Wed, 03 Jul 2019 20:42:37 +0800
Merge
1 /*
2 * Copyright (c) 1999, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_share_linux.hpp"
40 #include "osContainer_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/interfaceSupport.hpp"
48 #include "runtime/init.hpp"
49 #include "runtime/java.hpp"
50 #include "runtime/javaCalls.hpp"
51 #include "runtime/mutexLocker.hpp"
52 #include "runtime/objectMonitor.hpp"
53 #include "runtime/orderAccess.inline.hpp"
54 #include "runtime/osThread.hpp"
55 #include "runtime/perfMemory.hpp"
56 #include "runtime/sharedRuntime.hpp"
57 #include "runtime/statSampler.hpp"
58 #include "runtime/stubRoutines.hpp"
59 #include "runtime/thread.inline.hpp"
60 #include "runtime/threadCritical.hpp"
61 #include "runtime/timer.hpp"
62 #include "services/attachListener.hpp"
63 #include "services/memTracker.hpp"
64 #include "services/runtimeService.hpp"
65 #include "utilities/decoder.hpp"
66 #include "utilities/defaultStream.hpp"
67 #include "utilities/events.hpp"
68 #include "utilities/elfFile.hpp"
69 #include "utilities/growableArray.hpp"
70 #include "utilities/vmError.hpp"
72 // put OS-includes here
73 # include <sys/types.h>
74 # include <sys/mman.h>
75 # include <sys/stat.h>
76 # include <sys/select.h>
77 # include <pthread.h>
78 # include <signal.h>
79 # include <errno.h>
80 # include <dlfcn.h>
81 # include <stdio.h>
82 # include <unistd.h>
83 # include <sys/resource.h>
84 # include <pthread.h>
85 # include <sys/stat.h>
86 # include <sys/time.h>
87 # include <sys/times.h>
88 # include <sys/utsname.h>
89 # include <sys/socket.h>
90 # include <sys/wait.h>
91 # include <pwd.h>
92 # include <poll.h>
93 # include <semaphore.h>
94 # include <fcntl.h>
95 # include <string.h>
96 # include <syscall.h>
97 # include <sys/sysinfo.h>
98 # include <gnu/libc-version.h>
99 # include <sys/ipc.h>
100 # include <sys/shm.h>
101 # include <link.h>
102 # include <stdint.h>
103 # include <inttypes.h>
104 # include <sys/ioctl.h>
106 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
108 #ifndef _GNU_SOURCE
109 #define _GNU_SOURCE
110 #include <sched.h>
111 #undef _GNU_SOURCE
112 #else
113 #include <sched.h>
114 #endif
116 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
117 // getrusage() is prepared to handle the associated failure.
118 #ifndef RUSAGE_THREAD
119 #define RUSAGE_THREAD (1) /* only the calling thread */
120 #endif
122 #define MAX_PATH (2 * K)
124 #define MAX_SECS 100000000
126 // for timer info max values which include all bits
127 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
129 #define LARGEPAGES_BIT (1 << 6)
130 ////////////////////////////////////////////////////////////////////////////////
131 // global variables
132 julong os::Linux::_physical_memory = 0;
134 address os::Linux::_initial_thread_stack_bottom = NULL;
135 uintptr_t os::Linux::_initial_thread_stack_size = 0;
137 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
138 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
139 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 void os::Linux::expand_stack_to(address bottom) {
728 _expand_stack_to(bottom);
729 }
731 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
732 assert(t!=NULL, "just checking");
733 assert(t->osthread()->expanding_stack(), "expand should be set");
734 assert(t->stack_base() != NULL, "stack_base was not initialized");
736 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
737 sigset_t mask_all, old_sigset;
738 sigfillset(&mask_all);
739 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
740 _expand_stack_to(addr);
741 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
742 return true;
743 }
744 return false;
745 }
747 //////////////////////////////////////////////////////////////////////////////
748 // create new thread
750 static address highest_vm_reserved_address();
752 // check if it's safe to start a new thread
753 static bool _thread_safety_check(Thread* thread) {
754 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
755 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
756 // Heap is mmap'ed at lower end of memory space. Thread stacks are
757 // allocated (MAP_FIXED) from high address space. Every thread stack
758 // occupies a fixed size slot (usually 2Mbytes, but user can change
759 // it to other values if they rebuild LinuxThreads).
