maps does not exist
1314 warning("Can't detect primordial thread stack location - find_vma failed");
1315 // best effort: stack_start is normally within a few pages below the real
1316 // stack top, use it as stack top, and reduce stack size so we won't put
1317 // guard page outside stack.
1318 stack_top = stack_start;
1319 stack_size -= 16 * page_size();
1320 }
1322 // stack_top could be partially down the page so align it
1323 stack_top = align_size_up(stack_top, page_size());
1325 // Allowed stack value is minimum of max_size and what we derived from rlimit
1326 if (max_size > 0) {
1327 _initial_thread_stack_size = MIN2(max_size, stack_size);
1328 } else {
1329 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1330 // clamp it at 8MB as we do on Solaris
1331 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1332 }
1334 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1335 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1336 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1337 }
1339 ////////////////////////////////////////////////////////////////////////////////
1340 // time support
1342 // Time since start-up in seconds to a fine granularity.
1343 // Used by VMSelfDestructTimer and the MemProfiler.
1344 double os::elapsedTime() {
1346 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1347 }
1349 jlong os::elapsed_counter() {
1350 return javaTimeNanos() - initial_time_count;
1351 }
1353 jlong os::elapsed_frequency() {
1354 return NANOSECS_PER_SEC; // nanosecond resolution
1355 }
1357 bool os::supports_vtime() { return true; }
1358 bool os::enable_vtime() { return false; }
1359 bool os::vtime_enabled() { return false; }
1361 double os::elapsedVTime() {
1362 struct rusage usage;
1363 int retval = getrusage(RUSAGE_THREAD, &usage);
1364 if (retval == 0) {
1365 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);
1366 } else {
1367 // better than nothing, but not much
1368 return elapsedTime();
1369 }
1370 }
1372 jlong os::javaTimeMillis() {
1373 timeval time;
1374 int status = gettimeofday(&time, NULL);
1375 assert(status != -1, "linux error");
1376 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1377 }
1379 #ifndef CLOCK_MONOTONIC
1380 #define CLOCK_MONOTONIC (1)
1381 #endif
1383 void os::Linux::clock_init() {
1384 // we do dlopen's in this particular order due to bug in linux
1385 // dynamical loader (see 6348968) leading to crash on exit
1386 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1387 if (handle == NULL) {
1388 handle = dlopen("librt.so", RTLD_LAZY);
1389 }
1391 if (handle) {
1392 int (*clock_getres_func)(clockid_t, struct timespec*) =
1393 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1394 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1395 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1396 if (clock_getres_func && clock_gettime_func) {
1397 // See if monotonic clock is supported by the kernel. Note that some
1398 // early implementations simply return kernel jiffies (updated every
1399 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1400 // for nano time (though the monotonic property is still nice to have).
1401 // It's fixed in newer kernels, however clock_getres() still returns
1402 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1403 // resolution for now. Hopefully as people move to new kernels, this
1404 // won't be a problem.
1405 struct timespec res;
1406 struct timespec tp;
1407 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1408 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1409 // yes, monotonic clock is supported
1410 _clock_gettime = clock_gettime_func;
1411 return;
1412 } else {
1413 // close librt if there is no monotonic clock
1414 dlclose(handle);
1415 }
1416 }
1417 }
1418 warning("No monotonic clock was available - timed services may " \
1419 "be adversely affected if the time-of-day clock changes");
1420 }
1422 #ifndef SYS_clock_getres
1424 #if defined(IA32) || defined(AMD64)
1425 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1426 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1427 #else
1428 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1429 #define sys_clock_getres(x,y) -1
1430 #endif
1432 #else
1433 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1434 #endif
1436 void os::Linux::fast_thread_clock_init() {
1437 if (!UseLinuxPosixThreadCPUClocks) {
1438 return;
1439 }
1440 clockid_t clockid;
1441 struct timespec tp;
1442 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1443 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1445 // Switch to using fast clocks for thread cpu time if
1446 // the sys_clock_getres() returns 0 error code.
1447 // Note, that some kernels may support the current thread
1448 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1449 // returned by the pthread_getcpuclockid().
1450 // If the fast Posix clocks are supported then the sys_clock_getres()
1451 // must return at least tp.tv_sec == 0 which means a resolution
1452 // better than 1 sec. This is extra check for reliability.
1454 if(pthread_getcpuclockid_func &&
1455 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1456 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1458 _supports_fast_thread_cpu_time = true;
1459 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1460 }
1461 }
1463 jlong os::javaTimeNanos() {
1464 if (Linux::supports_monotonic_clock()) {
1465 struct timespec tp;
1466 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1467 assert(status == 0, "gettime error");
1468 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1469 return result;
1470 } else {
1471 timeval time;
1472 int status = gettimeofday(&time, NULL);
1473 assert(status != -1, "linux error");
1474 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1475 return 1000 * usecs;
1476 }
1477 }
1479 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1480 if (Linux::supports_monotonic_clock()) {
1481 info_ptr->max_value = ALL_64_BITS;
1483 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1484 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1485 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1486 } else {
1487 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1488 info_ptr->max_value = ALL_64_BITS;
1490 // gettimeofday is a real time clock so it skips
1491 info_ptr->may_skip_backward = true;
1492 info_ptr->may_skip_forward = true;
1493 }
1495 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1496 }
1498 // Return the real, user, and system times in seconds from an
1499 // arbitrary fixed point in the past.
1500 bool os::getTimesSecs(double* process_real_time,
1501 double* process_user_time,
1502 double* process_system_time) {
1503 struct tms ticks;
1504 clock_t real_ticks = times(&ticks);
1506 if (real_ticks == (clock_t) (-1)) {
1507 return false;
1508 } else {
1509 double ticks_per_second = (double) clock_tics_per_sec;
1510 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1511 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1512 *process_real_time = ((double) real_ticks) / ticks_per_second;
1514 return true;
1515 }
1516 }
1519 char * os::local_time_string(char *buf, size_t buflen) {
1520 struct tm t;
1521 time_t long_time;
1522 time(&long_time);
1523 localtime_r(&long_time, &t);
1524 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1525 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1526 t.tm_hour, t.tm_min, t.tm_sec);
1527 return buf;
1528 }
1530 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1531 return localtime_r(clock, res);
1532 }
1534 ////////////////////////////////////////////////////////////////////////////////
1535 // runtime exit support
1537 // Note: os::shutdown() might be called very early during initialization, or
1538 // called from signal handler. Before adding something to os::shutdown(), make
1539 // sure it is async-safe and can handle partially initialized VM.
1540 void os::shutdown() {
1542 // allow PerfMemory to attempt cleanup of any persistent resources
1543 perfMemory_exit();
1545 // needs to remove object in file system
1546 AttachListener::abort();
1548 // flush buffered output, finish log files
1549 ostream_abort();
1551 // Check for abort hook
1552 abort_hook_t abort_hook = Arguments::abort_hook();
1553 if (abort_hook != NULL) {
1554 abort_hook();
1555 }
1557 }
1559 // Note: os::abort() might be called very early during initialization, or
1560 // called from signal handler. Before adding something to os::abort(), make
1561 // sure it is async-safe and can handle partially initialized VM.
1562 void os::abort(bool dump_core) {
1563 os::shutdown();
1564 if (dump_core) {
1565 #ifndef PRODUCT
1566 fdStream out(defaultStream::output_fd());
1567 out.print_raw("Current thread is ");
1568 char buf[16];
1569 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1570 out.print_raw_cr(buf);
1571 out.print_raw_cr("Dumping core ...");
1572 #endif
1573 ::abort(); // dump core
1574 }
1576 ::exit(1);
1577 }
1579 // Die immediately, no exit hook, no abort hook, no cleanup.
1580 void os::die() {
1581 // _exit() on LinuxThreads only kills current thread
1582 ::abort();
1583 }
1586 // This method is a copy of JDK's sysGetLastErrorString
1587 // from src/solaris/hpi/src/system_md.c
1589 size_t os::lasterror(char *buf, size_t len) {
1591 if (errno == 0) return 0;
1593 const char *s = ::strerror(errno);
1594 size_t n = ::strlen(s);
1595 if (n >= len) {
1596 n = len - 1;
1597 }
1598 ::strncpy(buf, s, n);
1599 buf[n] = '\0';
1600 return n;
1601 }
1603 intx os::current_thread_id() { return (intx)pthread_self(); }
1604 int os::current_process_id() {
1606 // Under the old linux thread library, linux gives each thread
1607 // its own process id. Because of this each thread will return
1608 // a different pid if this method were to return the result
1609 // of getpid(2). Linux provides no api that returns the pid
1610 // of the launcher thread for the vm. This implementation
1611 // returns a unique pid, the pid of the launcher thread
1612 // that starts the vm 'process'.
1614 // Under the NPTL, getpid() returns the same pid as the
1615 // launcher thread rather than a unique pid per thread.
1616 // Use gettid() if you want the old pre NPTL behaviour.
1618 // if you are looking for the result of a call to getpid() that
1619 // returns a unique pid for the calling thread, then look at the
1620 // OSThread::thread_id() method in osThread_linux.hpp file
1622 return (int)(_initial_pid ? _initial_pid : getpid());
1623 }
1625 // DLL functions
1627 const char* os::dll_file_extension() { return ".so"; }
1629 // This must be hard coded because it's the system's temporary
1630 // directory not the java application's temp directory, ala java.io.tmpdir.
1631 const char* os::get_temp_directory() { return "/tmp"; }
1633 static bool file_exists(const char* filename) {
1634 struct stat statbuf;
1635 if (filename == NULL || strlen(filename) == 0) {
1636 return false;
1637 }
1638 return os::stat(filename, &statbuf) == 0;
1639 }
1641 bool os::dll_build_name(char* buffer, size_t buflen,
1642 const char* pname, const char* fname) {
1643 bool retval = false;
1644 // Copied from libhpi
1645 const size_t pnamelen = pname ? strlen(pname) : 0;
1647 // Return error on buffer overflow.
1648 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1649 return retval;
1650 }
1652 if (pnamelen == 0) {
1653 snprintf(buffer, buflen, "lib%s.so", fname);
1654 retval = true;
1655 } else if (strchr(pname, *os::path_separator()) != NULL) {
1656 int n;
1657 char** pelements = split_path(pname, &n);
1658 if (pelements == NULL) {
1659 return false;
1660 }
1661 for (int i = 0 ; i < n ; i++) {
1662 // Really shouldn't be NULL, but check can't hurt
1663 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1664 continue; // skip the empty path values
1665 }
1666 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1667 if (file_exists(buffer)) {
1668 retval = true;
1669 break;
1670 }
1671 }
1672 // release the storage
1673 for (int i = 0 ; i < n ; i++) {
1674 if (pelements[i] != NULL) {
1675 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1676 }
1677 }
1678 if (pelements != NULL) {
1679 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1680 }
1681 } else {
1682 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1683 retval = true;
1684 }
1685 return retval;
1686 }
1688 // check if addr is inside libjvm.so
1689 bool os::address_is_in_vm(address addr) {
1690 static address libjvm_base_addr;
1691 Dl_info dlinfo;
1693 if (libjvm_base_addr == NULL) {
1694 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1695 libjvm_base_addr = (address)dlinfo.dli_fbase;
1696 }
1697 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1698 }
1700 if (dladdr((void *)addr, &dlinfo) != 0) {
1701 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1702 }
1704 return false;
1705 }
1707 bool os::dll_address_to_function_name(address addr, char *buf,
1708 int buflen, int *offset) {
1709 // buf is not optional, but offset is optional
1710 assert(buf != NULL, "sanity check");
1712 Dl_info dlinfo;
1714 if (dladdr((void*)addr, &dlinfo) != 0) {
1715 // see if we have a matching symbol
1716 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1717 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1718 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1719 }
1720 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1721 return true;
1722 }
1723 // no matching symbol so try for just file info
1724 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1725 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1726 buf, buflen, offset, dlinfo.dli_fname)) {
1727 return true;
1728 }
1729 }
1730 }
1732 buf[0] = '\0';
1733 if (offset != NULL) *offset = -1;
1734 return false;
1735 }
1737 struct _address_to_library_name {
1738 address addr; // input : memory address
1739 size_t buflen; // size of fname
1740 char* fname; // output: library name
1741 address base; // library base addr
1742 };
1744 static int address_to_library_name_callback(struct dl_phdr_info *info,
1745 size_t size, void *data) {
1746 int i;
1747 bool found = false;
1748 address libbase = NULL;
1749 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1751 // iterate through all loadable segments
1752 for (i = 0; i < info->dlpi_phnum; i++) {
1753 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1754 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1755 // base address of a library is the lowest address of its loaded
1756 // segments.
1757 if (libbase == NULL || libbase > segbase) {
1758 libbase = segbase;
1759 }
1760 // see if 'addr' is within current segment
1761 if (segbase <= d->addr &&
1762 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1763 found = true;
1764 }
1765 }
1766 }
1768 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1769 // so dll_address_to_library_name() can fall through to use dladdr() which
1770 // can figure out executable name from argv[0].
1771 if (found && info->dlpi_name && info->dlpi_name[0]) {
1772 d->base = libbase;
1773 if (d->fname) {
1774 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1775 }
1776 return 1;
1777 }
1778 return 0;
1779 }
1781 bool os::dll_address_to_library_name(address addr, char* buf,
1782 int buflen, int* offset) {
1783 // buf is not optional, but offset is optional
1784 assert(buf != NULL, "sanity check");
1786 Dl_info dlinfo;
1787 struct _address_to_library_name data;
1789 // There is a bug in old glibc dladdr() implementation that it could resolve
1790 // to wrong library name if the .so file has a base address != NULL. Here
1791 // we iterate through the program headers of all loaded libraries to find
1792 // out which library 'addr' really belongs to. This workaround can be
1793 // removed once the minimum requirement for glibc is moved to 2.3.x.
1794 data.addr = addr;
1795 data.fname = buf;
1796 data.buflen = buflen;
1797 data.base = NULL;
1798 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1800 if (rslt) {
1801 // buf already contains library name
1802 if (offset) *offset = addr - data.base;
1803 return true;
1804 }
1805 if (dladdr((void*)addr, &dlinfo) != 0) {
1806 if (dlinfo.dli_fname != NULL) {
1807 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1808 }
1809 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1810 *offset = addr - (address)dlinfo.dli_fbase;
1811 }
1812 return true;
1813 }
1815 buf[0] = '\0';
1816 if (offset) *offset = -1;
1817 return false;
1818 }
1820 // Loads .dll/.so and
1821 // in case of error it checks if .dll/.so was built for the
1822 // same architecture as Hotspot is running on
1825 // Remember the stack's state. The Linux dynamic linker will change
1826 // the stack to 'executable' at most once, so we must safepoint only once.
1827 bool os::Linux::_stack_is_executable = false;
1829 // VM operation that loads a library. This is necessary if stack protection
1830 // of the Java stacks can be lost during loading the library. If we
1831 // do not stop the Java threads, they can stack overflow before the stacks
1832 // are protected again.
1833 class VM_LinuxDllLoad: public VM_Operation {
1834 private:
1835 const char *_filename;
1836 char *_ebuf;
1837 int _ebuflen;
1838 void *_lib;
1839 public:
1840 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1841 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1842 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1843 void doit() {
1844 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1845 os::Linux::_stack_is_executable = true;
1846 }
1847 void* loaded_library() { return _lib; }
1848 };
1850 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1851 {
1852 void * result = NULL;
1853 bool load_attempted = false;
1855 // Check whether the library to load might change execution rights
1856 // of the stack. If they are changed, the protection of the stack
1857 // guard pages will be lost. We need a safepoint to fix this.
1858 //
1859 // See Linux man page execstack(8) for more info.
1860 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1861 ElfFile ef(filename);
1862 if (!ef.specifies_noexecstack()) {
1863 if (!is_init_completed()) {
1864 os::Linux::_stack_is_executable = true;
1865 // This is OK - No Java threads have been created yet, and hence no
1866 // stack guard pages to fix.
1867 //
1868 // This should happen only when you are building JDK7 using a very
1869 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1870 //
1871 // Dynamic loader will make all stacks executable after
1872 // this function returns, and will not do that again.
1873 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1874 } else {
1875 warning("You have loaded library %s which might have disabled stack guard. "
1876 "The VM will try to fix the stack guard now.\n"
1877 "It's highly recommended that you fix the library with "
1878 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1879 filename);
1881 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1882 JavaThread *jt = JavaThread::current();
1883 if (jt->thread_state() != _thread_in_native) {
1884 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1885 // that requires ExecStack. Cannot enter safe point. Let's give up.
1886 warning("Unable to fix stack guard. Giving up.");
1887 } else {
1888 if (!LoadExecStackDllInVMThread) {
1889 // This is for the case where the DLL has an static
1890 // constructor function that executes JNI code. We cannot
1891 // load such DLLs in the VMThread.