760 //
761 // Problem with MAP_FIXED is that mmap() can still succeed even part of
762 // the memory region has already been mmap'ed. That means if we have too
763 // many threads and/or very large heap, eventually thread stack will
764 // collide with heap.
765 //
766 // Here we try to prevent heap/stack collision by comparing current
767 // stack bottom with the highest address that has been mmap'ed by JVM
768 // plus a safety margin for memory maps created by native code.
769 //
770 // This feature can be disabled by setting ThreadSafetyMargin to 0
771 //
772 if (ThreadSafetyMargin > 0) {
773 address stack_bottom = os::current_stack_base() - os::current_stack_size();
775 // not safe if our stack extends below the safety margin
776 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
777 } else {
778 return true;
779 }
780 } else {
781 // Floating stack LinuxThreads or NPTL:
782 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
783 // there's not enough space left, pthread_create() will fail. If we come
784 // here, that means enough space has been reserved for stack.
785 return true;
786 }
787 }
789 // Thread start routine for all newly created threads
790 static void *java_start(Thread *thread) {
791 // Try to randomize the cache line index of hot stack frames.
792 // This helps when threads of the same stack traces evict each other's
793 // cache lines. The threads can be either from the same JVM instance, or
794 // from different JVM instances. The benefit is especially true for
795 // processors with hyperthreading technology.
796 static int counter = 0;
797 int pid = os::current_process_id();
798 alloca(((pid ^ counter++) & 7) * 128);
800 ThreadLocalStorage::set_thread(thread);
802 OSThread* osthread = thread->osthread();
803 Monitor* sync = osthread->startThread_lock();
805 // non floating stack LinuxThreads needs extra check, see above
806 if (!_thread_safety_check(thread)) {
807 // notify parent thread
808 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
809 osthread->set_state(ZOMBIE);
810 sync->notify_all();
811 return NULL;
812 }
814 // thread_id is kernel thread id (similar to Solaris LWP id)
815 osthread->set_thread_id(os::Linux::gettid());
817 if (UseNUMA) {
818 int lgrp_id = os::numa_get_group_id();
819 if (lgrp_id != -1) {
820 thread->set_lgrp_id(lgrp_id);
821 }
822 }
823 // initialize signal mask for this thread
824 os::Linux::hotspot_sigmask(thread);
826 // initialize floating point control register
827 os::Linux::init_thread_fpu_state();
829 // handshaking with parent thread
830 {
831 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
833 // notify parent thread
834 osthread->set_state(INITIALIZED);
835 sync->notify_all();
837 // wait until os::start_thread()
838 while (osthread->get_state() == INITIALIZED) {
839 sync->wait(Mutex::_no_safepoint_check_flag);
840 }
841 }
843 // call one more level start routine
844 thread->run();
846 return 0;
847 }
849 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
850 assert(thread->osthread() == NULL, "caller responsible");
852 // Allocate the OSThread object
853 OSThread* osthread = new OSThread(NULL, NULL);
854 if (osthread == NULL) {
855 return false;
856 }
858 // set the correct thread state
859 osthread->set_thread_type(thr_type);
861 // Initial state is ALLOCATED but not INITIALIZED
862 osthread->set_state(ALLOCATED);
864 thread->set_osthread(osthread);
866 // init thread attributes
867 pthread_attr_t attr;
868 pthread_attr_init(&attr);
869 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
871 // stack size
872 if (os::Linux::supports_variable_stack_size()) {
873 // calculate stack size if it's not specified by caller
874 if (stack_size == 0) {
875 stack_size = os::Linux::default_stack_size(thr_type);
877 switch (thr_type) {
878 case os::java_thread:
879 // Java threads use ThreadStackSize which default value can be
880 // changed with the flag -Xss
881 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
882 stack_size = JavaThread::stack_size_at_create();
883 break;
884 case os::compiler_thread:
885 if (CompilerThreadStackSize > 0) {
886 stack_size = (size_t)(CompilerThreadStackSize * K);
887 break;
888 } // else fall through:
889 // use VMThreadStackSize if CompilerThreadStackSize is not defined
890 case os::vm_thread:
891 case os::pgc_thread:
892 case os::cgc_thread:
893 case os::watcher_thread:
894 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
895 break;
896 }
897 }
899 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
900 pthread_attr_setstacksize(&attr, stack_size);
901 } else {
902 // let pthread_create() pick the default value.
903 }
905 // glibc guard page
906 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
908 ThreadState state;
910 {
911 // Serialize thread creation if we are running with fixed stack LinuxThreads
912 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
913 if (lock) {
914 os::Linux::createThread_lock()->lock_without_safepoint_check();
915 }
917 pthread_t tid;
918 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
920 pthread_attr_destroy(&attr);
922 if (ret != 0) {
923 if (PrintMiscellaneous && (Verbose || WizardMode)) {
924 perror("pthread_create()");
925 }
926 // Need to clean up stuff we've allocated so far
927 thread->set_osthread(NULL);
928 delete osthread;
929 if (lock) os::Linux::createThread_lock()->unlock();
930 return false;
931 }
933 // Store pthread info into the OSThread
934 osthread->set_pthread_id(tid);
936 // Wait until child thread is either initialized or aborted
937 {
938 Monitor* sync_with_child = osthread->startThread_lock();
939 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
940 while ((state = osthread->get_state()) == ALLOCATED) {
941 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
942 }
943 }
945 if (lock) {
946 os::Linux::createThread_lock()->unlock();
947 }
948 }
950 // Aborted due to thread limit being reached
951 if (state == ZOMBIE) {
952 thread->set_osthread(NULL);
953 delete osthread;
954 return false;
955 }
957 // The thread is returned suspended (in state INITIALIZED),
958 // and is started higher up in the call chain
959 assert(state == INITIALIZED, "race condition");
960 return true;
961 }
963 /////////////////////////////////////////////////////////////////////////////
964 // attach existing thread
966 // bootstrap the main thread
967 bool os::create_main_thread(JavaThread* thread) {
968 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
969 return create_attached_thread(thread);
970 }
972 bool os::create_attached_thread(JavaThread* thread) {
973 #ifdef ASSERT
974 thread->verify_not_published();
975 #endif
977 // Allocate the OSThread object
978 OSThread* osthread = new OSThread(NULL, NULL);
980 if (osthread == NULL) {
981 return false;
982 }
984 // Store pthread info into the OSThread
985 osthread->set_thread_id(os::Linux::gettid());
986 osthread->set_pthread_id(::pthread_self());
988 // initialize floating point control register
989 os::Linux::init_thread_fpu_state();
991 // Initial thread state is RUNNABLE
992 osthread->set_state(RUNNABLE);
994 thread->set_osthread(osthread);
996 if (UseNUMA) {
997 int lgrp_id = os::numa_get_group_id();
998 if (lgrp_id != -1) {
999 thread->set_lgrp_id(lgrp_id);
1000 }
1001 }
1003 if (os::is_primordial_thread()) {
1004 // If current thread is primordial thread, its stack is mapped on demand,
1005 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1006 // the entire stack region to avoid SEGV in stack banging.