1892 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1893 }
1895 ThreadInVMfromNative tiv(jt);
1896 debug_only(VMNativeEntryWrapper vew;)
1898 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1899 VMThread::execute(&op);
1900 if (LoadExecStackDllInVMThread) {
1901 result = op.loaded_library();
1902 }
1903 load_attempted = true;
1904 }
1905 }
1906 }
1907 }
1909 if (!load_attempted) {
1910 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1911 }
1913 if (result != NULL) {
1914 // Successful loading
1915 return result;
1916 }
1918 Elf32_Ehdr elf_head;
1919 int diag_msg_max_length=ebuflen-strlen(ebuf);
1920 char* diag_msg_buf=ebuf+strlen(ebuf);
1922 if (diag_msg_max_length==0) {
1923 // No more space in ebuf for additional diagnostics message
1924 return NULL;
1925 }
1928 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1930 if (file_descriptor < 0) {
1931 // Can't open library, report dlerror() message
1932 return NULL;
1933 }
1935 bool failed_to_read_elf_head=
1936 (sizeof(elf_head)!=
1937 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1939 ::close(file_descriptor);
1940 if (failed_to_read_elf_head) {
1941 // file i/o error - report dlerror() msg
1942 return NULL;
1943 }
1945 typedef struct {
1946 Elf32_Half code; // Actual value as defined in elf.h
1947 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1948 char elf_class; // 32 or 64 bit
1949 char endianess; // MSB or LSB
1950 char* name; // String representation
1951 } arch_t;
1953 #ifndef EM_486
1954 #define EM_486 6 /* Intel 80486 */
1955 #endif
1957 static const arch_t arch_array[]={
1958 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1959 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1960 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1961 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1962 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1963 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1964 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1965 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1966 #if defined(VM_LITTLE_ENDIAN)
1967 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1968 #else
1969 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1970 #endif
1971 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1972 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1973 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1974 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1975 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1976 {EM_MIPS, EM_MIPS, ELFCLASS64, ELFDATA2LSB, (char*)"MIPS64 LE"},
1977 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1978 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1979 };
1981 #if (defined IA32)
1982 static Elf32_Half running_arch_code=EM_386;
1983 #elif (defined AMD64)
1984 static Elf32_Half running_arch_code=EM_X86_64;
1985 #elif (defined IA64)
1986 static Elf32_Half running_arch_code=EM_IA_64;
1987 #elif (defined __sparc) && (defined _LP64)
1988 static Elf32_Half running_arch_code=EM_SPARCV9;
1989 #elif (defined __sparc) && (!defined _LP64)
1990 static Elf32_Half running_arch_code=EM_SPARC;
1991 #elif (defined MIPS64)
1992 static Elf32_Half running_arch_code=EM_MIPS;
1993 #elif (defined __powerpc64__)
1994 static Elf32_Half running_arch_code=EM_PPC64;
1995 #elif (defined __powerpc__)
1996 static Elf32_Half running_arch_code=EM_PPC;
1997 #elif (defined ARM)
1998 static Elf32_Half running_arch_code=EM_ARM;
1999 #elif (defined S390)
2000 static Elf32_Half running_arch_code=EM_S390;
2001 #elif (defined ALPHA)
2002 static Elf32_Half running_arch_code=EM_ALPHA;
2003 #elif (defined MIPSEL)
2004 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2005 #elif (defined PARISC)
2006 static Elf32_Half running_arch_code=EM_PARISC;
2007 #elif (defined MIPS)
2008 static Elf32_Half running_arch_code=EM_MIPS;
2009 #elif (defined M68K)
2010 static Elf32_Half running_arch_code=EM_68K;
2011 #else
2012 #error Method os::dll_load requires that one of following is defined:\
2013 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, __mips64, PARISC, M68K
2014 #endif
2016 // Identify compatability class for VM's architecture and library's architecture
2017 // Obtain string descriptions for architectures
2019 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2020 int running_arch_index=-1;
2022 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2023 if (running_arch_code == arch_array[i].code) {
2024 running_arch_index = i;
2025 }
2026 if (lib_arch.code == arch_array[i].code) {
2027 lib_arch.compat_class = arch_array[i].compat_class;
2028 lib_arch.name = arch_array[i].name;
2029 }
2030 }
2032 assert(running_arch_index != -1,
2033 "Didn't find running architecture code (running_arch_code) in arch_array");
2034 if (running_arch_index == -1) {
2035 // Even though running architecture detection failed
2036 // we may still continue with reporting dlerror() message
2037 return NULL;
2038 }
2040 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2041 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2042 return NULL;
2043 }
2045 #ifndef S390
2046 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2047 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2048 return NULL;
2049 }
2050 #endif // !S390
2052 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2053 if ( lib_arch.name!=NULL ) {
2054 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2055 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2056 lib_arch.name, arch_array[running_arch_index].name);
2057 } else {
2058 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2059 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2060 lib_arch.code,
2061 arch_array[running_arch_index].name);
2062 }
2063 }
2065 return NULL;
2066 }
2068 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2069 void * result = ::dlopen(filename, RTLD_LAZY);
2070 if (result == NULL) {
2071 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2072 ebuf[ebuflen-1] = '\0';
2073 }
2074 return result;
2075 }
2077 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2078 void * result = NULL;
2079 if (LoadExecStackDllInVMThread) {
2080 result = dlopen_helper(filename, ebuf, ebuflen);
2081 }
2083 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2084 // library that requires an executable stack, or which does not have this
2085 // stack attribute set, dlopen changes the stack attribute to executable. The
2086 // read protection of the guard pages gets lost.
2087 //
2088 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2089 // may have been queued at the same time.
2091 if (!_stack_is_executable) {
2092 JavaThread *jt = Threads::first();
2094 while (jt) {
2095 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2096 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2097 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2098 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2099 warning("Attempt to reguard stack yellow zone failed.");
2100 }
2101 }
2102 jt = jt->next();
2103 }
2104 }
2106 return result;
2107 }
2109 /*
2110 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2111 * chances are you might want to run the generated bits against glibc-2.0
2112 * libdl.so, so always use locking for any version of glibc.
2113 */
2114 void* os::dll_lookup(void* handle, const char* name) {
2115 pthread_mutex_lock(&dl_mutex);
2116 void* res = dlsym(handle, name);
2117 pthread_mutex_unlock(&dl_mutex);
2118 return res;
2119 }
2121 void* os::get_default_process_handle() {
2122 return (void*)::dlopen(NULL, RTLD_LAZY);
2123 }
2125 static bool _print_ascii_file(const char* filename, outputStream* st) {
2126 int fd = ::open(filename, O_RDONLY);
2127 if (fd == -1) {
2128 return false;
2129 }
2131 char buf[32];
2132 int bytes;
2133 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2134 st->print_raw(buf, bytes);
2135 }
2137 ::close(fd);
2139 return true;
2140 }
2142 void os::print_dll_info(outputStream *st) {
2143 st->print_cr("Dynamic libraries:");
2145 char fname[32];
2146 pid_t pid = os::Linux::gettid();
2148 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2150 if (!_print_ascii_file(fname, st)) {
2151 st->print("Can not get library information for pid = %d\n", pid);
2152 }
2153 }
2155 void os::print_os_info_brief(outputStream* st) {
2156 os::Linux::print_distro_info(st);
2158 os::Posix::print_uname_info(st);
2160 os::Linux::print_libversion_info(st);
2162 }
2164 void os::print_os_info(outputStream* st) {
2165 st->print("OS:");
2167 os::Linux::print_distro_info(st);
2169 os::Posix::print_uname_info(st);
2171 // Print warning if unsafe chroot environment detected
2172 if (unsafe_chroot_detected) {
2173 st->print("WARNING!! ");
2174 st->print_cr("%s", unstable_chroot_error);
2175 }
2177 os::Linux::print_libversion_info(st);
2179 os::Posix::print_rlimit_info(st);
2181 os::Posix::print_load_average(st);
2183 os::Linux::print_full_memory_info(st);
2185 os::Linux::print_container_info(st);
2186 }
2188 // Try to identify popular distros.
2189 // Most Linux distributions have a /etc/XXX-release file, which contains
2190 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2191 // file that also contains the OS version string. Some have more than one
2192 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2193 // /etc/redhat-release.), so the order is important.
2194 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2195 // their own specific XXX-release file as well as a redhat-release file.
2196 // Because of this the XXX-release file needs to be searched for before the
2197 // redhat-release file.
2198 // Since Red Hat has a lsb-release file that is not very descriptive the
2199 // search for redhat-release needs to be before lsb-release.
2200 // Since the lsb-release file is the new standard it needs to be searched
2201 // before the older style release files.
2202 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2203 // next to last resort. The os-release file is a new standard that contains
2204 // distribution information and the system-release file seems to be an old
2205 // standard that has been replaced by the lsb-release and os-release files.
2206 // Searching for the debian_version file is the last resort. It contains
2207 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2208 // "Debian " is printed before the contents of the debian_version file.
2209 void os::Linux::print_distro_info(outputStream* st) {
2210 if (!_print_ascii_file("/etc/oracle-release", st) &&
2211 !_print_ascii_file("/etc/mandriva-release", st) &&
2212 !_print_ascii_file("/etc/mandrake-release", st) &&
2213 !_print_ascii_file("/etc/sun-release", st) &&
2214 !_print_ascii_file("/etc/redhat-release", st) &&
2215 !_print_ascii_file("/etc/lsb-release", st) &&
2216 !_print_ascii_file("/etc/SuSE-release", st) &&
2217 !_print_ascii_file("/etc/turbolinux-release", st) &&
2218 !_print_ascii_file("/etc/gentoo-release", st) &&
2219 !_print_ascii_file("/etc/ltib-release", st) &&
2220 !_print_ascii_file("/etc/angstrom-version", st) &&
2221 !_print_ascii_file("/etc/system-release", st) &&
2222 !_print_ascii_file("/etc/os-release", st)) {
2224 if (file_exists("/etc/debian_version")) {
2225 st->print("Debian ");
2226 _print_ascii_file("/etc/debian_version", st);
2227 } else {
2228 st->print("Linux");
2229 }
2230 }
2231 st->cr();
2232 }
2234 void os::Linux::print_libversion_info(outputStream* st) {
2235 // libc, pthread
2236 st->print("libc:");
2237 st->print("%s ", os::Linux::glibc_version());
2238 st->print("%s ", os::Linux::libpthread_version());
2239 if (os::Linux::is_LinuxThreads()) {
2240 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2241 }
2242 st->cr();
2243 }
2245 void os::Linux::print_full_memory_info(outputStream* st) {
2246 st->print("\n/proc/meminfo:\n");
2247 _print_ascii_file("/proc/meminfo", st);
2248 st->cr();
2249 }
2251 void os::Linux::print_container_info(outputStream* st) {
2252 if (!OSContainer::is_containerized()) {
2253 return;
2254 }
2256 st->print("container (cgroup) information:\n");
2258 const char *p_ct = OSContainer::container_type();
2259 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2261 char *p = OSContainer::cpu_cpuset_cpus();
2262 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2263 free(p);
2265 p = OSContainer::cpu_cpuset_memory_nodes();
2266 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2267 free(p);
2269 int i = OSContainer::active_processor_count();
2270 if (i > 0) {
2271 st->print("active_processor_count: %d\n", i);
2272 } else {
2273 st->print("active_processor_count: failed\n");
2274 }
2276 i = OSContainer::cpu_quota();
2277 st->print("cpu_quota: %d\n", i);
2279 i = OSContainer::cpu_period();
2280 st->print("cpu_period: %d\n", i);
2282 i = OSContainer::cpu_shares();
2283 st->print("cpu_shares: %d\n", i);
2285 jlong j = OSContainer::memory_limit_in_bytes();
2286 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2288 j = OSContainer::memory_and_swap_limit_in_bytes();
2289 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2291 j = OSContainer::memory_soft_limit_in_bytes();
2292 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2294 j = OSContainer::OSContainer::memory_usage_in_bytes();
2295 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2297 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2298 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2299 st->cr();
2300 }
2302 void os::print_memory_info(outputStream* st) {
2304 st->print("Memory:");
2305 st->print(" %dk page", os::vm_page_size()>>10);
2307 // values in struct sysinfo are "unsigned long"
2308 struct sysinfo si;
2309 sysinfo(&si);
2311 st->print(", physical " UINT64_FORMAT "k",
2312 os::physical_memory() >> 10);
2313 st->print("(" UINT64_FORMAT "k free)",
2314 os::available_memory() >> 10);
2315 st->print(", swap " UINT64_FORMAT "k",
2316 ((jlong)si.totalswap * si.mem_unit) >> 10);
2317 st->print("(" UINT64_FORMAT "k free)",
2318 ((jlong)si.freeswap * si.mem_unit) >> 10);
2319 st->cr();
2320 }
2322 void os::pd_print_cpu_info(outputStream* st) {
2323 st->print("\n/proc/cpuinfo:\n");
2324 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2325 st->print(" <Not Available>");
2326 }
2327 st->cr();
2328 }
2330 void os::print_siginfo(outputStream* st, void* siginfo) {
2331 const siginfo_t* si = (const siginfo_t*)siginfo;
2333 os::Posix::print_siginfo_brief(st, si);
2334 #if INCLUDE_CDS
2335 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2336 UseSharedSpaces) {
2337 FileMapInfo* mapinfo = FileMapInfo::current_info();
2338 if (mapinfo->is_in_shared_space(si->si_addr)) {
2339 st->print("\n\nError accessing class data sharing archive." \
2340 " Mapped file inaccessible during execution, " \
2341 " possible disk/network problem.");
2342 }
2343 }
2344 #endif
2345 st->cr();
2346 }
2349 static void print_signal_handler(outputStream* st, int sig,
2350 char* buf, size_t buflen);
2352 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2353 st->print_cr("Signal Handlers:");
2354 print_signal_handler(st, SIGSEGV, buf, buflen);
2355 print_signal_handler(st, SIGBUS , buf, buflen);
2356 print_signal_handler(st, SIGFPE , buf, buflen);
2357 print_signal_handler(st, SIGPIPE, buf, buflen);
2358 print_signal_handler(st, SIGXFSZ, buf, buflen);
2359 print_signal_handler(st, SIGILL , buf, buflen);
2360 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2361 print_signal_handler(st, SR_signum, buf, buflen);
2362 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2363 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2364 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2365 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2366 #if defined(PPC64)
2367 print_signal_handler(st, SIGTRAP, buf, buflen);
2368 #endif
2369 }
2371 static char saved_jvm_path[MAXPATHLEN] = {0};
2373 // Find the full path to the current module, libjvm.so
2374 void os::jvm_path(char *buf, jint buflen) {
2375 // Error checking.
2376 if (buflen < MAXPATHLEN) {
2377 assert(false, "must use a large-enough buffer");
2378 buf[0] = '\0';
2379 return;
2380 }
2381 // Lazy resolve the path to current module.
2382 if (saved_jvm_path[0] != 0) {
2383 strcpy(buf, saved_jvm_path);
2384 return;
2385 }
2387 char dli_fname[MAXPATHLEN];
2388 bool ret = dll_address_to_library_name(
2389 CAST_FROM_FN_PTR(address, os::jvm_path),
2390 dli_fname, sizeof(dli_fname), NULL);
2391 assert(ret, "cannot locate libjvm");
2392 char *rp = NULL;
2393 if (ret && dli_fname[0] != '\0') {
2394 rp = realpath(dli_fname, buf);
2395 }
2396 if (rp == NULL)
2397 return;
2399 if (Arguments::created_by_gamma_launcher()) {
2400 // Support for the gamma launcher. Typical value for buf is
2401 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2402 // the right place in the string, then assume we are installed in a JDK and
2403 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2404 // up the path so it looks like libjvm.so is installed there (append a
2405 // fake suffix hotspot/libjvm.so).
2406 const char *p = buf + strlen(buf) - 1;
2407 for (int count = 0; p > buf && count < 5; ++count) {
2408 for (--p; p > buf && *p != '/'; --p)
2409 /* empty */ ;
2410 }
2412 if (strncmp(p, "/jre/lib/", 9) != 0) {
2413 // Look for JAVA_HOME in the environment.
2414 char* java_home_var = ::getenv("JAVA_HOME");
2415 if (java_home_var != NULL && java_home_var[0] != 0) {
2416 char* jrelib_p;
2417 int len;
2419 // Check the current module name "libjvm.so".
2420 p = strrchr(buf, '/');
2421 assert(strstr(p, "/libjvm") == p, "invalid library name");
2423 rp = realpath(java_home_var, buf);
2424 if (rp == NULL)
2425 return;
2427 // determine if this is a legacy image or modules image
2428 // modules image doesn't have "jre" subdirectory
2429 len = strlen(buf);
2430 assert(len < buflen, "Ran out of buffer room");
2431 jrelib_p = buf + len;
2432 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2433 if (0 != access(buf, F_OK)) {
2434 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2435 }
2437 if (0 == access(buf, F_OK)) {
2438 // Use current module name "libjvm.so"
2439 len = strlen(buf);
2440 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2441 } else {
2442 // Go back to path of .so
2443 rp = realpath(dli_fname, buf);
2444 if (rp == NULL)
2445 return;
2446 }
2447 }
2448 }
2449 }
2451 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2452 }
2454 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2455 // no prefix required, not even "_"
2456 }
2458 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2459 // no suffix required
2460 }
2462 ////////////////////////////////////////////////////////////////////////////////
2463 // sun.misc.Signal support
2465 static volatile jint sigint_count = 0;
2467 static void
2468 UserHandler(int sig, void *siginfo, void *context) {
2469 // 4511530 - sem_post is serialized and handled by the manager thread. When
2470 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2471 // don't want to flood the manager thread with sem_post requests.
2472 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2473 return;
2475 // Ctrl-C is pressed during error reporting, likely because the error
2476 // handler fails to abort. Let VM die immediately.
2477 if (sig == SIGINT && is_error_reported()) {
2478 os::die();
2479 }
2481 os::signal_notify(sig);
2482 }
2484 void* os::user_handler() {
2485 return CAST_FROM_FN_PTR(void*, UserHandler);
2486 }
2488 class Semaphore : public StackObj {
2489 public:
2490 Semaphore();
2491 ~Semaphore();
2492 void signal();
2493 void wait();
2494 bool trywait();
2495 bool timedwait(unsigned int sec, int nsec);
2496 private:
2497 sem_t _semaphore;
2498 };
2500 Semaphore::Semaphore() {
2501 sem_init(&_semaphore, 0, 0);
2502 }
2504 Semaphore::~Semaphore() {
2505 sem_destroy(&_semaphore);
2506 }
2508 void Semaphore::signal() {
2509 sem_post(&_semaphore);
2510 }
2512 void Semaphore::wait() {
2513 sem_wait(&_semaphore);
2514 }
2516 bool Semaphore::trywait() {
2517 return sem_trywait(&_semaphore) == 0;
2518 }
2520 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2522 struct timespec ts;
2523 // Semaphore's are always associated with CLOCK_REALTIME
2524 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2525 // see unpackTime for discussion on overflow checking
2526 if (sec >= MAX_SECS) {
2527 ts.tv_sec += MAX_SECS;
2528 ts.tv_nsec = 0;
2529 } else {
2530 ts.tv_sec += sec;
2531 ts.tv_nsec += nsec;
2532 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2533 ts.tv_nsec -= NANOSECS_PER_SEC;
2534 ++ts.tv_sec; // note: this must be <= max_secs
2535 }
2536 }
2538 while (1) {
2539 int result = sem_timedwait(&_semaphore, &ts);
2540 if (result == 0) {
2541 return true;
2542 } else if (errno == EINTR) {
2543 continue;
2544 } else if (errno == ETIMEDOUT) {
2545 return false;
2546 } else {
2547 return false;
2548 }
2549 }
2550 }
2552 extern "C" {
2553 typedef void (*sa_handler_t)(int);
2554 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2555 }
2557 void* os::signal(int signal_number, void* handler) {
2558 struct sigaction sigAct, oldSigAct;
2560 sigfillset(&(sigAct.sa_mask));
2561 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2562 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2564 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2565 // -1 means registration failed
2566 return (void *)-1;
2567 }
2569 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2570 }
2572 void os::signal_raise(int signal_number) {
2573 ::raise(signal_number);
2574 }
2576 /*
2577 * The following code is moved from os.cpp for making this
2578 * code platform specific, which it is by its very nature.
2579 */
2581 // Will be modified when max signal is changed to be dynamic
2582 int os::sigexitnum_pd() {
2583 return NSIG;
2584 }
2586 // a counter for each possible signal value
2587 static volatile jint pending_signals[NSIG+1] = { 0 };
2589 // Linux(POSIX) specific hand shaking semaphore.
2590 static sem_t sig_sem;
2591 static Semaphore sr_semaphore;
2593 void os::signal_init_pd() {
2594 // Initialize signal structures
2595 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2597 // Initialize signal semaphore
2598 ::sem_init(&sig_sem, 0, 0);
2599 }
2601 void os::signal_notify(int sig) {
2602 Atomic::inc(&pending_signals[sig]);
2603 ::sem_post(&sig_sem);
2604 }
2606 static int check_pending_signals(bool wait) {
2607 Atomic::store(0, &sigint_count);
2608 for (;;) {
2609 for (int i = 0; i < NSIG + 1; i++) {
2610 jint n = pending_signals[i];
2611 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2612 return i;
2613 }
2614 }
2615 if (!wait) {
2616 return -1;
2617 }
2618 JavaThread *thread = JavaThread::current();
2619 ThreadBlockInVM tbivm(thread);
2621 bool threadIsSuspended;
2622 do {
2623 thread->set_suspend_equivalent();
2624 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2625 ::sem_wait(&sig_sem);
2627 // were we externally suspended while we were waiting?
2628 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2629 if (threadIsSuspended) {
2630 //
2631 // The semaphore has been incremented, but while we were waiting
2632 // another thread suspended us. We don't want to continue running
2633 // while suspended because that would surprise the thread that
2634 // suspended us.