1007 // It is also useful to get around the heap-stack-gap problem on SuSE
1008 // kernel (see 4821821 for details). We first expand stack to the top
1009 // of yellow zone, then enable stack yellow zone (order is significant,
1010 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1011 // is no gap between the last two virtual memory regions.
1013 JavaThread *jt = (JavaThread *)thread;
1014 address addr = jt->stack_yellow_zone_base();
1015 assert(addr != NULL, "initialization problem?");
1016 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1018 osthread->set_expanding_stack();
1019 os::Linux::manually_expand_stack(jt, addr);
1020 osthread->clear_expanding_stack();
1021 }
1023 // initialize signal mask for this thread
1024 // and save the caller's signal mask
1025 os::Linux::hotspot_sigmask(thread);
1027 return true;
1028 }
1030 void os::pd_start_thread(Thread* thread) {
1031 OSThread * osthread = thread->osthread();
1032 assert(osthread->get_state() != INITIALIZED, "just checking");
1033 Monitor* sync_with_child = osthread->startThread_lock();
1034 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1035 sync_with_child->notify();
1037 #if defined MIPS && !defined ZERO
1038 //To be accessed in NativeGeneralJump::patch_verified_entry()
1039 if (thread->is_Java_thread())
1040 {
1041 ((JavaThread*)thread)->set_handle_wrong_method_stub(SharedRuntime::get_handle_wrong_method_stub());
1042 }
1043 #endif
1044 }
1046 // Free Linux resources related to the OSThread
1047 void os::free_thread(OSThread* osthread) {
1048 assert(osthread != NULL, "osthread not set");
1050 if (Thread::current()->osthread() == osthread) {
1051 // Restore caller's signal mask
1052 sigset_t sigmask = osthread->caller_sigmask();
1053 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1054 }
1056 delete osthread;
1057 }
1059 //////////////////////////////////////////////////////////////////////////////
1060 // thread local storage
1062 // Restore the thread pointer if the destructor is called. This is in case
1063 // someone from JNI code sets up a destructor with pthread_key_create to run
1064 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1065 // will hang or crash. When detachCurrentThread is called the key will be set
1066 // to null and we will not be called again. If detachCurrentThread is never
1067 // called we could loop forever depending on the pthread implementation.
1068 static void restore_thread_pointer(void* p) {
1069 Thread* thread = (Thread*) p;
1070 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1071 }
1073 int os::allocate_thread_local_storage() {
1074 pthread_key_t key;
1075 int rslt = pthread_key_create(&key, restore_thread_pointer);
1076 assert(rslt == 0, "cannot allocate thread local storage");
1077 return (int)key;
1078 }
1080 // Note: This is currently not used by VM, as we don't destroy TLS key
1081 // on VM exit.
1082 void os::free_thread_local_storage(int index) {
1083 int rslt = pthread_key_delete((pthread_key_t)index);
1084 assert(rslt == 0, "invalid index");
1085 }
1087 void os::thread_local_storage_at_put(int index, void* value) {
1088 int rslt = pthread_setspecific((pthread_key_t)index, value);
1089 assert(rslt == 0, "pthread_setspecific failed");
1090 }
1092 extern "C" Thread* get_thread() {
1093 return ThreadLocalStorage::thread();
1094 }
1096 //////////////////////////////////////////////////////////////////////////////
1097 // primordial thread
1099 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1100 bool os::is_primordial_thread(void) {
1101 char dummy;
1102 // If called before init complete, thread stack bottom will be null.
1103 // Can be called if fatal error occurs before initialization.
1104 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1105 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1106 os::Linux::initial_thread_stack_size() != 0,
1107 "os::init did not locate primordial thread's stack region");
1108 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1109 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1110 os::Linux::initial_thread_stack_size()) {
1111 return true;
1112 } else {
1113 return false;
1114 }
1115 }
1117 // Find the virtual memory area that contains addr
1118 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1119 FILE *fp = fopen("/proc/self/maps", "r");
1120 if (fp) {
1121 address low, high;
1122 while (!feof(fp)) {
1123 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1124 if (low <= addr && addr < high) {
1125 if (vma_low) *vma_low = low;
1126 if (vma_high) *vma_high = high;
1127 fclose (fp);
1128 return true;
1129 }
1130 }
1131 for (;;) {
1132 int ch = fgetc(fp);
1133 if (ch == EOF || ch == (int)'\n') break;
1134 }
1135 }
1136 fclose(fp);
1137 }
1138 return false;
1139 }
1141 // Locate primordial thread stack. This special handling of primordial thread stack
1142 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1143 // bogus value for the primordial process thread. While the launcher has created
1144 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1145 // JNI invocation API from a primordial thread.
1146 void os::Linux::capture_initial_stack(size_t max_size) {
1148 // max_size is either 0 (which means accept OS default for thread stacks) or
1149 // a user-specified value known to be at least the minimum needed. If we
1150 // are actually on the primordial thread we can make it appear that we have a
1151 // smaller max_size stack by inserting the guard pages at that location. But we
1152 // cannot do anything to emulate a larger stack than what has been provided by
1153 // the OS or threading library. In fact if we try to use a stack greater than
1154 // what is set by rlimit then we will crash the hosting process.
1156 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1157 // If this is "unlimited" then it will be a huge value.
1158 struct rlimit rlim;
1159 getrlimit(RLIMIT_STACK, &rlim);
1160 size_t stack_size = rlim.rlim_cur;
1162 // 6308388: a bug in ld.so will relocate its own .data section to the
1163 // lower end of primordial stack; reduce ulimit -s value a little bit
1164 // so we won't install guard page on ld.so's data section.
1165 // But ensure we don't underflow the stack size - allow 1 page spare
1166 if (stack_size >= (size_t)(3 * page_size())) {
1167 stack_size -= 2 * page_size();
1168 }
1170 // Try to figure out where the stack base (top) is. This is harder.
1171 //
1172 // When an application is started, glibc saves the initial stack pointer in
1173 // a global variable "__libc_stack_end", which is then used by system
1174 // libraries. __libc_stack_end should be pretty close to stack top. The
1175 // variable is available since the very early days. However, because it is
1176 // a private interface, it could disappear in the future.
1177 //
1178 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1179 // to __libc_stack_end, it is very close to stack top, but isn't the real
1180 // stack top. Note that /proc may not exist if VM is running as a chroot
1181 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1182 // /proc/<pid>/stat could change in the future (though unlikely).
1183 //
1184 // We try __libc_stack_end first. If that doesn't work, look for
1185 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1186 // as a hint, which should work well in most cases.
1188 uintptr_t stack_start;
1190 // try __libc_stack_end first
1191 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1192 if (p && *p) {
1193 stack_start = *p;
1194 } else {
1195 // see if we can get the start_stack field from /proc/self/stat
1196 FILE *fp;
1197 int pid;
1198 char state;
1199 int ppid;
1200 int pgrp;
1201 int session;
1202 int nr;
1203 int tpgrp;
1204 unsigned long flags;
1205 unsigned long minflt;
1206 unsigned long cminflt;
1207 unsigned long majflt;
1208 unsigned long cmajflt;
1209 unsigned long utime;
1210 unsigned long stime;
1211 long cutime;
1212 long cstime;
1213 long prio;
1214 long nice;
1215 long junk;
1216 long it_real;
1217 uintptr_t start;
1218 uintptr_t vsize;
1219 intptr_t rss;
1220 uintptr_t rsslim;
1221 uintptr_t scodes;
1222 uintptr_t ecode;
1223 int i;
1225 // Figure what the primordial thread stack base is. Code is inspired
1226 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1227 // followed by command name surrounded by parentheses, state, etc.