2635 //
2636 ::sem_post(&sig_sem);
2638 thread->java_suspend_self();
2639 }
2640 } while (threadIsSuspended);
2641 }
2642 }
2644 int os::signal_lookup() {
2645 return check_pending_signals(false);
2646 }
2648 int os::signal_wait() {
2649 return check_pending_signals(true);
2650 }
2652 ////////////////////////////////////////////////////////////////////////////////
2653 // Virtual Memory
2655 int os::vm_page_size() {
2656 // Seems redundant as all get out
2657 assert(os::Linux::page_size() != -1, "must call os::init");
2658 return os::Linux::page_size();
2659 }
2661 // Solaris allocates memory by pages.
2662 int os::vm_allocation_granularity() {
2663 assert(os::Linux::page_size() != -1, "must call os::init");
2664 return os::Linux::page_size();
2665 }
2667 // Rationale behind this function:
2668 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2669 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2670 // samples for JITted code. Here we create private executable mapping over the code cache
2671 // and then we can use standard (well, almost, as mapping can change) way to provide
2672 // info for the reporting script by storing timestamp and location of symbol
2673 void linux_wrap_code(char* base, size_t size) {
2674 static volatile jint cnt = 0;
2676 if (!UseOprofile) {
2677 return;
2678 }
2680 char buf[PATH_MAX+1];
2681 int num = Atomic::add(1, &cnt);
2683 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2684 os::get_temp_directory(), os::current_process_id(), num);
2685 unlink(buf);
2687 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2689 if (fd != -1) {
2690 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2691 if (rv != (off_t)-1) {
2692 if (::write(fd, "", 1) == 1) {
2693 mmap(base, size,
2694 PROT_READ|PROT_WRITE|PROT_EXEC,
2695 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2696 }
2697 }
2698 ::close(fd);
2699 unlink(buf);
2700 }
2701 }
2703 static bool recoverable_mmap_error(int err) {
2704 // See if the error is one we can let the caller handle. This
2705 // list of errno values comes from JBS-6843484. I can't find a
2706 // Linux man page that documents this specific set of errno
2707 // values so while this list currently matches Solaris, it may
2708 // change as we gain experience with this failure mode.
2709 switch (err) {
2710 case EBADF:
2711 case EINVAL:
2712 case ENOTSUP:
2713 // let the caller deal with these errors
2714 return true;
2716 default:
2717 // Any remaining errors on this OS can cause our reserved mapping
2718 // to be lost. That can cause confusion where different data
2719 // structures think they have the same memory mapped. The worst
2720 // scenario is if both the VM and a library think they have the
2721 // same memory mapped.
2722 return false;
2723 }
2724 }
2726 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2727 int err) {
2728 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2729 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2730 strerror(err), err);
2731 }
2733 static void warn_fail_commit_memory(char* addr, size_t size,
2734 size_t alignment_hint, bool exec,
2735 int err) {
2736 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2737 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2738 alignment_hint, exec, strerror(err), err);
2739 }
2741 // NOTE: Linux kernel does not really reserve the pages for us.
2742 // All it does is to check if there are enough free pages
2743 // left at the time of mmap(). This could be a potential
2744 // problem.
2745 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2746 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2747 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2748 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2749 if (res != (uintptr_t) MAP_FAILED) {
2750 if (UseNUMAInterleaving) {
2751 numa_make_global(addr, size);
2752 }
2753 return 0;
2754 }
2756 int err = errno; // save errno from mmap() call above
2758 if (!recoverable_mmap_error(err)) {
2759 warn_fail_commit_memory(addr, size, exec, err);
2760 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2761 }
2763 return err;
2764 }
2766 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2767 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2768 }
2770 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2771 const char* mesg) {
2772 assert(mesg != NULL, "mesg must be specified");
2773 int err = os::Linux::commit_memory_impl(addr, size, exec);
2774 if (err != 0) {
2775 // the caller wants all commit errors to exit with the specified mesg:
2776 warn_fail_commit_memory(addr, size, exec, err);
2777 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2778 }
2779 }
2781 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2782 #ifndef MAP_HUGETLB
2783 #define MAP_HUGETLB 0x40000
2784 #endif
2786 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2787 #ifndef MADV_HUGEPAGE
2788 #define MADV_HUGEPAGE 14
2789 #endif
2791 int os::Linux::commit_memory_impl(char* addr, size_t size,
2792 size_t alignment_hint, bool exec) {
2793 int err = os::Linux::commit_memory_impl(addr, size, exec);
2794 if (err == 0) {
2795 realign_memory(addr, size, alignment_hint);
2796 }
2797 return err;
2798 }
2800 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2801 bool exec) {
2802 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2803 }
2805 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2806 size_t alignment_hint, bool exec,
2807 const char* mesg) {
2808 assert(mesg != NULL, "mesg must be specified");
2809 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2810 if (err != 0) {
2811 // the caller wants all commit errors to exit with the specified mesg:
2812 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2813 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2814 }
2815 }
2817 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2818 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2819 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2820 // be supported or the memory may already be backed by huge pages.
2821 ::madvise(addr, bytes, MADV_HUGEPAGE);
2822 }
2823 }
2825 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2826 // This method works by doing an mmap over an existing mmaping and effectively discarding
2827 // the existing pages. However it won't work for SHM-based large pages that cannot be
2828 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2829 // small pages on top of the SHM segment. This method always works for small pages, so we
2830 // allow that in any case.
2831 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2832 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2833 }
2834 }
2836 void os::numa_make_global(char *addr, size_t bytes) {
2837 Linux::numa_interleave_memory(addr, bytes);
2838 }
2840 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2841 // bind policy to MPOL_PREFERRED for the current thread.
2842 #define USE_MPOL_PREFERRED 0
2844 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2845 // To make NUMA and large pages more robust when both enabled, we need to ease
2846 // the requirements on where the memory should be allocated. MPOL_BIND is the
2847 // default policy and it will force memory to be allocated on the specified
2848 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2849 // the specified node, but will not force it. Using this policy will prevent
2850 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2851 // free large pages.
2852 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2853 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2854 }
2856 bool os::numa_topology_changed() { return false; }
2858 size_t os::numa_get_groups_num() {
2859 // Return just the number of nodes in which it's possible to allocate memory
2860 // (in numa terminology, configured nodes).
2861 return Linux::numa_num_configured_nodes();
2862 }
2864 int os::numa_get_group_id() {
2865 int cpu_id = Linux::sched_getcpu();
2866 if (cpu_id != -1) {
2867 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2868 if (lgrp_id != -1) {
2869 return lgrp_id;
2870 }
2871 }
2872 return 0;
2873 }
2875 int os::Linux::get_existing_num_nodes() {
2876 size_t node;
2877 size_t highest_node_number = Linux::numa_max_node();
2878 int num_nodes = 0;
2880 // Get the total number of nodes in the system including nodes without memory.
2881 for (node = 0; node <= highest_node_number; node++) {
2882 if (isnode_in_existing_nodes(node)) {
2883 num_nodes++;
2884 }
2885 }
2886 return num_nodes;
2887 }
2889 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2890 size_t highest_node_number = Linux::numa_max_node();
2891 size_t i = 0;
2893 // Map all node ids in which is possible to allocate memory. Also nodes are
2894 // not always consecutively available, i.e. available from 0 to the highest
2895 // node number.
2896 for (size_t node = 0; node <= highest_node_number; node++) {
2897 if (Linux::isnode_in_configured_nodes(node)) {
2898 ids[i++] = node;
2899 }
2900 }
2901 return i;
2902 }
2904 bool os::get_page_info(char *start, page_info* info) {
2905 return false;
2906 }
2908 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2909 return end;
2910 }
2913 int os::Linux::sched_getcpu_syscall(void) {
2914 unsigned int cpu = 0;
2915 int retval = -1;
2917 #if defined(IA32)
2918 # ifndef SYS_getcpu
2919 # define SYS_getcpu 318
2920 # endif
2921 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2922 #elif defined(AMD64)
2923 // Unfortunately we have to bring all these macros here from vsyscall.h
2924 // to be able to compile on old linuxes.
2925 # define __NR_vgetcpu 2
2926 # define VSYSCALL_START (-10UL << 20)
2927 # define VSYSCALL_SIZE 1024
2928 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2929 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2930 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2931 retval = vgetcpu(&cpu, NULL, NULL);
2932 #endif
2934 return (retval == -1) ? retval : cpu;
2935 }
2937 // Something to do with the numa-aware allocator needs these symbols
2938 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2939 extern "C" JNIEXPORT void numa_error(char *where) { }
2940 extern "C" JNIEXPORT int fork1() { return fork(); }
2942 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2943 // load symbol from base version instead.
2944 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2945 void *f = dlvsym(handle, name, "libnuma_1.1");
2946 if (f == NULL) {
2947 f = dlsym(handle, name);
2948 }
2949 return f;
2950 }
2952 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2953 // Return NULL if the symbol is not defined in this particular version.
2954 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2955 return dlvsym(handle, name, "libnuma_1.2");
2956 }
2958 bool os::Linux::libnuma_init() {
2959 // sched_getcpu() should be in libc.
2960 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2961 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2963 // If it's not, try a direct syscall.
2964 if (sched_getcpu() == -1)
2965 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2967 if (sched_getcpu() != -1) { // Does it work?
2968 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2969 if (handle != NULL) {
2970 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2971 libnuma_dlsym(handle, "numa_node_to_cpus")));
2972 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2973 libnuma_dlsym(handle, "numa_max_node")));
2974 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2975 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2976 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2977 libnuma_dlsym(handle, "numa_available")));
2978 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2979 libnuma_dlsym(handle, "numa_tonode_memory")));
2980 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2981 libnuma_dlsym(handle, "numa_interleave_memory")));
2982 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2983 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2984 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2985 libnuma_dlsym(handle, "numa_set_bind_policy")));
2986 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2987 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2988 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2989 libnuma_dlsym(handle, "numa_distance")));
2991 if (numa_available() != -1) {
2992 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2993 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2994 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2995 // Create an index -> node mapping, since nodes are not always consecutive
2996 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2997 rebuild_nindex_to_node_map();
2998 // Create a cpu -> node mapping
2999 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3000 rebuild_cpu_to_node_map();
3001 return true;
3002 }
3003 }
3004 }
3005 return false;
3006 }
3008 void os::Linux::rebuild_nindex_to_node_map() {
3009 int highest_node_number = Linux::numa_max_node();
3011 nindex_to_node()->clear();
3012 for (int node = 0; node <= highest_node_number; node++) {
3013 if (Linux::isnode_in_existing_nodes(node)) {
3014 nindex_to_node()->append(node);
3015 }
3016 }
3017 }
3019 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3020 // The table is later used in get_node_by_cpu().
3021 void os::Linux::rebuild_cpu_to_node_map() {
3022 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3023 // in libnuma (possible values are starting from 16,
3024 // and continuing up with every other power of 2, but less
3025 // than the maximum number of CPUs supported by kernel), and
3026 // is a subject to change (in libnuma version 2 the requirements
3027 // are more reasonable) we'll just hardcode the number they use
3028 // in the library.
3029 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3031 size_t cpu_num = processor_count();
3032 size_t cpu_map_size = NCPUS / BitsPerCLong;
3033 size_t cpu_map_valid_size =
3034 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3036 cpu_to_node()->clear();
3037 cpu_to_node()->at_grow(cpu_num - 1);
3039 size_t node_num = get_existing_num_nodes();
3041 int distance = 0;
3042 int closest_distance = INT_MAX;
3043 int closest_node = 0;
3044 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3045 for (size_t i = 0; i < node_num; i++) {
3046 // Check if node is configured (not a memory-less node). If it is not, find
3047 // the closest configured node.
3048 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3049 closest_distance = INT_MAX;
3050 // Check distance from all remaining nodes in the system. Ignore distance
3051 // from itself and from another non-configured node.
3052 for (size_t m = 0; m < node_num; m++) {
3053 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3054 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3055 // If a closest node is found, update. There is always at least one
3056 // configured node in the system so there is always at least one node
3057 // close.
3058 if (distance != 0 && distance < closest_distance) {
3059 closest_distance = distance;
3060 closest_node = nindex_to_node()->at(m);
3061 }
3062 }
3063 }
3064 } else {
3065 // Current node is already a configured node.
3066 closest_node = nindex_to_node()->at(i);
3067 }
3069 // Get cpus from the original node and map them to the closest node. If node
3070 // is a configured node (not a memory-less node), then original node and
3071 // closest node are the same.
3072 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3073 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3074 if (cpu_map[j] != 0) {
3075 for (size_t k = 0; k < BitsPerCLong; k++) {
3076 if (cpu_map[j] & (1UL << k)) {
3077 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3078 }
3079 }
3080 }
3081 }
3082 }
3083 }
3084 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
3085 }
3087 int os::Linux::get_node_by_cpu(int cpu_id) {
3088 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3089 return cpu_to_node()->at(cpu_id);
3090 }
3091 return -1;
3092 }
3094 GrowableArray<int>* os::Linux::_cpu_to_node;
3095 GrowableArray<int>* os::Linux::_nindex_to_node;
3096 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3097 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3098 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3099 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3100 os::Linux::numa_available_func_t os::Linux::_numa_available;
3101 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3102 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3103 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3104 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3105 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3106 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3107 unsigned long* os::Linux::_numa_all_nodes;
3108 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3109 struct bitmask* os::Linux::_numa_nodes_ptr;
3111 bool os::pd_uncommit_memory(char* addr, size_t size) {
3112 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3113 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3114 return res != (uintptr_t) MAP_FAILED;
3115 }
3117 static
3118 address get_stack_commited_bottom(address bottom, size_t size) {
3119 address nbot = bottom;
3120 address ntop = bottom + size;
3122 size_t page_sz = os::vm_page_size();
3123 unsigned pages = size / page_sz;
3125 unsigned char vec[1];
3126 unsigned imin = 1, imax = pages + 1, imid;
3127 int mincore_return_value = 0;
3129 assert(imin <= imax, "Unexpected page size");
3131 while (imin < imax) {
3132 imid = (imax + imin) / 2;
3133 nbot = ntop - (imid * page_sz);
3135 // Use a trick with mincore to check whether the page is mapped or not.
3136 // mincore sets vec to 1 if page resides in memory and to 0 if page
3137 // is swapped output but if page we are asking for is unmapped
3138 // it returns -1,ENOMEM
3139 mincore_return_value = mincore(nbot, page_sz, vec);
3141 if (mincore_return_value == -1) {
3142 // Page is not mapped go up
3143 // to find first mapped page
3144 if (errno != EAGAIN) {
3145 assert(errno == ENOMEM, "Unexpected mincore errno");
3146 imax = imid;
3147 }
3148 } else {
3149 // Page is mapped go down
3150 // to find first not mapped page
3151 imin = imid + 1;
3152 }
3153 }
3155 nbot = nbot + page_sz;
3157 // Adjust stack bottom one page up if last checked page is not mapped
3158 if (mincore_return_value == -1) {
3159 nbot = nbot + page_sz;
3160 }
3162 return nbot;
3163 }
3166 // Linux uses a growable mapping for the stack, and if the mapping for
3167 // the stack guard pages is not removed when we detach a thread the
3168 // stack cannot grow beyond the pages where the stack guard was
3169 // mapped. If at some point later in the process the stack expands to
3170 // that point, the Linux kernel cannot expand the stack any further
3171 // because the guard pages are in the way, and a segfault occurs.
3172 //
3173 // However, it's essential not to split the stack region by unmapping
3174 // a region (leaving a hole) that's already part of the stack mapping,
3175 // so if the stack mapping has already grown beyond the guard pages at
3176 // the time we create them, we have to truncate the stack mapping.
3177 // So, we need to know the extent of the stack mapping when
3178 // create_stack_guard_pages() is called.
3180 // We only need this for stacks that are growable: at the time of
3181 // writing thread stacks don't use growable mappings (i.e. those
3182 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3183 // only applies to the main thread.
3185 // If the (growable) stack mapping already extends beyond the point
3186 // where we're going to put our guard pages, truncate the mapping at
3187 // that point by munmap()ping it. This ensures that when we later
3188 // munmap() the guard pages we don't leave a hole in the stack
3189 // mapping. This only affects the main/primordial thread
3191 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3193 if (os::is_primordial_thread()) {
3194 // As we manually grow stack up to bottom inside create_attached_thread(),
3195 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3196 // we don't need to do anything special.
3197 // Check it first, before calling heavy function.
3198 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3199 unsigned char vec[1];
3201 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3202 // Fallback to slow path on all errors, including EAGAIN
3203 stack_extent = (uintptr_t) get_stack_commited_bottom(
3204 os::Linux::initial_thread_stack_bottom(),
3205 (size_t)addr - stack_extent);
3206 }
3208 if (stack_extent < (uintptr_t)addr) {
3209 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3210 }
3211 }
3213 return os::commit_memory(addr, size, !ExecMem);
3214 }
3216 // If this is a growable mapping, remove the guard pages entirely by
3217 // munmap()ping them. If not, just call uncommit_memory(). This only
3218 // affects the main/primordial thread, but guard against future OS changes.
3219 // It's safe to always unmap guard pages for primordial thread because we
3220 // always place it right after end of the mapped region.
3222 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3223 uintptr_t stack_extent, stack_base;
3225 if (os::is_primordial_thread()) {
3226 return ::munmap(addr, size) == 0;
3227 }
3229 return os::uncommit_memory(addr, size);
3230 }
3232 static address _highest_vm_reserved_address = NULL;
3234 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3235 // at 'requested_addr'. If there are existing memory mappings at the same
3236 // location, however, they will be overwritten. If 'fixed' is false,
3237 // 'requested_addr' is only treated as a hint, the return value may or
3238 // may not start from the requested address. Unlike Linux mmap(), this
3239 // function returns NULL to indicate failure.
3240 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3241 char * addr;
3242 int flags;
3244 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3245 if (fixed) {
3246 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3247 flags |= MAP_FIXED;
3248 }
3250 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3251 // touch an uncommitted page. Otherwise, the read/write might
3252 // succeed if we have enough swap space to back the physical page.
3253 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3254 flags, -1, 0);
3256 if (addr != MAP_FAILED) {
3257 // anon_mmap() should only get called during VM initialization,
3258 // don't need lock (actually we can skip locking even it can be called
3259 // from multiple threads, because _highest_vm_reserved_address is just a
3260 // hint about the upper limit of non-stack memory regions.)
3261 if ((address)addr + bytes > _highest_vm_reserved_address) {
3262 _highest_vm_reserved_address = (address)addr + bytes;
3263 }
3264 }
3266 return addr == MAP_FAILED ? NULL : addr;
3267 }
3269 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3270 // (req_addr != NULL) or with a given alignment.
3271 // - bytes shall be a multiple of alignment.
3272 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3273 // - alignment sets the alignment at which memory shall be allocated.
3274 // It must be a multiple of allocation granularity.
3275 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3276 // req_addr or NULL.
3277 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3279 size_t extra_size = bytes;
3280 if (req_addr == NULL && alignment > 0) {
3281 extra_size += alignment;
3282 }
3284 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3285 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3286 -1, 0);
3287 if (start == MAP_FAILED) {
3288 start = NULL;
3289 } else {
3290 if (req_addr != NULL) {
3291 if (start != req_addr) {
3292 ::munmap(start, extra_size);
3293 start = NULL;
3294 }
3295 } else {
3296 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3297 char* const end_aligned = start_aligned + bytes;
3298 char* const end = start + extra_size;
3299 if (start_aligned > start) {
3300 ::munmap(start, start_aligned - start);
3301 }
3302 if (end_aligned < end) {
3303 ::munmap(end_aligned, end - end_aligned);
3304 }
3305 start = start_aligned;
3306 }
3307 }
3308 return start;
3309 }
3311 // Don't update _highest_vm_reserved_address, because there might be memory
3312 // regions above addr + size. If so, releasing a memory region only creates
3313 // a hole in the address space, it doesn't help prevent heap-stack collision.