1228 char stat[2048];
1229 int statlen;
1231 fp = fopen("/proc/self/stat", "r");
1232 if (fp) {
1233 statlen = fread(stat, 1, 2047, fp);
1234 stat[statlen] = '\0';
1235 fclose(fp);
1237 // Skip pid and the command string. Note that we could be dealing with
1238 // weird command names, e.g. user could decide to rename java launcher
1239 // to "java 1.4.2 :)", then the stat file would look like
1240 // 1234 (java 1.4.2 :)) R ... ...
1241 // We don't really need to know the command string, just find the last
1242 // occurrence of ")" and then start parsing from there. See bug 4726580.
1243 char * s = strrchr(stat, ')');
1245 i = 0;
1246 if (s) {
1247 // Skip blank chars
1248 do s++; while (isspace(*s));
1250 #define _UFM UINTX_FORMAT
1251 #define _DFM INTX_FORMAT
1253 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1254 /* 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 */
1255 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,
1256 &state, /* 3 %c */
1257 &ppid, /* 4 %d */
1258 &pgrp, /* 5 %d */
1259 &session, /* 6 %d */
1260 &nr, /* 7 %d */
1261 &tpgrp, /* 8 %d */
1262 &flags, /* 9 %lu */
1263 &minflt, /* 10 %lu */
1264 &cminflt, /* 11 %lu */
1265 &majflt, /* 12 %lu */
1266 &cmajflt, /* 13 %lu */
1267 &utime, /* 14 %lu */
1268 &stime, /* 15 %lu */
1269 &cutime, /* 16 %ld */
1270 &cstime, /* 17 %ld */
1271 &prio, /* 18 %ld */
1272 &nice, /* 19 %ld */
1273 &junk, /* 20 %ld */
1274 &it_real, /* 21 %ld */
1275 &start, /* 22 UINTX_FORMAT */
1276 &vsize, /* 23 UINTX_FORMAT */
1277 &rss, /* 24 INTX_FORMAT */
1278 &rsslim, /* 25 UINTX_FORMAT */
1279 &scodes, /* 26 UINTX_FORMAT */
1280 &ecode, /* 27 UINTX_FORMAT */
1281 &stack_start); /* 28 UINTX_FORMAT */
1282 }
1284 #undef _UFM
1285 #undef _DFM
1287 if (i != 28 - 2) {
1288 assert(false, "Bad conversion from /proc/self/stat");
1289 // product mode - assume we are the primordial thread, good luck in the
1290 // embedded case.
1291 warning("Can't detect primordial thread stack location - bad conversion");
1292 stack_start = (uintptr_t) &rlim;
1293 }
1294 } else {
1295 // For some reason we can't open /proc/self/stat (for example, running on
1296 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1297 // most cases, so don't abort:
1298 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1299 stack_start = (uintptr_t) &rlim;
1300 }
1301 }
1303 // Now we have a pointer (stack_start) very close to the stack top, the
1304 // next thing to do is to figure out the exact location of stack top. We
1305 // can find out the virtual memory area that contains stack_start by
1306 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1307 // and its upper limit is the real stack top. (again, this would fail if
1308 // running inside chroot, because /proc may not exist.)
1310 uintptr_t stack_top;
1311 address low, high;
1312 if (find_vma((address)stack_start, &low, &high)) {
1313 // success, "high" is the true stack top. (ignore "low", because initial
1314 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1315 stack_top = (uintptr_t)high;
1316 } else {
1317 // failed, likely because /proc/self/maps does not exist
1318 warning("Can't detect primordial thread stack location - find_vma failed");
1319 // best effort: stack_start is normally within a few pages below the real
1320 // stack top, use it as stack top, and reduce stack size so we won't put
1321 // guard page outside stack.