3314 //
3315 static int anon_munmap(char * addr, size_t size) {
3316 return ::munmap(addr, size) == 0;
3317 }
3319 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3320 size_t alignment_hint) {
3321 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3322 }
3324 bool os::pd_release_memory(char* addr, size_t size) {
3325 return anon_munmap(addr, size);
3326 }
3328 static address highest_vm_reserved_address() {
3329 return _highest_vm_reserved_address;
3330 }
3332 static bool linux_mprotect(char* addr, size_t size, int prot) {
3333 // Linux wants the mprotect address argument to be page aligned.
3334 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3336 // According to SUSv3, mprotect() should only be used with mappings
3337 // established by mmap(), and mmap() always maps whole pages. Unaligned
3338 // 'addr' likely indicates problem in the VM (e.g. trying to change
3339 // protection of malloc'ed or statically allocated memory). Check the
3340 // caller if you hit this assert.
3341 assert(addr == bottom, "sanity check");
3343 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3344 return ::mprotect(bottom, size, prot) == 0;
3345 }
3347 // Set protections specified
3348 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3349 bool is_committed) {
3350 unsigned int p = 0;
3351 switch (prot) {
3352 case MEM_PROT_NONE: p = PROT_NONE; break;
3353 case MEM_PROT_READ: p = PROT_READ; break;
3354 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3355 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3356 default:
3357 ShouldNotReachHere();
3358 }
3359 // is_committed is unused.
3360 return linux_mprotect(addr, bytes, p);
3361 }
3363 bool os::guard_memory(char* addr, size_t size) {
3364 return linux_mprotect(addr, size, PROT_NONE);
3365 }
3367 bool os::unguard_memory(char* addr, size_t size) {
3368 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3369 }
3371 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3372 bool result = false;
3373 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3374 MAP_ANONYMOUS|MAP_PRIVATE,
3375 -1, 0);
3376 if (p != MAP_FAILED) {
3377 void *aligned_p = align_ptr_up(p, page_size);
3379 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3381 munmap(p, page_size * 2);
3382 }
3384 if (warn && !result) {
3385 warning("TransparentHugePages is not supported by the operating system.");
3386 }
3388 return result;
3389 }
3391 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3392 bool result = false;
3393 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3394 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3395 -1, 0);
3397 if (p != MAP_FAILED) {
3398 // We don't know if this really is a huge page or not.
3399 FILE *fp = fopen("/proc/self/maps", "r");
3400 if (fp) {
3401 while (!feof(fp)) {
3402 char chars[257];
3403 long x = 0;
3404 if (fgets(chars, sizeof(chars), fp)) {
3405 if (sscanf(chars, "%lx-%*x", &x) == 1
3406 && x == (long)p) {
3407 if (strstr (chars, "hugepage")) {
3408 result = true;
3409 break;
3410 }
3411 }
3412 }
3413 }
3414 fclose(fp);
3415 }
3416 munmap(p, page_size);
3417 }
3419 if (warn && !result) {
3420 warning("HugeTLBFS is not supported by the operating system.");
3421 }
3423 return result;
3424 }
3426 /*
3427 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3428 *
3429 * From the coredump_filter documentation:
3430 *
3431 * - (bit 0) anonymous private memory
3432 * - (bit 1) anonymous shared memory
3433 * - (bit 2) file-backed private memory
3434 * - (bit 3) file-backed shared memory
3435 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3436 * effective only if the bit 2 is cleared)
3437 * - (bit 5) hugetlb private memory
3438 * - (bit 6) hugetlb shared memory
3439 */
3440 static void set_coredump_filter(void) {
3441 FILE *f;
3442 long cdm;
3444 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3445 return;
3446 }
3448 if (fscanf(f, "%lx", &cdm) != 1) {
3449 fclose(f);
3450 return;
3451 }
3453 rewind(f);
3455 if ((cdm & LARGEPAGES_BIT) == 0) {
3456 cdm |= LARGEPAGES_BIT;
3457 fprintf(f, "%#lx", cdm);
3458 }
3460 fclose(f);
3461 }
3463 // Large page support
3465 static size_t _large_page_size = 0;
3467 size_t os::Linux::find_large_page_size() {
3468 size_t large_page_size = 0;
3470 // large_page_size on Linux is used to round up heap size. x86 uses either
3471 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3472 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3473 // page as large as 256M.
3474 //
3475 // Here we try to figure out page size by parsing /proc/meminfo and looking
3476 // for a line with the following format:
3477 // Hugepagesize: 2048 kB
3478 //
3479 // If we can't determine the value (e.g. /proc is not mounted, or the text
3480 // format has been changed), we'll use the largest page size supported by
3481 // the processor.
3483 #ifndef ZERO
3484 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3485 ARM_ONLY(2 * M) PPC_ONLY(4 * M) MIPS64_ONLY(4 * M); //In MIPS _large_page_size is seted 4*M.
3486 #endif // ZERO
3488 FILE *fp = fopen("/proc/meminfo", "r");
3489 if (fp) {
3490 while (!feof(fp)) {
3491 int x = 0;
3492 char buf[16];
3493 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3494 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3495 large_page_size = x * K;
3496 break;
3497 }
3498 } else {
3499 // skip to next line
3500 for (;;) {
3501 int ch = fgetc(fp);
3502 if (ch == EOF || ch == (int)'\n') break;
3503 }
3504 }
3505 }
3506 fclose(fp);
3507 }
3509 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3510 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3511 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3512 proper_unit_for_byte_size(large_page_size));
3513 }
3515 return large_page_size;
3516 }
3518 size_t os::Linux::setup_large_page_size() {
3519 _large_page_size = Linux::find_large_page_size();
3520 const size_t default_page_size = (size_t)Linux::page_size();
3521 if (_large_page_size > default_page_size) {
3522 _page_sizes[0] = _large_page_size;
3523 _page_sizes[1] = default_page_size;
3524 _page_sizes[2] = 0;
3525 }
3527 return _large_page_size;
3528 }
3530 bool os::Linux::setup_large_page_type(size_t page_size) {
3531 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3532 FLAG_IS_DEFAULT(UseSHM) &&
3533 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3535 // The type of large pages has not been specified by the user.
3537 // Try UseHugeTLBFS and then UseSHM.
3538 UseHugeTLBFS = UseSHM = true;
3540 // Don't try UseTransparentHugePages since there are known
3541 // performance issues with it turned on. This might change in the future.
3542 UseTransparentHugePages = false;
3543 }
3545 if (UseTransparentHugePages) {
3546 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3547 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3548 UseHugeTLBFS = false;
3549 UseSHM = false;
3550 return true;
3551 }
3552 UseTransparentHugePages = false;
3553 }
3555 if (UseHugeTLBFS) {
3556 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3557 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3558 UseSHM = false;
3559 return true;
3560 }
3561 UseHugeTLBFS = false;
3562 }
3564 return UseSHM;
3565 }
3567 void os::large_page_init() {
3568 if (!UseLargePages &&
3569 !UseTransparentHugePages &&
3570 !UseHugeTLBFS &&
3571 !UseSHM) {
3572 // Not using large pages.
3573 return;
3574 }
3576 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3577 // The user explicitly turned off large pages.
3578 // Ignore the rest of the large pages flags.
3579 UseTransparentHugePages = false;
3580 UseHugeTLBFS = false;
3581 UseSHM = false;
3582 return;
3583 }
3585 size_t large_page_size = Linux::setup_large_page_size();
3586 UseLargePages = Linux::setup_large_page_type(large_page_size);
3588 set_coredump_filter();
3589 }
3591 #ifndef SHM_HUGETLB
3592 #define SHM_HUGETLB 04000
3593 #endif
3595 #define shm_warning_format(format, ...) \
3596 do { \
3597 if (UseLargePages && \
3598 (!FLAG_IS_DEFAULT(UseLargePages) || \
3599 !FLAG_IS_DEFAULT(UseSHM) || \
3600 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3601 warning(format, __VA_ARGS__); \
3602 } \
3603 } while (0)
3605 #define shm_warning(str) shm_warning_format("%s", str)
3607 #define shm_warning_with_errno(str) \
3608 do { \
3609 int err = errno; \
3610 shm_warning_format(str " (error = %d)", err); \
3611 } while (0)
3613 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3614 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3616 if (!is_size_aligned(alignment, SHMLBA)) {
3617 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3618 return NULL;
3619 }
3621 // To ensure that we get 'alignment' aligned memory from shmat,
3622 // we pre-reserve aligned virtual memory and then attach to that.
3624 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3625 if (pre_reserved_addr == NULL) {
3626 // Couldn't pre-reserve aligned memory.
3627 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3628 return NULL;
3629 }
3631 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3632 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3634 if ((intptr_t)addr == -1) {
3635 int err = errno;
3636 shm_warning_with_errno("Failed to attach shared memory.");
3638 assert(err != EACCES, "Unexpected error");
3639 assert(err != EIDRM, "Unexpected error");
3640 assert(err != EINVAL, "Unexpected error");
3642 // Since we don't know if the kernel unmapped the pre-reserved memory area
3643 // we can't unmap it, since that would potentially unmap memory that was
3644 // mapped from other threads.
3645 return NULL;
3646 }
3648 return addr;
3649 }
3651 static char* shmat_at_address(int shmid, char* req_addr) {
3652 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3653 assert(false, "Requested address needs to be SHMLBA aligned");
3654 return NULL;
3655 }
3657 char* addr = (char*)shmat(shmid, req_addr, 0);
3659 if ((intptr_t)addr == -1) {
3660 shm_warning_with_errno("Failed to attach shared memory.");
3661 return NULL;
3662 }
3664 return addr;
3665 }
3667 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3668 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3669 if (req_addr != NULL) {
3670 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3671 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3672 return shmat_at_address(shmid, req_addr);
3673 }
3675 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3676 // return large page size aligned memory addresses when req_addr == NULL.
3677 // However, if the alignment is larger than the large page size, we have
3678 // to manually ensure that the memory returned is 'alignment' aligned.
3679 if (alignment > os::large_page_size()) {
3680 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3681 return shmat_with_alignment(shmid, bytes, alignment);
3682 } else {
3683 return shmat_at_address(shmid, NULL);
3684 }
3685 }
3687 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3688 // "exec" is passed in but not used. Creating the shared image for
3689 // the code cache doesn't have an SHM_X executable permission to check.
3690 assert(UseLargePages && UseSHM, "only for SHM large pages");
3691 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3692 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3694 if (!is_size_aligned(bytes, os::large_page_size())) {
3695 return NULL; // Fallback to small pages.
3696 }
3698 // Create a large shared memory region to attach to based on size.
3699 // Currently, size is the total size of the heap.
3700 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3701 if (shmid == -1) {
3702 // Possible reasons for shmget failure:
3703 // 1. shmmax is too small for Java heap.
3704 // > check shmmax value: cat /proc/sys/kernel/shmmax
3705 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3706 // 2. not enough large page memory.
3707 // > check available large pages: cat /proc/meminfo
3708 // > increase amount of large pages:
3709 // echo new_value > /proc/sys/vm/nr_hugepages
3710 // Note 1: different Linux may use different name for this property,
3711 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3712 // Note 2: it's possible there's enough physical memory available but
3713 // they are so fragmented after a long run that they can't
3714 // coalesce into large pages. Try to reserve large pages when
3715 // the system is still "fresh".
3716 shm_warning_with_errno("Failed to reserve shared memory.");
3717 return NULL;
3718 }
3720 // Attach to the region.
3721 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3723 // Remove shmid. If shmat() is successful, the actual shared memory segment
3724 // will be deleted when it's detached by shmdt() or when the process
3725 // terminates. If shmat() is not successful this will remove the shared
3726 // segment immediately.
3727 shmctl(shmid, IPC_RMID, NULL);
3729 return addr;
3730 }
3732 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3733 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3735 bool warn_on_failure = UseLargePages &&
3736 (!FLAG_IS_DEFAULT(UseLargePages) ||
3737 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3738 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3740 if (warn_on_failure) {
3741 char msg[128];
3742 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3743 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3744 warning("%s", msg);
3745 }
3746 }
3748 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3749 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3750 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3751 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3753 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3754 char* addr = (char*)::mmap(req_addr, bytes, prot,
3755 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3756 -1, 0);
3758 if (addr == MAP_FAILED) {
3759 warn_on_large_pages_failure(req_addr, bytes, errno);
3760 return NULL;
3761 }
3763 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3765 return addr;
3766 }
3768 // Reserve memory using mmap(MAP_HUGETLB).
3769 // - bytes shall be a multiple of alignment.
3770 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3771 // - alignment sets the alignment at which memory shall be allocated.
3772 // It must be a multiple of allocation granularity.
3773 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3774 // req_addr or NULL.
3775 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3776 size_t large_page_size = os::large_page_size();
3777 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3779 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3780 assert(is_size_aligned(bytes, alignment), "Must be");
3782 // First reserve - but not commit - the address range in small pages.
3783 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3785 if (start == NULL) {
3786 return NULL;
3787 }
3789 assert(is_ptr_aligned(start, alignment), "Must be");
3791 char* end = start + bytes;
3793 // Find the regions of the allocated chunk that can be promoted to large pages.
3794 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3795 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3797 size_t lp_bytes = lp_end - lp_start;
3799 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3801 if (lp_bytes == 0) {
3802 // The mapped region doesn't even span the start and the end of a large page.
3803 // Fall back to allocate a non-special area.
3804 ::munmap(start, end - start);
3805 return NULL;
3806 }
3808 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3810 void* result;
3812 // Commit small-paged leading area.
3813 if (start != lp_start) {
3814 result = ::mmap(start, lp_start - start, prot,
3815 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3816 -1, 0);
3817 if (result == MAP_FAILED) {
3818 ::munmap(lp_start, end - lp_start);
3819 return NULL;
3820 }
3821 }
3823 // Commit large-paged area.
3824 result = ::mmap(lp_start, lp_bytes, prot,
3825 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3826 -1, 0);
3827 if (result == MAP_FAILED) {
3828 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3829 // If the mmap above fails, the large pages region will be unmapped and we
3830 // have regions before and after with small pages. Release these regions.
3831 //
3832 // | mapped | unmapped | mapped |
3833 // ^ ^ ^ ^
3834 // start lp_start lp_end end
3835 //
3836 ::munmap(start, lp_start - start);
3837 ::munmap(lp_end, end - lp_end);
3838 return NULL;
3839 }
3841 // Commit small-paged trailing area.
3842 if (lp_end != end) {
3843 result = ::mmap(lp_end, end - lp_end, prot,
3844 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3845 -1, 0);
3846 if (result == MAP_FAILED) {
3847 ::munmap(start, lp_end - start);
3848 return NULL;
3849 }
3850 }
3852 return start;
3853 }
3855 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3856 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3857 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3858 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3859 assert(is_power_of_2(os::large_page_size()), "Must be");
3860 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3862 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3863 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3864 } else {
3865 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3866 }
3867 }
3869 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3870 assert(UseLargePages, "only for large pages");
3872 char* addr;
3873 if (UseSHM) {
3874 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3875 } else {
3876 assert(UseHugeTLBFS, "must be");
3877 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3878 }
3880 if (addr != NULL) {
3881 if (UseNUMAInterleaving) {
3882 numa_make_global(addr, bytes);
3883 }
3885 // The memory is committed
3886 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3887 }
3889 return addr;
3890 }
3892 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3893 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3894 return shmdt(base) == 0;
3895 }
3897 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3898 return pd_release_memory(base, bytes);
3899 }
3901 bool os::release_memory_special(char* base, size_t bytes) {
3902 bool res;
3903 if (MemTracker::tracking_level() > NMT_minimal) {
3904 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3905 res = os::Linux::release_memory_special_impl(base, bytes);
3906 if (res) {
3907 tkr.record((address)base, bytes);
3908 }
3910 } else {
3911 res = os::Linux::release_memory_special_impl(base, bytes);
3912 }
3913 return res;
3914 }
3916 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3917 assert(UseLargePages, "only for large pages");
3918 bool res;
3920 if (UseSHM) {
3921 res = os::Linux::release_memory_special_shm(base, bytes);
3922 } else {
3923 assert(UseHugeTLBFS, "must be");
3924 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3925 }
3926 return res;
3927 }
3929 size_t os::large_page_size() {
3930 return _large_page_size;
3931 }
3933 // With SysV SHM the entire memory region must be allocated as shared
3934 // memory.
3935 // HugeTLBFS allows application to commit large page memory on demand.
3936 // However, when committing memory with HugeTLBFS fails, the region
3937 // that was supposed to be committed will lose the old reservation
3938 // and allow other threads to steal that memory region. Because of this
3939 // behavior we can't commit HugeTLBFS memory.
3940 bool os::can_commit_large_page_memory() {
3941 return UseTransparentHugePages;
3942 }
3944 bool os::can_execute_large_page_memory() {
3945 return UseTransparentHugePages || UseHugeTLBFS;
3946 }
3948 // Reserve memory at an arbitrary address, only if that area is
3949 // available (and not reserved for something else).
3951 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3952 const int max_tries = 10;
3953 char* base[max_tries];
3954 size_t size[max_tries];
3955 const size_t gap = 0x000000;
3957 // Assert only that the size is a multiple of the page size, since
3958 // that's all that mmap requires, and since that's all we really know
3959 // about at this low abstraction level. If we need higher alignment,
3960 // we can either pass an alignment to this method or verify alignment
3961 // in one of the methods further up the call chain. See bug 5044738.
3962 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3964 // Repeatedly allocate blocks until the block is allocated at the
3965 // right spot. Give up after max_tries. Note that reserve_memory() will
3966 // automatically update _highest_vm_reserved_address if the call is
3967 // successful. The variable tracks the highest memory address every reserved
3968 // by JVM. It is used to detect heap-stack collision if running with
3969 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3970 // space than needed, it could confuse the collision detecting code. To
3971 // solve the problem, save current _highest_vm_reserved_address and
3972 // calculate the correct value before return.
3973 address old_highest = _highest_vm_reserved_address;
3975 // Linux mmap allows caller to pass an address as hint; give it a try first,
3976 // if kernel honors the hint then we can return immediately.
3977 char * addr = anon_mmap(requested_addr, bytes, false);
3978 if (addr == requested_addr) {
3979 return requested_addr;
3980 }
3982 if (addr != NULL) {
3983 // mmap() is successful but it fails to reserve at the requested address
3984 anon_munmap(addr, bytes);
3985 }
3987 int i;
3988 for (i = 0; i < max_tries; ++i) {
3989 base[i] = reserve_memory(bytes);
3991 if (base[i] != NULL) {
3992 // Is this the block we wanted?
3993 if (base[i] == requested_addr) {
3994 size[i] = bytes;
3995 break;
3996 }
3998 // Does this overlap the block we wanted? Give back the overlapped
3999 // parts and try again.
4001 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
4002 if (top_overlap >= 0 && top_overlap < bytes) {
4003 unmap_memory(base[i], top_overlap);
4004 base[i] += top_overlap;
4005 size[i] = bytes - top_overlap;
4006 } else {
4007 size_t bottom_overlap = base[i] + bytes - requested_addr;
4008 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
4009 unmap_memory(requested_addr, bottom_overlap);
4010 size[i] = bytes - bottom_overlap;
4011 } else {
4012 size[i] = bytes;
4013 }
4014 }
4015 }
4016 }
4018 // Give back the unused reserved pieces.
4020 for (int j = 0; j < i; ++j) {
4021 if (base[j] != NULL) {
4022 unmap_memory(base[j], size[j]);
4023 }
4024 }
4026 if (i < max_tries) {
4027 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
4028 return requested_addr;
4029 } else {
4030 _highest_vm_reserved_address = old_highest;
4031 return NULL;
4032 }
4033 }
4035 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4036 return ::read(fd, buf, nBytes);
4037 }
4039 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
4040 // Solaris uses poll(), linux uses park().
4041 // Poll() is likely a better choice, assuming that Thread.interrupt()
4042 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
4043 // SIGSEGV, see 4355769.
4045 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
4046 assert(thread == Thread::current(), "thread consistency check");
4048 ParkEvent * const slp = thread->_SleepEvent ;
4049 slp->reset() ;
4050 OrderAccess::fence() ;
4052 if (interruptible) {
4053 jlong prevtime = javaTimeNanos();
4055 for (;;) {
4056 if (os::is_interrupted(thread, true)) {
4057 return OS_INTRPT;
4058 }
4060 jlong newtime = javaTimeNanos();
4062 if (newtime - prevtime < 0) {
4063 // time moving backwards, should only happen if no monotonic clock
4064 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4065 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4066 } else {
4067 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4068 }
4070 if(millis <= 0) {
4071 return OS_OK;
4072 }
4074 prevtime = newtime;
4076 {
4077 assert(thread->is_Java_thread(), "sanity check");
4078 JavaThread *jt = (JavaThread *) thread;
4079 ThreadBlockInVM tbivm(jt);
4080 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
4082 jt->set_suspend_equivalent();
4083 // cleared by handle_special_suspend_equivalent_condition() or
4084 // java_suspend_self() via check_and_wait_while_suspended()
4086 slp->park(millis);
4088 // were we externally suspended while we were waiting?
4089 jt->check_and_wait_while_suspended();
4090 }
4091 }
4092 } else {
4093 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4094 jlong prevtime = javaTimeNanos();
4096 for (;;) {
4097 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
4098 // the 1st iteration ...
4099 jlong newtime = javaTimeNanos();
4101 if (newtime - prevtime < 0) {
4102 // time moving backwards, should only happen if no monotonic clock
4103 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4104 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4105 } else {
4106 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4107 }
4109 if(millis <= 0) break ;
4111 prevtime = newtime;
4112 slp->park(millis);
4113 }
4114 return OS_OK ;
4115 }
4116 }
4118 //
4119 // Short sleep, direct OS call.
4120 //
4121 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4122 // sched_yield(2) will actually give up the CPU:
4123 //
4124 // * Alone on this pariticular CPU, keeps running.
4125 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4126 // (pre 2.6.39).
4127 //
4128 // So calling this with 0 is an alternative.
4129 //
4130 void os::naked_short_sleep(jlong ms) {
4131 struct timespec req;
4133 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4134 req.tv_sec = 0;
4135 if (ms > 0) {
4136 req.tv_nsec = (ms % 1000) * 1000000;
4137 }
4138 else {
4139 req.tv_nsec = 1;
4140 }
4142 nanosleep(&req, NULL);
4144 return;
4145 }
4147 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4148 void os::infinite_sleep() {
4149 while (true) { // sleep forever ...
4150 ::sleep(100); // ... 100 seconds at a time
4151 }
4152 }
4154 // Used to convert frequent JVM_Yield() to nops
4155 bool os::dont_yield() {
4156 return DontYieldALot;
4157 }
4159 void os::yield() {
4160 sched_yield();
4161 }
4163 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
4165 void os::yield_all(int attempts) {
4166 // Yields to all threads, including threads with lower priorities
4167 // Threads on Linux are all with same priority. The Solaris style
4168 // os::yield_all() with nanosleep(1ms) is not necessary.
4169 sched_yield();
4170 }
4172 // Called from the tight loops to possibly influence time-sharing heuristics
4173 void os::loop_breaker(int attempts) {
4174 os::yield_all(attempts);
4175 }
4177 ////////////////////////////////////////////////////////////////////////////////
4178 // thread priority support
4180 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4181 // only supports dynamic priority, static priority must be zero. For real-time
4182 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4183 // However, for large multi-threaded applications, SCHED_RR is not only slower
4184 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4185 // of 5 runs - Sep 2005).
4186 //
4187 // The following code actually changes the niceness of kernel-thread/LWP. It
4188 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4189 // not the entire user process, and user level threads are 1:1 mapped to kernel
4190 // threads. It has always been the case, but could change in the future. For
4191 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4192 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4194 int os::java_to_os_priority[CriticalPriority + 1] = {
4195 19, // 0 Entry should never be used
4197 4, // 1 MinPriority
4198 3, // 2
4199 2, // 3
4201 1, // 4
4202 0, // 5 NormPriority
4203 -1, // 6
4205 -2, // 7
4206 -3, // 8
4207 -4, // 9 NearMaxPriority
4209 -5, // 10 MaxPriority
4211 -5 // 11 CriticalPriority
4212 };
4214 static int prio_init() {
4215 if (ThreadPriorityPolicy == 1) {
4216 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4217 // if effective uid is not root. Perhaps, a more elegant way of doing
4218 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4219 if (geteuid() != 0) {
4220 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4221 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4222 }
4223 ThreadPriorityPolicy = 0;
4224 }
4225 }
4226 if (UseCriticalJavaThreadPriority) {
4227 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4228 }
4229 return 0;
4230 }
4232 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4233 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4235 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4236 return (ret == 0) ? OS_OK : OS_ERR;
4237 }
4239 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4240 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4241 *priority_ptr = java_to_os_priority[NormPriority];
4242 return OS_OK;
4243 }
4245 errno = 0;
4246 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4247 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4248 }
4250 // Hint to the underlying OS that a task switch would not be good.
4251 // Void return because it's a hint and can fail.
4252 void os::hint_no_preempt() {}
4254 ////////////////////////////////////////////////////////////////////////////////
4255 // suspend/resume support
4257 // the low-level signal-based suspend/resume support is a remnant from the
4258 // old VM-suspension that used to be for java-suspension, safepoints etc,
4259 // within hotspot. Now there is a single use-case for this:
4260 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4261 // that runs in the watcher thread.
4262 // The remaining code is greatly simplified from the more general suspension
4263 // code that used to be used.
4264 //
4265 // The protocol is quite simple:
4266 // - suspend:
4267 // - sends a signal to the target thread
4268 // - polls the suspend state of the osthread using a yield loop
4269 // - target thread signal handler (SR_handler) sets suspend state
4270 // and blocks in sigsuspend until continued
4271 // - resume:
4272 // - sets target osthread state to continue
4273 // - sends signal to end the sigsuspend loop in the SR_handler
4274 //
4275 // Note that the SR_lock plays no role in this suspend/resume protocol.
4276 //
4278 static void resume_clear_context(OSThread *osthread) {
4279 osthread->set_ucontext(NULL);
4280 osthread->set_siginfo(NULL);
4281 }
4283 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4284 osthread->set_ucontext(context);
4285 osthread->set_siginfo(siginfo);
4286 }
4288 //
4289 // Handler function invoked when a thread's execution is suspended or
4290 // resumed. We have to be careful that only async-safe functions are
4291 // called here (Note: most pthread functions are not async safe and
4292 // should be avoided.)
4293 //
4294 // Note: sigwait() is a more natural fit than sigsuspend() from an
4295 // interface point of view, but sigwait() prevents the signal hander
4296 // from being run. libpthread would get very confused by not having
4297 // its signal handlers run and prevents sigwait()'s use with the
4298 // mutex granting granting signal.
4299 //
4300 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4301 //
4302 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4303 // Save and restore errno to avoid confusing native code with EINTR
4304 // after sigsuspend.
4305 int old_errno = errno;
4307 Thread* thread = Thread::current();
4308 OSThread* osthread = thread->osthread();
4309 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4311 os::SuspendResume::State current = osthread->sr.state();
4312 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4313 suspend_save_context(osthread, siginfo, context);
4315 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4316 os::SuspendResume::State state = osthread->sr.suspended();
4317 if (state == os::SuspendResume::SR_SUSPENDED) {
4318 sigset_t suspend_set; // signals for sigsuspend()
4320 // get current set of blocked signals and unblock resume signal
4321 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4322 sigdelset(&suspend_set, SR_signum);
4324 sr_semaphore.signal();
4325 // wait here until we are resumed
4326 while (1) {
4327 sigsuspend(&suspend_set);
4329 os::SuspendResume::State result = osthread->sr.running();
4330 if (result == os::SuspendResume::SR_RUNNING) {
4331 sr_semaphore.signal();
4332 break;
4333 }
4334 }
4336 } else if (state == os::SuspendResume::SR_RUNNING) {
4337 // request was cancelled, continue
4338 } else {
4339 ShouldNotReachHere();
4340 }
4342 resume_clear_context(osthread);
4343 } else if (current == os::SuspendResume::SR_RUNNING) {
4344 // request was cancelled, continue
4345 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4346 // ignore
4347 } else {
4348 // ignore
4349 }
4351 errno = old_errno;
4352 }
4355 static int SR_initialize() {
4356 struct sigaction act;
4357 char *s;
4358 /* Get signal number to use for suspend/resume */
4359 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4360 int sig = ::strtol(s, 0, 10);
4361 if (sig > 0 || sig < _NSIG) {
4362 SR_signum = sig;
4363 }
4364 }
4366 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4367 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4369 sigemptyset(&SR_sigset);
4370 sigaddset(&SR_sigset, SR_signum);
4372 /* Set up signal handler for suspend/resume */
4373 act.sa_flags = SA_RESTART|SA_SIGINFO;
4374 act.sa_handler = (void (*)(int)) SR_handler;
4376 // SR_signum is blocked by default.
4377 // 4528190 - We also need to block pthread restart signal (32 on all
4378 // supported Linux platforms). Note that LinuxThreads need to block
4379 // this signal for all threads to work properly. So we don't have
4380 // to use hard-coded signal number when setting up the mask.
4381 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4383 if (sigaction(SR_signum, &act, 0) == -1) {
4384 return -1;
4385 }
4387 // Save signal flag
4388 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4389 return 0;
4390 }
4392 static int sr_notify(OSThread* osthread) {
4393 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4394 assert_status(status == 0, status, "pthread_kill");
4395 return status;
4396 }
4398 // "Randomly" selected value for how long we want to spin
4399 // before bailing out on suspending a thread, also how often
4400 // we send a signal to a thread we want to resume
4401 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4402 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4404 // returns true on success and false on error - really an error is fatal
4405 // but this seems the normal response to library errors
4406 static bool do_suspend(OSThread* osthread) {
4407 assert(osthread->sr.is_running(), "thread should be running");
4408 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4410 // mark as suspended and send signal
4411 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4412 // failed to switch, state wasn't running?
4413 ShouldNotReachHere();
4414 return false;
4415 }
4417 if (sr_notify(osthread) != 0) {
4418 ShouldNotReachHere();
4419 }
4421 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4422 while (true) {
4423 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4424 break;
4425 } else {
4426 // timeout
4427 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4428 if (cancelled == os::SuspendResume::SR_RUNNING) {
4429 return false;
4430 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4431 // make sure that we consume the signal on the semaphore as well
4432 sr_semaphore.wait();
4433 break;
4434 } else {
4435 ShouldNotReachHere();
4436 return false;
4437 }
4438 }
4439 }
4441 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4442 return true;
4443 }
4445 static void do_resume(OSThread* osthread) {
4446 assert(osthread->sr.is_suspended(), "thread should be suspended");
4447 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4449 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4450 // failed to switch to WAKEUP_REQUEST
4451 ShouldNotReachHere();
4452 return;
4453 }
4455 while (true) {
4456 if (sr_notify(osthread) == 0) {
4457 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4458 if (osthread->sr.is_running()) {
4459 return;
4460 }
4461 }
4462 } else {
4463 ShouldNotReachHere();
4464 }
4465 }
4467 guarantee(osthread->sr.is_running(), "Must be running!");
4468 }
4470 ////////////////////////////////////////////////////////////////////////////////
4471 // interrupt support
4473 void os::interrupt(Thread* thread) {
4474 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4475 "possibility of dangling Thread pointer");
4477 OSThread* osthread = thread->osthread();
4479 if (!osthread->interrupted()) {
4480 osthread->set_interrupted(true);
4481 // More than one thread can get here with the same value of osthread,
4482 // resulting in multiple notifications. We do, however, want the store
4483 // to interrupted() to be visible to other threads before we execute unpark().
4484 OrderAccess::fence();
4485 ParkEvent * const slp = thread->_SleepEvent ;
4486 if (slp != NULL) slp->unpark() ;
4487 }
4489 // For JSR166. Unpark even if interrupt status already was set
4490 if (thread->is_Java_thread())
4491 ((JavaThread*)thread)->parker()->unpark();
4493 ParkEvent * ev = thread->_ParkEvent ;
4494 if (ev != NULL) ev->unpark() ;
4496 }
4498 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4499 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4500 "possibility of dangling Thread pointer");
4502 OSThread* osthread = thread->osthread();
4504 bool interrupted = osthread->interrupted();
4506 if (interrupted && clear_interrupted) {
4507 osthread->set_interrupted(false);
4508 // consider thread->_SleepEvent->reset() ... optional optimization
4509 }
4511 return interrupted;
4512 }
4514 ///////////////////////////////////////////////////////////////////////////////////
4515 // signal handling (except suspend/resume)
4517 // This routine may be used by user applications as a "hook" to catch signals.
4518 // The user-defined signal handler must pass unrecognized signals to this
4519 // routine, and if it returns true (non-zero), then the signal handler must
4520 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4521 // routine will never retun false (zero), but instead will execute a VM panic
4522 // routine kill the process.
4523 //
4524 // If this routine returns false, it is OK to call it again. This allows
4525 // the user-defined signal handler to perform checks either before or after
4526 // the VM performs its own checks. Naturally, the user code would be making
4527 // a serious error if it tried to handle an exception (such as a null check
4528 // or breakpoint) that the VM was generating for its own correct operation.
4529 //
4530 // This routine may recognize any of the following kinds of signals:
4531 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4532 // It should be consulted by handlers for any of those signals.
4533 //
4534 // The caller of this routine must pass in the three arguments supplied
4535 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4536 // field of the structure passed to sigaction(). This routine assumes that
4537 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4538 //
4539 // Note that the VM will print warnings if it detects conflicting signal
4540 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4541 //
4542 extern "C" JNIEXPORT int
4543 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4544 void* ucontext, int abort_if_unrecognized);
4546 void signalHandler(int sig, siginfo_t* info, void* uc) {
4547 assert(info != NULL && uc != NULL, "it must be old kernel");
4548 int orig_errno = errno; // Preserve errno value over signal handler.
4549 JVM_handle_linux_signal(sig, info, uc, true);
4550 errno = orig_errno;
4551 }
4554 // This boolean allows users to forward their own non-matching signals
4555 // to JVM_handle_linux_signal, harmlessly.
4556 bool os::Linux::signal_handlers_are_installed = false;
4558 // For signal-chaining
4559 struct sigaction os::Linux::sigact[MAXSIGNUM];
4560 unsigned int os::Linux::sigs = 0;
4561 bool os::Linux::libjsig_is_loaded = false;
4562 typedef struct sigaction *(*get_signal_t)(int);
4563 get_signal_t os::Linux::get_signal_action = NULL;
4565 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4566 struct sigaction *actp = NULL;
4568 if (libjsig_is_loaded) {
4569 // Retrieve the old signal handler from libjsig
4570 actp = (*get_signal_action)(sig);
4571 }
4572 if (actp == NULL) {
4573 // Retrieve the preinstalled signal handler from jvm
4574 actp = get_preinstalled_handler(sig);
4575 }
4577 return actp;
4578 }
4580 static bool call_chained_handler(struct sigaction *actp, int sig,
4581 siginfo_t *siginfo, void *context) {
4582 // Call the old signal handler
4583 if (actp->sa_handler == SIG_DFL) {
4584 // It's more reasonable to let jvm treat it as an unexpected exception
4585 // instead of taking the default action.
4586 return false;
4587 } else if (actp->sa_handler != SIG_IGN) {
4588 if ((actp->sa_flags & SA_NODEFER) == 0) {
4589 // automaticlly block the signal
4590 sigaddset(&(actp->sa_mask), sig);
4591 }
4593 sa_handler_t hand = NULL;
4594 sa_sigaction_t sa = NULL;
4595 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4596 // retrieve the chained handler
4597 if (siginfo_flag_set) {
4598 sa = actp->sa_sigaction;
4599 } else {
4600 hand = actp->sa_handler;
4601 }
4603 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4604 actp->sa_handler = SIG_DFL;
4605 }
4607 // try to honor the signal mask
4608 sigset_t oset;
4609 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4611 // call into the chained handler
4612 if (siginfo_flag_set) {
4613 (*sa)(sig, siginfo, context);
4614 } else {
4615 (*hand)(sig);
4616 }
4618 // restore the signal mask
4619 pthread_sigmask(SIG_SETMASK, &oset, 0);
4620 }
4621 // Tell jvm's signal handler the signal is taken care of.
4622 return true;
4623 }
4625 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4626 bool chained = false;
4627 // signal-chaining
4628 if (UseSignalChaining) {
4629 struct sigaction *actp = get_chained_signal_action(sig);
4630 if (actp != NULL) {
4631 chained = call_chained_handler(actp, sig, siginfo, context);
4632 }
4633 }
4634 return chained;
4635 }
4637 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4638 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4639 return &sigact[sig];
4640 }
4641 return NULL;
4642 }
4644 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4645 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4646 sigact[sig] = oldAct;
4647 sigs |= (unsigned int)1 << sig;
4648 }
4650 // for diagnostic
4651 int os::Linux::sigflags[MAXSIGNUM];
4653 int os::Linux::get_our_sigflags(int sig) {
4654 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4655 return sigflags[sig];
4656 }
4658 void os::Linux::set_our_sigflags(int sig, int flags) {
4659 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4660 sigflags[sig] = flags;
4661 }
4663 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4664 // Check for overwrite.
4665 struct sigaction oldAct;
4666 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4668 void* oldhand = oldAct.sa_sigaction
4669 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4670 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4671 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4672 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4673 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4674 if (AllowUserSignalHandlers || !set_installed) {
4675 // Do not overwrite; user takes responsibility to forward to us.
4676 return;
4677 } else if (UseSignalChaining) {
4678 // save the old handler in jvm
4679 save_preinstalled_handler(sig, oldAct);
4680 // libjsig also interposes the sigaction() call below and saves the
4681 // old sigaction on it own.