1322 stack_top = stack_start;
1323 stack_size -= 16 * page_size();
1324 }
1326 // stack_top could be partially down the page so align it
1327 stack_top = align_size_up(stack_top, page_size());
1329 // Allowed stack value is minimum of max_size and what we derived from rlimit
1330 if (max_size > 0) {
1331 _initial_thread_stack_size = MIN2(max_size, stack_size);
1332 } else {
1333 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1334 // clamp it at 8MB as we do on Solaris
1335 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1336 }
1338 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1339 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1340 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1341 }
1343 ////////////////////////////////////////////////////////////////////////////////
1344 // time support
1346 // Time since start-up in seconds to a fine granularity.
1347 // Used by VMSelfDestructTimer and the MemProfiler.
1348 double os::elapsedTime() {
1350 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1351 }
1353 jlong os::elapsed_counter() {
1354 return javaTimeNanos() - initial_time_count;
1355 }
1357 jlong os::elapsed_frequency() {
1358 return NANOSECS_PER_SEC; // nanosecond resolution
1359 }
1361 bool os::supports_vtime() { return true; }
1362 bool os::enable_vtime() { return false; }
1363 bool os::vtime_enabled() { return false; }
1365 double os::elapsedVTime() {
1366 struct rusage usage;
1367 int retval = getrusage(RUSAGE_THREAD, &usage);
1368 if (retval == 0) {
1369 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);
1370 } else {
1371 // better than nothing, but not much
1372 return elapsedTime();
1373 }
1374 }
1376 jlong os::javaTimeMillis() {
1377 timeval time;
1378 int status = gettimeofday(&time, NULL);
1379 assert(status != -1, "linux error");
1380 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1381 }
1383 #ifndef CLOCK_MONOTONIC
1384 #define CLOCK_MONOTONIC (1)
1385 #endif
1387 void os::Linux::clock_init() {
1388 // we do dlopen's in this particular order due to bug in linux
1389 // dynamical loader (see 6348968) leading to crash on exit
1390 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1391 if (handle == NULL) {
1392 handle = dlopen("librt.so", RTLD_LAZY);
1393 }
1395 if (handle) {
1396 int (*clock_getres_func)(clockid_t, struct timespec*) =
1397 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1398 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1399 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1400 if (clock_getres_func && clock_gettime_func) {
1401 // See if monotonic clock is supported by the kernel. Note that some
1402 // early implementations simply return kernel jiffies (updated every
1403 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1404 // for nano time (though the monotonic property is still nice to have).
1405 // It's fixed in newer kernels, however clock_getres() still returns
1406 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1407 // resolution for now. Hopefully as people move to new kernels, this
1408 // won't be a problem.
1409 struct timespec res;
1410 struct timespec tp;
1411 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1412 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1413 // yes, monotonic clock is supported
1414 _clock_gettime = clock_gettime_func;
1415 return;
1416 } else {
1417 // close librt if there is no monotonic clock
1418 dlclose(handle);
1419 }
1420 }
1421 }
1422 warning("No monotonic clock was available - timed services may " \
1423 "be adversely affected if the time-of-day clock changes");
1424 }
1426 #ifndef SYS_clock_getres
1428 #if defined(IA32) || defined(AMD64)
1429 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1430 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1431 #else
1432 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1433 #define sys_clock_getres(x,y) -1
1434 #endif
1436 #else
1437 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1438 #endif
1440 void os::Linux::fast_thread_clock_init() {
1441 if (!UseLinuxPosixThreadCPUClocks) {
1442 return;
1443 }
1444 clockid_t clockid;
1445 struct timespec tp;
1446 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1447 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1449 // Switch to using fast clocks for thread cpu time if
1450 // the sys_clock_getres() returns 0 error code.
1451 // Note, that some kernels may support the current thread
1452 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1453 // returned by the pthread_getcpuclockid().
1454 // If the fast Posix clocks are supported then the sys_clock_getres()
1455 // must return at least tp.tv_sec == 0 which means a resolution
1456 // better than 1 sec. This is extra check for reliability.