4682 } else {
4683 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4684 "%#lx for signal %d.", (long)oldhand, sig));
4685 }
4686 }
4688 struct sigaction sigAct;
4689 sigfillset(&(sigAct.sa_mask));
4690 sigAct.sa_handler = SIG_DFL;
4691 if (!set_installed) {
4692 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4693 } else {
4694 sigAct.sa_sigaction = signalHandler;
4695 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4696 }
4697 // Save flags, which are set by ours
4698 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4699 sigflags[sig] = sigAct.sa_flags;
4701 int ret = sigaction(sig, &sigAct, &oldAct);
4702 assert(ret == 0, "check");
4704 void* oldhand2 = oldAct.sa_sigaction
4705 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4706 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4707 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4708 }
4710 // install signal handlers for signals that HotSpot needs to
4711 // handle in order to support Java-level exception handling.
4713 void os::Linux::install_signal_handlers() {
4714 if (!signal_handlers_are_installed) {
4715 signal_handlers_are_installed = true;
4717 // signal-chaining
4718 typedef void (*signal_setting_t)();
4719 signal_setting_t begin_signal_setting = NULL;
4720 signal_setting_t end_signal_setting = NULL;
4721 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4722 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4723 if (begin_signal_setting != NULL) {
4724 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4725 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4726 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4727 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4728 libjsig_is_loaded = true;
4729 assert(UseSignalChaining, "should enable signal-chaining");
4730 }
4731 if (libjsig_is_loaded) {
4732 // Tell libjsig jvm is setting signal handlers
4733 (*begin_signal_setting)();
4734 }
4736 set_signal_handler(SIGSEGV, true);
4737 set_signal_handler(SIGPIPE, true);
4738 set_signal_handler(SIGBUS, true);
4739 set_signal_handler(SIGILL, true);
4740 set_signal_handler(SIGFPE, true);
4741 #if defined(PPC64)
4742 set_signal_handler(SIGTRAP, true);
4743 #endif
4744 set_signal_handler(SIGXFSZ, true);
4746 if (libjsig_is_loaded) {
4747 // Tell libjsig jvm finishes setting signal handlers
4748 (*end_signal_setting)();
4749 }
4751 // We don't activate signal checker if libjsig is in place, we trust ourselves
4752 // and if UserSignalHandler is installed all bets are off.
4753 // Log that signal checking is off only if -verbose:jni is specified.
4754 if (CheckJNICalls) {
4755 if (libjsig_is_loaded) {
4756 if (PrintJNIResolving) {
4757 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4758 }
4759 check_signals = false;
4760 }
4761 if (AllowUserSignalHandlers) {
4762 if (PrintJNIResolving) {
4763 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4764 }
4765 check_signals = false;
4766 }
4767 }
4768 }
4769 }
4771 // This is the fastest way to get thread cpu time on Linux.
4772 // Returns cpu time (user+sys) for any thread, not only for current.
4773 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4774 // It might work on 2.6.10+ with a special kernel/glibc patch.
4775 // For reference, please, see IEEE Std 1003.1-2004:
4776 // http://www.unix.org/single_unix_specification
4778 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4779 struct timespec tp;
4780 int rc = os::Linux::clock_gettime(clockid, &tp);
4781 assert(rc == 0, "clock_gettime is expected to return 0 code");
4783 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4784 }
4786 /////
4787 // glibc on Linux platform uses non-documented flag
4788 // to indicate, that some special sort of signal
4789 // trampoline is used.
4790 // We will never set this flag, and we should
4791 // ignore this flag in our diagnostic
4792 #ifdef SIGNIFICANT_SIGNAL_MASK
4793 #undef SIGNIFICANT_SIGNAL_MASK
4794 #endif
4795 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4797 static const char* get_signal_handler_name(address handler,
4798 char* buf, int buflen) {
4799 int offset = 0;
4800 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4801 if (found) {
4802 // skip directory names
4803 const char *p1, *p2;
4804 p1 = buf;
4805 size_t len = strlen(os::file_separator());
4806 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4807 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4808 } else {
4809 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4810 }
4811 return buf;
4812 }
4814 static void print_signal_handler(outputStream* st, int sig,
4815 char* buf, size_t buflen) {
4816 struct sigaction sa;
4818 sigaction(sig, NULL, &sa);
4820 // See comment for SIGNIFICANT_SIGNAL_MASK define
4821 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4823 st->print("%s: ", os::exception_name(sig, buf, buflen));
4825 address handler = (sa.sa_flags & SA_SIGINFO)
4826 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4827 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4829 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4830 st->print("SIG_DFL");
4831 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4832 st->print("SIG_IGN");
4833 } else {
4834 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4835 }
4837 st->print(", sa_mask[0]=");
4838 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4840 address rh = VMError::get_resetted_sighandler(sig);
4841 // May be, handler was resetted by VMError?
4842 if(rh != NULL) {
4843 handler = rh;
4844 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4845 }
4847 st->print(", sa_flags=");
4848 os::Posix::print_sa_flags(st, sa.sa_flags);
4850 // Check: is it our handler?
4851 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4852 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4853 // It is our signal handler
4854 // check for flags, reset system-used one!
4855 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4856 st->print(
4857 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4858 os::Linux::get_our_sigflags(sig));
4859 }
4860 }
4861 st->cr();
4862 }
4865 #define DO_SIGNAL_CHECK(sig) \
4866 if (!sigismember(&check_signal_done, sig)) \
4867 os::Linux::check_signal_handler(sig)
4869 // This method is a periodic task to check for misbehaving JNI applications
4870 // under CheckJNI, we can add any periodic checks here
4872 void os::run_periodic_checks() {
4874 if (check_signals == false) return;
4876 // SEGV and BUS if overridden could potentially prevent
4877 // generation of hs*.log in the event of a crash, debugging
4878 // such a case can be very challenging, so we absolutely
4879 // check the following for a good measure:
4880 DO_SIGNAL_CHECK(SIGSEGV);
4881 DO_SIGNAL_CHECK(SIGILL);
4882 DO_SIGNAL_CHECK(SIGFPE);
4883 DO_SIGNAL_CHECK(SIGBUS);
4884 DO_SIGNAL_CHECK(SIGPIPE);
4885 DO_SIGNAL_CHECK(SIGXFSZ);
4886 #if defined(PPC64)
4887 DO_SIGNAL_CHECK(SIGTRAP);
4888 #endif
4890 // ReduceSignalUsage allows the user to override these handlers
4891 // see comments at the very top and jvm_solaris.h
4892 if (!ReduceSignalUsage) {
4893 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4894 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4895 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4896 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4897 }
4899 DO_SIGNAL_CHECK(SR_signum);
4900 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4901 }
4903 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4905 static os_sigaction_t os_sigaction = NULL;
4907 void os::Linux::check_signal_handler(int sig) {
4908 char buf[O_BUFLEN];
4909 address jvmHandler = NULL;
4912 struct sigaction act;
4913 if (os_sigaction == NULL) {
4914 // only trust the default sigaction, in case it has been interposed
4915 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4916 if (os_sigaction == NULL) return;
4917 }
4919 os_sigaction(sig, (struct sigaction*)NULL, &act);
4922 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4924 address thisHandler = (act.sa_flags & SA_SIGINFO)
4925 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4926 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4929 switch(sig) {
4930 case SIGSEGV:
4931 case SIGBUS:
4932 case SIGFPE:
4933 case SIGPIPE:
4934 case SIGILL:
4935 case SIGXFSZ:
4936 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4937 break;
4939 case SHUTDOWN1_SIGNAL:
4940 case SHUTDOWN2_SIGNAL:
4941 case SHUTDOWN3_SIGNAL:
4942 case BREAK_SIGNAL:
4943 jvmHandler = (address)user_handler();
4944 break;
4946 case INTERRUPT_SIGNAL:
4947 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4948 break;
4950 default:
4951 if (sig == SR_signum) {
4952 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4953 } else {
4954 return;
4955 }
4956 break;
4957 }
4959 if (thisHandler != jvmHandler) {
4960 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4961 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4962 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4963 // No need to check this sig any longer
4964 sigaddset(&check_signal_done, sig);
4965 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4966 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4967 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4968 exception_name(sig, buf, O_BUFLEN));
4969 }
4970 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4971 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4972 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4973 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4974 // No need to check this sig any longer
4975 sigaddset(&check_signal_done, sig);
4976 }
4978 // Dump all the signal
4979 if (sigismember(&check_signal_done, sig)) {
4980 print_signal_handlers(tty, buf, O_BUFLEN);
4981 }
4982 }
4984 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4986 extern bool signal_name(int signo, char* buf, size_t len);
4988 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4989 if (0 < exception_code && exception_code <= SIGRTMAX) {
4990 // signal
4991 if (!signal_name(exception_code, buf, size)) {
4992 jio_snprintf(buf, size, "SIG%d", exception_code);
4993 }
4994 return buf;
4995 } else {
4996 return NULL;
4997 }
4998 }
5000 // this is called _before_ most of the global arguments have been parsed
5001 void os::init(void) {
5002 char dummy; /* used to get a guess on initial stack address */
5004 // With LinuxThreads the JavaMain thread pid (primordial thread)
5005 // is different than the pid of the java launcher thread.
5006 // So, on Linux, the launcher thread pid is passed to the VM
5007 // via the sun.java.launcher.pid property.
5008 // Use this property instead of getpid() if it was correctly passed.
5009 // See bug 6351349.
5010 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
5012 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
5014 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5016 init_random(1234567);
5018 ThreadCritical::initialize();
5020 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5021 if (Linux::page_size() == -1) {
5022 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
5023 strerror(errno)));
5024 }
5025 init_page_sizes((size_t) Linux::page_size());
5027 Linux::initialize_system_info();
5029 // _main_thread points to the thread that created/loaded the JVM.
5030 Linux::_main_thread = pthread_self();
5032 Linux::clock_init();
5033 initial_time_count = javaTimeNanos();
5035 // pthread_condattr initialization for monotonic clock
5036 int status;
5037 pthread_condattr_t* _condattr = os::Linux::condAttr();
5038 if ((status = pthread_condattr_init(_condattr)) != 0) {
5039 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
5040 }
5041 // Only set the clock if CLOCK_MONOTONIC is available
5042 if (Linux::supports_monotonic_clock()) {
5043 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
5044 if (status == EINVAL) {
5045 warning("Unable to use monotonic clock with relative timed-waits" \
5046 " - changes to the time-of-day clock may have adverse affects");
5047 } else {
5048 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
5049 }
5050 }
5051 }
5052 // else it defaults to CLOCK_REALTIME
5054 pthread_mutex_init(&dl_mutex, NULL);
5056 // If the pagesize of the VM is greater than 8K determine the appropriate
5057 // number of initial guard pages. The user can change this with the
5058 // command line arguments, if needed.
5059 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
5060 StackYellowPages = 1;
5061 StackRedPages = 1;
5062 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
5063 }
5064 }
5066 // To install functions for atexit system call
5067 extern "C" {
5068 static void perfMemory_exit_helper() {
5069 perfMemory_exit();
5070 }
5071 }
5073 void os::pd_init_container_support() {
5074 OSContainer::init();
5075 }
5077 // this is called _after_ the global arguments have been parsed
5078 jint os::init_2(void)
5079 {
5080 Linux::fast_thread_clock_init();
5082 // Allocate a single page and mark it as readable for safepoint polling
5083 #ifdef OPT_SAFEPOINT
5084 void * p = (void *)(0x10000);
5085 address polling_page = (address) ::mmap(p, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5086 #else
5087 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5088 #endif
5089 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
5091 os::set_polling_page( polling_page );
5093 #ifndef PRODUCT
5094 if(Verbose && PrintMiscellaneous)
5095 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5096 #endif
5098 if (!UseMembar) {
5099 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5100 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
5101 os::set_memory_serialize_page( mem_serialize_page );
5103 #ifndef PRODUCT
5104 if(Verbose && PrintMiscellaneous)
5105 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5106 #endif
5107 }
5109 // initialize suspend/resume support - must do this before signal_sets_init()
5110 if (SR_initialize() != 0) {
5111 perror("SR_initialize failed");
5112 return JNI_ERR;
5113 }
5115 Linux::signal_sets_init();
5116 Linux::install_signal_handlers();
5118 // Check minimum allowable stack size for thread creation and to initialize
5119 // the java system classes, including StackOverflowError - depends on page
5120 // size. Add a page for compiler2 recursion in main thread.
5121 // Add in 2*BytesPerWord times page size to account for VM stack during
5122 // class initialization depending on 32 or 64 bit VM.
5124 /* 2014/1/2 Liao: JDK8 requires larger -Xss option.
5125 * TongWeb cannot run with -Xss192K.
5126 * We are not sure whether this causes errors, so simply print a warning. */
5127 size_t min_stack_allowed_jdk6 = os::Linux::min_stack_allowed;
5128 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
5129 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
5130 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
5132 size_t threadStackSizeInBytes = ThreadStackSize * K;
5133 if (threadStackSizeInBytes != 0 &&
5134 threadStackSizeInBytes < min_stack_allowed_jdk6) {
5135 tty->print_cr("\nThe stack size specified is too small, "
5136 "Specify at least %dk",
5137 os::Linux::min_stack_allowed/ K);
5138 return JNI_ERR;
5139 }
5141 // Make the stack size a multiple of the page size so that
5142 // the yellow/red zones can be guarded.
5143 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5144 vm_page_size()));
5146 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5148 #if defined(IA32)
5149 workaround_expand_exec_shield_cs_limit();
5150 #endif
5152 Linux::libpthread_init();
5153 if (PrintMiscellaneous && (Verbose || WizardMode)) {
5154 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
5155 Linux::glibc_version(), Linux::libpthread_version(),
5156 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
5157 }
5159 if (UseNUMA) {
5160 if (!Linux::libnuma_init()) {
5161 UseNUMA = false;
5162 } else {
5163 if ((Linux::numa_max_node() < 1)) {
5164 // There's only one node(they start from 0), disable NUMA.
5165 UseNUMA = false;
5166 }
5167 }
5168 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5169 // we can make the adaptive lgrp chunk resizing work. If the user specified
5170 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
5171 // disable adaptive resizing.
5172 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5173 if (FLAG_IS_DEFAULT(UseNUMA)) {
5174 UseNUMA = false;
5175 } else {
5176 if (FLAG_IS_DEFAULT(UseLargePages) &&
5177 FLAG_IS_DEFAULT(UseSHM) &&
5178 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
5179 UseLargePages = false;
5180 } else {
5181 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
5182 UseAdaptiveSizePolicy = false;
5183 UseAdaptiveNUMAChunkSizing = false;
5184 }
5185 }
5186 }
5187 if (!UseNUMA && ForceNUMA) {
5188 UseNUMA = true;
5189 }
5190 }
5192 if (MaxFDLimit) {
5193 // set the number of file descriptors to max. print out error
5194 // if getrlimit/setrlimit fails but continue regardless.
5195 struct rlimit nbr_files;
5196 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5197 if (status != 0) {
5198 if (PrintMiscellaneous && (Verbose || WizardMode))
5199 perror("os::init_2 getrlimit failed");
5200 } else {
5201 nbr_files.rlim_cur = nbr_files.rlim_max;
5202 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5203 if (status != 0) {
5204 if (PrintMiscellaneous && (Verbose || WizardMode))
5205 perror("os::init_2 setrlimit failed");
5206 }
5207 }
5208 }
5210 // Initialize lock used to serialize thread creation (see os::create_thread)
5211 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5213 // at-exit methods are called in the reverse order of their registration.
5214 // atexit functions are called on return from main or as a result of a
5215 // call to exit(3C). There can be only 32 of these functions registered
5216 // and atexit() does not set errno.
5218 if (PerfAllowAtExitRegistration) {
5219 // only register atexit functions if PerfAllowAtExitRegistration is set.
5220 // atexit functions can be delayed until process exit time, which
5221 // can be problematic for embedded VM situations. Embedded VMs should
5222 // call DestroyJavaVM() to assure that VM resources are released.
5224 // note: perfMemory_exit_helper atexit function may be removed in
5225 // the future if the appropriate cleanup code can be added to the
5226 // VM_Exit VMOperation's doit method.
5227 if (atexit(perfMemory_exit_helper) != 0) {
5228 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5229 }
5230 }
5232 // initialize thread priority policy
5233 prio_init();
5235 return JNI_OK;
5236 }
5238 // Mark the polling page as unreadable
5239 void os::make_polling_page_unreadable(void) {
5240 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5241 fatal("Could not disable polling page");
5242 };
5244 // Mark the polling page as readable
5245 void os::make_polling_page_readable(void) {
5246 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5247 fatal("Could not enable polling page");
5248 }
5249 };
5251 static int os_cpu_count(const cpu_set_t* cpus) {
5252 int count = 0;
5253 // only look up to the number of configured processors
5254 for (int i = 0; i < os::processor_count(); i++) {
5255 if (CPU_ISSET(i, cpus)) {
5256 count++;
5257 }
5258 }
5259 return count;
5260 }
5262 // Get the current number of available processors for this process.
5263 // This value can change at any time during a process's lifetime.
5264 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5265 // If anything goes wrong we fallback to returning the number of online
5266 // processors - which can be greater than the number available to the process.
5267 int os::Linux::active_processor_count() {
5268 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5269 int cpus_size = sizeof(cpu_set_t);
5270 int cpu_count = 0;
5272 // pid 0 means the current thread - which we have to assume represents the process
5273 if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
5274 cpu_count = os_cpu_count(&cpus);
5275 if (PrintActiveCpus) {
5276 tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5277 }
5278 }
5279 else {
5280 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5281 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5282 "which may exceed available processors", strerror(errno), cpu_count);
5283 }
5285 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5286 return cpu_count;
5287 }
5289 // Determine the active processor count from one of
5290 // three different sources:
5291 //
5292 // 1. User option -XX:ActiveProcessorCount
5293 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5294 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5295 //
5296 // Option 1, if specified, will always override.
5297 // If the cgroup subsystem is active and configured, we
5298 // will return the min of the cgroup and option 2 results.
5299 // This is required since tools, such as numactl, that
5300 // alter cpu affinity do not update cgroup subsystem
5301 // cpuset configuration files.
5302 int os::active_processor_count() {
5303 // User has overridden the number of active processors
5304 if (ActiveProcessorCount > 0) {
5305 if (PrintActiveCpus) {
5306 tty->print_cr("active_processor_count: "
5307 "active processor count set by user : %d",
5308 ActiveProcessorCount);
5309 }
5310 return ActiveProcessorCount;
5311 }
5313 int active_cpus;
5314 if (OSContainer::is_containerized()) {
5315 active_cpus = OSContainer::active_processor_count();
5316 if (PrintActiveCpus) {
5317 tty->print_cr("active_processor_count: determined by OSContainer: %d",
5318 active_cpus);
5319 }
5320 } else {
5321 active_cpus = os::Linux::active_processor_count();
5322 }
5324 return active_cpus;
5325 }
5327 void os::set_native_thread_name(const char *name) {
5328 // Not yet implemented.
5329 return;
5330 }
5332 bool os::distribute_processes(uint length, uint* distribution) {
5333 // Not yet implemented.
5334 return false;
5335 }
5337 bool os::bind_to_processor(uint processor_id) {
5338 // Not yet implemented.
5339 return false;
5340 }
5342 ///
5344 void os::SuspendedThreadTask::internal_do_task() {
5345 if (do_suspend(_thread->osthread())) {
5346 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5347 do_task(context);
5348 do_resume(_thread->osthread());
5349 }
5350 }
5352 class PcFetcher : public os::SuspendedThreadTask {
5353 public:
5354 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5355 ExtendedPC result();
5356 protected:
5357 void do_task(const os::SuspendedThreadTaskContext& context);
5358 private:
5359 ExtendedPC _epc;
5360 };
5362 ExtendedPC PcFetcher::result() {
5363 guarantee(is_done(), "task is not done yet.");
5364 return _epc;
5365 }
5367 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5368 Thread* thread = context.thread();
5369 OSThread* osthread = thread->osthread();
5370 if (osthread->ucontext() != NULL) {
5371 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5372 } else {
5373 // NULL context is unexpected, double-check this is the VMThread
5374 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5375 }
5376 }
5378 // Suspends the target using the signal mechanism and then grabs the PC before
5379 // resuming the target. Used by the flat-profiler only
5380 ExtendedPC os::get_thread_pc(Thread* thread) {
5381 // Make sure that it is called by the watcher for the VMThread
5382 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5383 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5385 PcFetcher fetcher(thread);
5386 fetcher.run();
5387 return fetcher.result();
5388 }
5390 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5391 {
5392 if (is_NPTL()) {
5393 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5394 } else {
5395 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5396 // word back to default 64bit precision if condvar is signaled. Java
5397 // wants 53bit precision. Save and restore current value.
5398 int fpu = get_fpu_control_word();
5399 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5400 set_fpu_control_word(fpu);
5401 return status;
5402 }
5403 }
5405 ////////////////////////////////////////////////////////////////////////////////
5406 // debug support
5408 bool os::find(address addr, outputStream* st) {
5409 Dl_info dlinfo;
5410 memset(&dlinfo, 0, sizeof(dlinfo));
5411 if (dladdr(addr, &dlinfo) != 0) {
5412 st->print(PTR_FORMAT ": ", addr);
5413 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5414 st->print("%s+%#x", dlinfo.dli_sname,
5415 addr - (intptr_t)dlinfo.dli_saddr);
5416 } else if (dlinfo.dli_fbase != NULL) {
5417 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5418 } else {
5419 st->print("<absolute address>");
5420 }
5421 if (dlinfo.dli_fname != NULL) {
5422 st->print(" in %s", dlinfo.dli_fname);
5423 }
5424 if (dlinfo.dli_fbase != NULL) {
5425 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5426 }
5427 st->cr();
5429 if (Verbose) {
5430 // decode some bytes around the PC
5431 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5432 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5433 address lowest = (address) dlinfo.dli_sname;
5434 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5435 if (begin < lowest) begin = lowest;
5436 Dl_info dlinfo2;
5437 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5438 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5439 end = (address) dlinfo2.dli_saddr;
5440 Disassembler::decode(begin, end, st);
5441 }
5442 return true;
5443 }
5444 return false;
5445 }
5447 ////////////////////////////////////////////////////////////////////////////////
5448 // misc
5450 // This does not do anything on Linux. This is basically a hook for being
5451 // able to use structured exception handling (thread-local exception filters)
5452 // on, e.g., Win32.
5453 void
5454 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5455 JavaCallArguments* args, Thread* thread) {
5456 f(value, method, args, thread);
5457 }
5459 void os::print_statistics() {
5460 }
5462 int os::message_box(const char* title, const char* message) {
5463 int i;
5464 fdStream err(defaultStream::error_fd());
5465 for (i = 0; i < 78; i++) err.print_raw("=");
5466 err.cr();
5467 err.print_raw_cr(title);
5468 for (i = 0; i < 78; i++) err.print_raw("-");
5469 err.cr();
5470 err.print_raw_cr(message);
5471 for (i = 0; i < 78; i++) err.print_raw("=");
5472 err.cr();
5474 char buf[16];
5475 // Prevent process from exiting upon "read error" without consuming all CPU
5476 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5478 return buf[0] == 'y' || buf[0] == 'Y';
5479 }
5481 int os::stat(const char *path, struct stat *sbuf) {
5482 char pathbuf[MAX_PATH];
5483 if (strlen(path) > MAX_PATH - 1) {
5484 errno = ENAMETOOLONG;
5485 return -1;
5486 }
5487 os::native_path(strcpy(pathbuf, path));
5488 return ::stat(pathbuf, sbuf);
5489 }
5491 bool os::check_heap(bool force) {
5492 return true;
5493 }
5495 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5496 return ::vsnprintf(buf, count, format, args);
5497 }
5499 // Is a (classpath) directory empty?
5500 bool os::dir_is_empty(const char* path) {
5501 DIR *dir = NULL;
5502 struct dirent *ptr;
5504 dir = opendir(path);
5505 if (dir == NULL) return true;
5507 /* Scan the directory */
5508 bool result = true;
5509 char buf[sizeof(struct dirent) + MAX_PATH];
5510 while (result && (ptr = ::readdir(dir)) != NULL) {
5511 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5512 result = false;
5513 }
5514 }
5515 closedir(dir);
5516 return result;
5517 }
5519 // This code originates from JDK's sysOpen and open64_w
5520 // from src/solaris/hpi/src/system_md.c
5522 #ifndef O_DELETE
5523 #define O_DELETE 0x10000
5524 #endif
5526 // Open a file. Unlink the file immediately after open returns
5527 // if the specified oflag has the O_DELETE flag set.
5528 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5530 int os::open(const char *path, int oflag, int mode) {
5532 if (strlen(path) > MAX_PATH - 1) {
5533 errno = ENAMETOOLONG;
5534 return -1;
5535 }
5536 int fd;
5537 int o_delete = (oflag & O_DELETE);
5538 oflag = oflag & ~O_DELETE;
5540 fd = ::open64(path, oflag, mode);
5541 if (fd == -1) return -1;
5543 //If the open succeeded, the file might still be a directory
5544 {
5545 struct stat64 buf64;
5546 int ret = ::fstat64(fd, &buf64);
5547 int st_mode = buf64.st_mode;
5549 if (ret != -1) {
5550 if ((st_mode & S_IFMT) == S_IFDIR) {
5551 errno = EISDIR;
5552 ::close(fd);
5553 return -1;
5554 }
5555 } else {
5556 ::close(fd);
5557 return -1;
5558 }
5559 }
5561 /*
5562 * All file descriptors that are opened in the JVM and not
5563 * specifically destined for a subprocess should have the
5564 * close-on-exec flag set. If we don't set it, then careless 3rd
5565 * party native code might fork and exec without closing all
5566 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5567 * UNIXProcess.c), and this in turn might:
5568 *
5569 * - cause end-of-file to fail to be detected on some file
5570 * descriptors, resulting in mysterious hangs, or
5571 *
5572 * - might cause an fopen in the subprocess to fail on a system
5573 * suffering from bug 1085341.
5574 *
5575 * (Yes, the default setting of the close-on-exec flag is a Unix
5576 * design flaw)
5577 *
5578 * See:
5579 * 1085341: 32-bit stdio routines should support file descriptors >255
5580 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5581 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5582 */
5583 #ifdef FD_CLOEXEC
5584 {
5585 int flags = ::fcntl(fd, F_GETFD);
5586 if (flags != -1)
5587 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5588 }
5589 #endif
5591 if (o_delete != 0) {
5592 ::unlink(path);
5593 }
5594 return fd;
5595 }
5598 // create binary file, rewriting existing file if required
5599 int os::create_binary_file(const char* path, bool rewrite_existing) {
5600 int oflags = O_WRONLY | O_CREAT;
5601 if (!rewrite_existing) {
5602 oflags |= O_EXCL;
5603 }
5604 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5605 }
5607 // return current position of file pointer
5608 jlong os::current_file_offset(int fd) {
5609 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5610 }
5612 // move file pointer to the specified offset
5613 jlong os::seek_to_file_offset(int fd, jlong offset) {
5614 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5615 }
5617 // This code originates from JDK's sysAvailable
5618 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5620 int os::available(int fd, jlong *bytes) {
5621 jlong cur, end;
5622 int mode;
5623 struct stat64 buf64;
5625 if (::fstat64(fd, &buf64) >= 0) {
5626 mode = buf64.st_mode;
5627 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5628 /*
5629 * XXX: is the following call interruptible? If so, this might
5630 * need to go through the INTERRUPT_IO() wrapper as for other
5631 * blocking, interruptible calls in this file.
5632 */
5633 int n;
5634 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5635 *bytes = n;
5636 return 1;
5637 }
5638 }
5639 }
5640 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5641 return 0;
5642 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5643 return 0;
5644 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5645 return 0;
5646 }
5647 *bytes = end - cur;
5648 return 1;
5649 }
5651 int os::socket_available(int fd, jint *pbytes) {
5652 // Linux doc says EINTR not returned, unlike Solaris
5653 int ret = ::ioctl(fd, FIONREAD, pbytes);
5655 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5656 // is expected to return 0 on failure and 1 on success to the jdk.
5657 return (ret < 0) ? 0 : 1;
5658 }
5660 // Map a block of memory.
5661 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5662 char *addr, size_t bytes, bool read_only,
5663 bool allow_exec) {
5664 int prot;
5665 int flags = MAP_PRIVATE;
5667 if (read_only) {
5668 prot = PROT_READ;
5669 } else {
5670 prot = PROT_READ | PROT_WRITE;
5671 }
5673 if (allow_exec) {
5674 prot |= PROT_EXEC;
5675 }
5677 if (addr != NULL) {
5678 flags |= MAP_FIXED;
5679 }
5681 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5682 fd, file_offset);
5683 if (mapped_address == MAP_FAILED) {
5684 return NULL;
5685 }
5686 return mapped_address;
5687 }
5690 // Remap a block of memory.
5691 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5692 char *addr, size_t bytes, bool read_only,
5693 bool allow_exec) {
5694 // same as map_memory() on this OS
5695 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5696 allow_exec);
5697 }
5700 // Unmap a block of memory.
5701 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5702 return munmap(addr, bytes) == 0;
5703 }
5705 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5707 static clockid_t thread_cpu_clockid(Thread* thread) {
5708 pthread_t tid = thread->osthread()->pthread_id();
5709 clockid_t clockid;
5711 // Get thread clockid
5712 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5713 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5714 return clockid;
5715 }
5717 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5718 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5719 // of a thread.
5720 //
5721 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5722 // the fast estimate available on the platform.
5724 jlong os::current_thread_cpu_time() {
5725 if (os::Linux::supports_fast_thread_cpu_time()) {
5726 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5727 } else {
5728 // return user + sys since the cost is the same
5729 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5730 }
5731 }
5733 jlong os::thread_cpu_time(Thread* thread) {
5734 // consistent with what current_thread_cpu_time() returns
5735 if (os::Linux::supports_fast_thread_cpu_time()) {
5736 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5737 } else {
5738 return slow_thread_cpu_time(thread, true /* user + sys */);
5739 }
5740 }
5742 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5743 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5744 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5745 } else {
5746 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5747 }
5748 }
5750 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5751 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5752 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5753 } else {
5754 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5755 }
5756 }
5758 //
5759 // -1 on error.
5760 //
5762 PRAGMA_DIAG_PUSH
5763 PRAGMA_FORMAT_NONLITERAL_IGNORED
5764 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5765 static bool proc_task_unchecked = true;
5766 static const char *proc_stat_path = "/proc/%d/stat";
5767 pid_t tid = thread->osthread()->thread_id();
5768 char *s;
5769 char stat[2048];
5770 int statlen;
5771 char proc_name[64];
5772 int count;
5773 long sys_time, user_time;
5774 char cdummy;
5775 int idummy;
5776 long ldummy;
5777 FILE *fp;
5779 // The /proc/<tid>/stat aggregates per-process usage on
5780 // new Linux kernels 2.6+ where NPTL is supported.
5781 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5782 // See bug 6328462.
5783 // There possibly can be cases where there is no directory
5784 // /proc/self/task, so we check its availability.
5785 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5786 // This is executed only once
5787 proc_task_unchecked = false;
5788 fp = fopen("/proc/self/task", "r");
5789 if (fp != NULL) {
5790 proc_stat_path = "/proc/self/task/%d/stat";
5791 fclose(fp);
5792 }
5793 }
5795 sprintf(proc_name, proc_stat_path, tid);
5796 fp = fopen(proc_name, "r");
5797 if ( fp == NULL ) return -1;
5798 statlen = fread(stat, 1, 2047, fp);
5799 stat[statlen] = '\0';
5800 fclose(fp);
5802 // Skip pid and the command string. Note that we could be dealing with
5803 // weird command names, e.g. user could decide to rename java launcher
5804 // to "java 1.4.2 :)", then the stat file would look like
5805 // 1234 (java 1.4.2 :)) R ... ...
5806 // We don't really need to know the command string, just find the last
5807 // occurrence of ")" and then start parsing from there. See bug 4726580.
5808 s = strrchr(stat, ')');
5809 if (s == NULL ) return -1;
5811 // Skip blank chars
5812 do s++; while (isspace(*s));
5814 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5815 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5816 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5817 &user_time, &sys_time);
5818 if ( count != 13 ) return -1;
5819 if (user_sys_cpu_time) {
5820 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5821 } else {
5822 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5823 }
5824 }
5825 PRAGMA_DIAG_POP
5827 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5828 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5829 info_ptr->may_skip_backward = false; // elapsed time not wall time
5830 info_ptr->may_skip_forward = false; // elapsed time not wall time
5831 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5832 }
5834 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5835 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5836 info_ptr->may_skip_backward = false; // elapsed time not wall time
5837 info_ptr->may_skip_forward = false; // elapsed time not wall time
5838 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5839 }
5841 bool os::is_thread_cpu_time_supported() {
5842 return true;
5843 }
5845 // System loadavg support. Returns -1 if load average cannot be obtained.
5846 // Linux doesn't yet have a (official) notion of processor sets,
5847 // so just return the system wide load average.
5848 int os::loadavg(double loadavg[], int nelem) {
5849 return ::getloadavg(loadavg, nelem);
5850 }
5852 void os::pause() {
5853 char filename[MAX_PATH];
5854 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5855 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5856 } else {
5857 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5858 }
5860 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5861 if (fd != -1) {
5862 struct stat buf;
5863 ::close(fd);
5864 while (::stat(filename, &buf) == 0) {
5865 (void)::poll(NULL, 0, 100);
5866 }
5867 } else {
5868 jio_fprintf(stderr,
5869 "Could not open pause file '%s', continuing immediately.\n", filename);
5870 }
5871 }
5874 // Refer to the comments in os_solaris.cpp park-unpark.
5875 //
5876 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5877 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5878 // For specifics regarding the bug see GLIBC BUGID 261237 :
5879 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5880 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5881 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5882 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5883 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5884 // and monitorenter when we're using 1-0 locking. All those operations may result in
5885 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5886 // of libpthread avoids the problem, but isn't practical.
5887 //
5888 // Possible remedies:
5889 //
5890 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5891 // This is palliative and probabilistic, however. If the thread is preempted
5892 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5893 // than the minimum period may have passed, and the abstime may be stale (in the
5894 // past) resultin in a hang. Using this technique reduces the odds of a hang
5895 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5896 //
5897 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5898 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5899 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5900 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5901 // thread.
5902 //
5903 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5904 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5905 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5906 // This also works well. In fact it avoids kernel-level scalability impediments
5907 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5908 // timers in a graceful fashion.
5909 //
5910 // 4. When the abstime value is in the past it appears that control returns
5911 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5912 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5913 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5914 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5915 // It may be possible to avoid reinitialization by checking the return
5916 // value from pthread_cond_timedwait(). In addition to reinitializing the
5917 // condvar we must establish the invariant that cond_signal() is only called
5918 // within critical sections protected by the adjunct mutex. This prevents
5919 // cond_signal() from "seeing" a condvar that's in the midst of being
5920 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5921 // desirable signal-after-unlock optimization that avoids futile context switching.
5922 //
5923 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5924 // structure when a condvar is used or initialized. cond_destroy() would
5925 // release the helper structure. Our reinitialize-after-timedwait fix
5926 // put excessive stress on malloc/free and locks protecting the c-heap.
5927 //
5928 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5929 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5930 // and only enabling the work-around for vulnerable environments.
5932 // utility to compute the abstime argument to timedwait:
5933 // millis is the relative timeout time
5934 // abstime will be the absolute timeout time
5935 // TODO: replace compute_abstime() with unpackTime()
5937 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5938 if (millis < 0) millis = 0;
5940 jlong seconds = millis / 1000;
5941 millis %= 1000;
5942 if (seconds > 50000000) { // see man cond_timedwait(3T)
5943 seconds = 50000000;
5944 }
5946 if (os::Linux::supports_monotonic_clock()) {
5947 struct timespec now;
5948 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5949 assert_status(status == 0, status, "clock_gettime");
5950 abstime->tv_sec = now.tv_sec + seconds;
5951 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5952 if (nanos >= NANOSECS_PER_SEC) {
5953 abstime->tv_sec += 1;
5954 nanos -= NANOSECS_PER_SEC;
5955 }
5956 abstime->tv_nsec = nanos;
5957 } else {
5958 struct timeval now;
5959 int status = gettimeofday(&now, NULL);
5960 assert(status == 0, "gettimeofday");
5961 abstime->tv_sec = now.tv_sec + seconds;
5962 long usec = now.tv_usec + millis * 1000;
5963 if (usec >= 1000000) {
5964 abstime->tv_sec += 1;
5965 usec -= 1000000;
5966 }
5967 abstime->tv_nsec = usec * 1000;
5968 }
5969 return abstime;
5970 }
5973 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5974 // Conceptually TryPark() should be equivalent to park(0).
5976 int os::PlatformEvent::TryPark() {
5977 for (;;) {
5978 const int v = _Event ;
5979 guarantee ((v == 0) || (v == 1), "invariant") ;
5980 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5981 }
5982 }
5984 void os::PlatformEvent::park() { // AKA "down()"
5985 // Invariant: Only the thread associated with the Event/PlatformEvent
5986 // may call park().
5987 // TODO: assert that _Assoc != NULL or _Assoc == Self
5988 int v ;
5989 for (;;) {
5990 v = _Event ;
5991 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5992 }
5993 guarantee (v >= 0, "invariant") ;
5994 if (v == 0) {
5995 // Do this the hard way by blocking ...
5996 int status = pthread_mutex_lock(_mutex);
5997 assert_status(status == 0, status, "mutex_lock");
5998 guarantee (_nParked == 0, "invariant") ;
5999 ++ _nParked ;
6000 while (_Event < 0) {
6001 status = pthread_cond_wait(_cond, _mutex);
6002 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6003 // Treat this the same as if the wait was interrupted
6004 if (status == ETIME) { status = EINTR; }
6005 assert_status(status == 0 || status == EINTR, status, "cond_wait");
6006 }
6007 -- _nParked ;
6009 _Event = 0 ;
6010 status = pthread_mutex_unlock(_mutex);
6011 assert_status(status == 0, status, "mutex_unlock");
6012 // Paranoia to ensure our locked and lock-free paths interact
6013 // correctly with each other.
6014 OrderAccess::fence();
6015 }
6016 guarantee (_Event >= 0, "invariant") ;
6017 }
6019 int os::PlatformEvent::park(jlong millis) {
6020 guarantee (_nParked == 0, "invariant") ;
6022 int v ;
6023 for (;;) {
6024 v = _Event ;
6025 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6026 }
6027 guarantee (v >= 0, "invariant") ;
6028 if (v != 0) return OS_OK ;
6030 // We do this the hard way, by blocking the thread.
6031 // Consider enforcing a minimum timeout value.
6032 struct timespec abst;
6033 compute_abstime(&abst, millis);
6035 int ret = OS_TIMEOUT;
6036 int status = pthread_mutex_lock(_mutex);
6037 assert_status(status == 0, status, "mutex_lock");
6038 guarantee (_nParked == 0, "invariant") ;
6039 ++_nParked ;
6041 // Object.wait(timo) will return because of
6042 // (a) notification
6043 // (b) timeout
6044 // (c) thread.interrupt
6045 //
6046 // Thread.interrupt and object.notify{All} both call Event::set.
6047 // That is, we treat thread.interrupt as a special case of notification.
6048 // The underlying Solaris implementation, cond_timedwait, admits
6049 // spurious/premature wakeups, but the JLS/JVM spec prevents the
6050 // JVM from making those visible to Java code. As such, we must
6051 // filter out spurious wakeups. We assume all ETIME returns are valid.
6052 //
6053 // TODO: properly differentiate simultaneous notify+interrupt.
6054 // In that case, we should propagate the notify to another waiter.
6056 while (_Event < 0) {
6057 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
6058 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6059 pthread_cond_destroy (_cond);
6060 pthread_cond_init (_cond, os::Linux::condAttr()) ;
6061 }
6062 assert_status(status == 0 || status == EINTR ||
6063 status == ETIME || status == ETIMEDOUT,
6064 status, "cond_timedwait");
6065 if (!FilterSpuriousWakeups) break ; // previous semantics
6066 if (status == ETIME || status == ETIMEDOUT) break ;
6067 // We consume and ignore EINTR and spurious wakeups.
6068 }
6069 --_nParked ;
6070 if (_Event >= 0) {
6071 ret = OS_OK;
6072 }
6073 _Event = 0 ;
6074 status = pthread_mutex_unlock(_mutex);
6075 assert_status(status == 0, status, "mutex_unlock");
6076 assert (_nParked == 0, "invariant") ;
6077 // Paranoia to ensure our locked and lock-free paths interact
6078 // correctly with each other.
6079 OrderAccess::fence();
6080 return ret;
6081 }
6083 void os::PlatformEvent::unpark() {
6084 // Transitions for _Event:
6085 // 0 :=> 1
6086 // 1 :=> 1
6087 // -1 :=> either 0 or 1; must signal target thread
6088 // That is, we can safely transition _Event from -1 to either
6089 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
6090 // unpark() calls.
6091 // See also: "Semaphores in Plan 9" by Mullender & Cox
6092 //
6093 // Note: Forcing a transition from "-1" to "1" on an unpark() means
6094 // that it will take two back-to-back park() calls for the owning
6095 // thread to block. This has the benefit of forcing a spurious return
6096 // from the first park() call after an unpark() call which will help
6097 // shake out uses of park() and unpark() without condition variables.
6099 if (Atomic::xchg(1, &_Event) >= 0) return;
6101 // Wait for the thread associated with the event to vacate
6102 int status = pthread_mutex_lock(_mutex);
6103 assert_status(status == 0, status, "mutex_lock");
6104 int AnyWaiters = _nParked;
6105 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6106 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
6107 AnyWaiters = 0;
6108 pthread_cond_signal(_cond);
6109 }
6110 status = pthread_mutex_unlock(_mutex);
6111 assert_status(status == 0, status, "mutex_unlock");
6112 if (AnyWaiters != 0) {
6113 status = pthread_cond_signal(_cond);
6114 assert_status(status == 0, status, "cond_signal");
6115 }
6117 // Note that we signal() _after dropping the lock for "immortal" Events.
6118 // This is safe and avoids a common class of futile wakeups. In rare
6119 // circumstances this can cause a thread to return prematurely from
6120 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
6121 // simply re-test the condition and re-park itself.
6122 }
6125 // JSR166
6126 // -------------------------------------------------------
6128 /*
6129 * The solaris and linux implementations of park/unpark are fairly
6130 * conservative for now, but can be improved. They currently use a
6131 * mutex/condvar pair, plus a a count.
6132 * Park decrements count if > 0, else does a condvar wait. Unpark
6133 * sets count to 1 and signals condvar. Only one thread ever waits
6134 * on the condvar. Contention seen when trying to park implies that someone
6135 * is unparking you, so don't wait. And spurious returns are fine, so there
6136 * is no need to track notifications.
6137 */
6139 /*
6140 * This code is common to linux and solaris and will be moved to a
6141 * common place in dolphin.
6142 *
6143 * The passed in time value is either a relative time in nanoseconds
6144 * or an absolute time in milliseconds. Either way it has to be unpacked
6145 * into suitable seconds and nanoseconds components and stored in the
6146 * given timespec structure.
6147 * Given time is a 64-bit value and the time_t used in the timespec is only
6148 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6149 * overflow if times way in the future are given. Further on Solaris versions
6150 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6151 * number of seconds, in abstime, is less than current_time + 100,000,000.
6152 * As it will be 28 years before "now + 100000000" will overflow we can
6153 * ignore overflow and just impose a hard-limit on seconds using the value
6154 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6155 * years from "now".
6156 */
6158 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6159 assert (time > 0, "convertTime");
6160 time_t max_secs = 0;
6162 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
6163 struct timeval now;
6164 int status = gettimeofday(&now, NULL);
6165 assert(status == 0, "gettimeofday");
6167 max_secs = now.tv_sec + MAX_SECS;
6169 if (isAbsolute) {
6170 jlong secs = time / 1000;
6171 if (secs > max_secs) {
6172 absTime->tv_sec = max_secs;
6173 } else {
6174 absTime->tv_sec = secs;
6175 }
6176 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6177 } else {
6178 jlong secs = time / NANOSECS_PER_SEC;
6179 if (secs >= MAX_SECS) {
6180 absTime->tv_sec = max_secs;
6181 absTime->tv_nsec = 0;
6182 } else {
6183 absTime->tv_sec = now.tv_sec + secs;
6184 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6185 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6186 absTime->tv_nsec -= NANOSECS_PER_SEC;
6187 ++absTime->tv_sec; // note: this must be <= max_secs
6188 }
6189 }
6190 }
6191 } else {
6192 // must be relative using monotonic clock
6193 struct timespec now;
6194 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6195 assert_status(status == 0, status, "clock_gettime");
6196 max_secs = now.tv_sec + MAX_SECS;
6197 jlong secs = time / NANOSECS_PER_SEC;
6198 if (secs >= MAX_SECS) {
6199 absTime->tv_sec = max_secs;
6200 absTime->tv_nsec = 0;
6201 } else {
6202 absTime->tv_sec = now.tv_sec + secs;
6203 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
6204 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6205 absTime->tv_nsec -= NANOSECS_PER_SEC;
6206 ++absTime->tv_sec; // note: this must be <= max_secs
6207 }
6208 }
6209 }
6210 assert(absTime->tv_sec >= 0, "tv_sec < 0");
6211 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6212 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6213 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6214 }
6216 void Parker::park(bool isAbsolute, jlong time) {
6217 // Ideally we'd do something useful while spinning, such
6218 // as calling unpackTime().
6220 // Optional fast-path check:
6221 // Return immediately if a permit is available.
6222 // We depend on Atomic::xchg() having full barrier semantics
6223 // since we are doing a lock-free update to _counter.
6224 if (Atomic::xchg(0, &_counter) > 0) return;
6226 Thread* thread = Thread::current();
6227 assert(thread->is_Java_thread(), "Must be JavaThread");
6228 JavaThread *jt = (JavaThread *)thread;
6230 // Optional optimization -- avoid state transitions if there's an interrupt pending.
6231 // Check interrupt before trying to wait
6232 if (Thread::is_interrupted(thread, false)) {
6233 return;
6234 }
6236 // Next, demultiplex/decode time arguments
6237 timespec absTime;
6238 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6239 return;
6240 }
6241 if (time > 0) {
6242 unpackTime(&absTime, isAbsolute, time);
6243 }
6246 // Enter safepoint region
6247 // Beware of deadlocks such as 6317397.
6248 // The per-thread Parker:: mutex is a classic leaf-lock.
6249 // In particular a thread must never block on the Threads_lock while
6250 // holding the Parker:: mutex. If safepoints are pending both the
6251 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6252 ThreadBlockInVM tbivm(jt);
6254 // Don't wait if cannot get lock since interference arises from
6255 // unblocking. Also. check interrupt before trying wait
6256 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6257 return;
6258 }
6260 int status ;
6261 if (_counter > 0) { // no wait needed
6262 _counter = 0;
6263 status = pthread_mutex_unlock(_mutex);
6264 assert (status == 0, "invariant") ;
6265 // Paranoia to ensure our locked and lock-free paths interact
6266 // correctly with each other and Java-level accesses.
6267 OrderAccess::fence();
6268 return;
6269 }
6271 #ifdef ASSERT
6272 // Don't catch signals while blocked; let the running threads have the signals.
6273 // (This allows a debugger to break into the running thread.)
6274 sigset_t oldsigs;
6275 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6276 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6277 #endif
6279 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6280 jt->set_suspend_equivalent();
6281 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6283 assert(_cur_index == -1, "invariant");
6284 if (time == 0) {
6285 _cur_index = REL_INDEX; // arbitrary choice when not timed
6286 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6287 } else {
6288 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6289 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6290 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6291 pthread_cond_destroy (&_cond[_cur_index]) ;
6292 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6293 }
6294 }
6295 _cur_index = -1;
6296 assert_status(status == 0 || status == EINTR ||
6297 status == ETIME || status == ETIMEDOUT,
6298 status, "cond_timedwait");
6300 #ifdef ASSERT
6301 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6302 #endif
6304 _counter = 0 ;
6305 status = pthread_mutex_unlock(_mutex) ;
6306 assert_status(status == 0, status, "invariant") ;
6307 // Paranoia to ensure our locked and lock-free paths interact
6308 // correctly with each other and Java-level accesses.
6309 OrderAccess::fence();
6311 // If externally suspended while waiting, re-suspend
6312 if (jt->handle_special_suspend_equivalent_condition()) {
6313 jt->java_suspend_self();
6314 }
6315 }
6317 void Parker::unpark() {
6318 int s, status ;
6319 status = pthread_mutex_lock(_mutex);
6320 assert (status == 0, "invariant") ;
6321 s = _counter;
6322 _counter = 1;
6323 if (s < 1) {
6324 // thread might be parked
6325 if (_cur_index != -1) {
6326 // thread is definitely parked
6327 if (WorkAroundNPTLTimedWaitHang) {
6328 status = pthread_cond_signal (&_cond[_cur_index]);
6329 assert (status == 0, "invariant");
6330 status = pthread_mutex_unlock(_mutex);
6331 assert (status == 0, "invariant");
6332 } else {
6333 // must capture correct index before unlocking
6334 int index = _cur_index;
6335 status = pthread_mutex_unlock(_mutex);
6336 assert (status == 0, "invariant");
6337 status = pthread_cond_signal (&_cond[index]);
6338 assert (status == 0, "invariant");
6339 }
6340 } else {
6341 pthread_mutex_unlock(_mutex);
6342 assert (status == 0, "invariant") ;
6343 }
6344 } else {
6345 pthread_mutex_unlock(_mutex);
6346 assert (status == 0, "invariant") ;
6347 }
6348 }
6351 extern char** environ;
6353 // Run the specified command in a separate process. Return its exit value,
6354 // or -1 on failure (e.g. can't fork a new process).
6355 // Unlike system(), this function can be called from signal handler. It
6356 // doesn't block SIGINT et al.
6357 int os::fork_and_exec(char* cmd) {
6358 const char * argv[4] = {"sh", "-c", cmd, NULL};
6360 pid_t pid = fork();
6362 if (pid < 0) {
6363 // fork failed
6364 return -1;
6366 } else if (pid == 0) {
6367 // child process
6369 execve("/bin/sh", (char* const*)argv, environ);
6371 // execve failed
6372 _exit(-1);
6374 } else {
6375 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6376 // care about the actual exit code, for now.
6378 int status;
6380 // Wait for the child process to exit. This returns immediately if
6381 // the child has already exited. */
6382 while (waitpid(pid, &status, 0) < 0) {
6383 switch (errno) {
6384 case ECHILD: return 0;
6385 case EINTR: break;
6386 default: return -1;
6387 }
6388 }
6390 if (WIFEXITED(status)) {
6391 // The child exited normally; get its exit code.
6392 return WEXITSTATUS(status);
6393 } else if (WIFSIGNALED(status)) {
6394 // The child exited because of a signal
6395 // The best value to return is 0x80 + signal number,
6396 // because that is what all Unix shells do, and because
6397 // it allows callers to distinguish between process exit and
6398 // process death by signal.
6399 return 0x80 + WTERMSIG(status);
6400 } else {
6401 // Unknown exit code; pass it through
6402 return status;
6403 }
6404 }
6405 }
6407 // is_headless_jre()
6408 //
6409 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6410 // in order to report if we are running in a headless jre
6411 //
6412 // Since JDK8 xawt/libmawt.so was moved into the same directory
6413 // as libawt.so, and renamed libawt_xawt.so
6414 //
6415 bool os::is_headless_jre() {
6416 struct stat statbuf;
6417 char buf[MAXPATHLEN];
6418 char libmawtpath[MAXPATHLEN];
6419 const char *xawtstr = "/xawt/libmawt.so";
6420 const char *new_xawtstr = "/libawt_xawt.so";
6421 char *p;
6423 // Get path to libjvm.so
6424 os::jvm_path(buf, sizeof(buf));
6426 // Get rid of libjvm.so
6427 p = strrchr(buf, '/');
6428 if (p == NULL) return false;
6429 else *p = '\0';
6431 // Get rid of client or server
6432 p = strrchr(buf, '/');
6433 if (p == NULL) return false;
6434 else *p = '\0';
6436 // check xawt/libmawt.so
6437 strcpy(libmawtpath, buf);
6438 strcat(libmawtpath, xawtstr);
6439 if (::stat(libmawtpath, &statbuf) == 0) return false;
6441 // check libawt_xawt.so
6442 strcpy(libmawtpath, buf);
6443 strcat(libmawtpath, new_xawtstr);
6444 if (::stat(libmawtpath, &statbuf) == 0) return false;
6446 return true;
6447 }
6449 // Get the default path to the core file
6450 // Returns the length of the string
6451 int os::get_core_path(char* buffer, size_t bufferSize) {
6452 const char* p = get_current_directory(buffer, bufferSize);
6454 if (p == NULL) {
6455 assert(p != NULL, "failed to get current directory");
6456 return 0;
6457 }
6459 return strlen(buffer);
6460 }
6462 /////////////// Unit tests ///////////////
6464 #ifndef PRODUCT
6466 #define test_log(...) \
6467 do {\
6468 if (VerboseInternalVMTests) { \
6469 tty->print_cr(__VA_ARGS__); \
6470 tty->flush(); \
6471 }\
6472 } while (false)
6474 class TestReserveMemorySpecial : AllStatic {
6475 public:
6476 static void small_page_write(void* addr, size_t size) {
6477 size_t page_size = os::vm_page_size();
6479 char* end = (char*)addr + size;
6480 for (char* p = (char*)addr; p < end; p += page_size) {
6481 *p = 1;
6482 }
6483 }
6485 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6486 if (!UseHugeTLBFS) {
6487 return;
6488 }
6490 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6492 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6494 if (addr != NULL) {
6495 small_page_write(addr, size);
6497 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6498 }
6499 }
6501 static void test_reserve_memory_special_huge_tlbfs_only() {
6502 if (!UseHugeTLBFS) {
6503 return;
6504 }
6506 size_t lp = os::large_page_size();
6508 for (size_t size = lp; size <= lp * 10; size += lp) {
6509 test_reserve_memory_special_huge_tlbfs_only(size);
6510 }
6511 }
6513 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6514 size_t lp = os::large_page_size();
6515 size_t ag = os::vm_allocation_granularity();
6517 // sizes to test
6518 const size_t sizes[] = {
6519 lp, lp + ag, lp + lp / 2, lp * 2,
6520 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6521 lp * 10, lp * 10 + lp / 2
6522 };
6523 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6525 // For each size/alignment combination, we test three scenarios:
6526 // 1) with req_addr == NULL
6527 // 2) with a non-null req_addr at which we expect to successfully allocate
6528 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6529 // expect the allocation to either fail or to ignore req_addr
6531 // Pre-allocate two areas; they shall be as large as the largest allocation
6532 // and aligned to the largest alignment we will be testing.
6533 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6534 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6535 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6536 -1, 0);
6537 assert(mapping1 != MAP_FAILED, "should work");
6539 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6540 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6541 -1, 0);
6542 assert(mapping2 != MAP_FAILED, "should work");
6544 // Unmap the first mapping, but leave the second mapping intact: the first
6545 // mapping will serve as a value for a "good" req_addr (case 2). The second
6546 // mapping, still intact, as "bad" req_addr (case 3).
6547 ::munmap(mapping1, mapping_size);
6549 // Case 1
6550 test_log("%s, req_addr NULL:", __FUNCTION__);
6551 test_log("size align result");
6553 for (int i = 0; i < num_sizes; i++) {
6554 const size_t size = sizes[i];
6555 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6556 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6557 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6558 size, alignment, p, (p != NULL ? "" : "(failed)"));
6559 if (p != NULL) {
6560 assert(is_ptr_aligned(p, alignment), "must be");
6561 small_page_write(p, size);
6562 os::Linux::release_memory_special_huge_tlbfs(p, size);
6563 }
6564 }
6565 }
6567 // Case 2
6568 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6569 test_log("size align req_addr result");
6571 for (int i = 0; i < num_sizes; i++) {
6572 const size_t size = sizes[i];
6573 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6574 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6575 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6576 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6577 size, alignment, req_addr, p,
6578 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6579 if (p != NULL) {
6580 assert(p == req_addr, "must be");
6581 small_page_write(p, size);
6582 os::Linux::release_memory_special_huge_tlbfs(p, size);
6583 }
6584 }
6585 }
6587 // Case 3
6588 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6589 test_log("size align req_addr result");
6591 for (int i = 0; i < num_sizes; i++) {
6592 const size_t size = sizes[i];
6593 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6594 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6595 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6596 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6597 size, alignment, req_addr, p,
6598 ((p != NULL ? "" : "(failed)")));
6599 // as the area around req_addr contains already existing mappings, the API should always
6600 // return NULL (as per contract, it cannot return another address)
6601 assert(p == NULL, "must be");
6602 }
6603 }
6605 ::munmap(mapping2, mapping_size);
6607 }
6609 static void test_reserve_memory_special_huge_tlbfs() {
6610 if (!UseHugeTLBFS) {
6611 return;
6612 }
6614 test_reserve_memory_special_huge_tlbfs_only();
6615 test_reserve_memory_special_huge_tlbfs_mixed();
6616 }
6618 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6619 if (!UseSHM) {
6620 return;
6621 }
6623 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6625 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6627 if (addr != NULL) {
6628 assert(is_ptr_aligned(addr, alignment), "Check");
6629 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6631 small_page_write(addr, size);
6633 os::Linux::release_memory_special_shm(addr, size);
6634 }
6635 }
6637 static void test_reserve_memory_special_shm() {
6638 size_t lp = os::large_page_size();
6639 size_t ag = os::vm_allocation_granularity();
6641 for (size_t size = ag; size < lp * 3; size += ag) {
6642 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6643 test_reserve_memory_special_shm(size, alignment);
6644 }
6645 }
6646 }
6648 static void test() {
6649 test_reserve_memory_special_huge_tlbfs();
6650 test_reserve_memory_special_shm();
6651 }
6652 };
6654 void TestReserveMemorySpecial_test() {
6655 TestReserveMemorySpecial::test();
6656 }
6658 #endif