1458 if(pthread_getcpuclockid_func &&
1459 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1460 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1462 _supports_fast_thread_cpu_time = true;
1463 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1464 }
1465 }
1467 jlong os::javaTimeNanos() {
1468 if (Linux::supports_monotonic_clock()) {
1469 struct timespec tp;
1470 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1471 assert(status == 0, "gettime error");
1472 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1473 return result;
1474 } else {
1475 timeval time;
1476 int status = gettimeofday(&time, NULL);
1477 assert(status != -1, "linux error");
1478 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1479 return 1000 * usecs;
1480 }
1481 }
1483 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1484 if (Linux::supports_monotonic_clock()) {
1485 info_ptr->max_value = ALL_64_BITS;
1487 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1488 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1489 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1490 } else {
1491 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1492 info_ptr->max_value = ALL_64_BITS;
1494 // gettimeofday is a real time clock so it skips
1495 info_ptr->may_skip_backward = true;
1496 info_ptr->may_skip_forward = true;
1497 }
1499 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1500 }
1502 // Return the real, user, and system times in seconds from an
1503 // arbitrary fixed point in the past.
1504 bool os::getTimesSecs(double* process_real_time,
1505 double* process_user_time,
1506 double* process_system_time) {
1507 struct tms ticks;
1508 clock_t real_ticks = times(&ticks);
1510 if (real_ticks == (clock_t) (-1)) {
1511 return false;
1512 } else {
1513 double ticks_per_second = (double) clock_tics_per_sec;
1514 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1515 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1516 *process_real_time = ((double) real_ticks) / ticks_per_second;
1518 return true;
1519 }
1520 }
1523 char * os::local_time_string(char *buf, size_t buflen) {
1524 struct tm t;
1525 time_t long_time;
1526 time(&long_time);
1527 localtime_r(&long_time, &t);
1528 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1529 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1530 t.tm_hour, t.tm_min, t.tm_sec);
1531 return buf;
1532 }
1534 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1535 return localtime_r(clock, res);
1536 }
1538 ////////////////////////////////////////////////////////////////////////////////
1539 // runtime exit support
1541 // Note: os::shutdown() might be called very early during initialization, or
1542 // called from signal handler. Before adding something to os::shutdown(), make
1543 // sure it is async-safe and can handle partially initialized VM.
1544 void os::shutdown() {
1546 // allow PerfMemory to attempt cleanup of any persistent resources
1547 perfMemory_exit();
1549 // needs to remove object in file system
1550 AttachListener::abort();
1552 // flush buffered output, finish log files
1553 ostream_abort();
1555 // Check for abort hook
1556 abort_hook_t abort_hook = Arguments::abort_hook();
1557 if (abort_hook != NULL) {
1558 abort_hook();
1559 }
1561 }
1563 // Note: os::abort() might be called very early during initialization, or
1564 // called from signal handler. Before adding something to os::abort(), make
1565 // sure it is async-safe and can handle partially initialized VM.
1566 void os::abort(bool dump_core) {
1567 os::shutdown();
1568 if (dump_core) {
1569 #ifndef PRODUCT
1570 fdStream out(defaultStream::output_fd());
1571 out.print_raw("Current thread is ");
1572 char buf[16];
1573 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1574 out.print_raw_cr(buf);
1575 out.print_raw_cr("Dumping core ...");
1576 #endif
1577 ::abort(); // dump core
1578 }
1580 ::exit(1);
1581 }
1583 // Die immediately, no exit hook, no abort hook, no cleanup.
1584 void os::die() {
1585 // _exit() on LinuxThreads only kills current thread
1586 ::abort();
1587 }
1590 // This method is a copy of JDK's sysGetLastErrorString
1591 // from src/solaris/hpi/src/system_md.c
1593 size_t os::lasterror(char *buf, size_t len) {
1595 if (errno == 0) return 0;
1597 const char *s = ::strerror(errno);
1598 size_t n = ::strlen(s);
1599 if (n >= len) {
1600 n = len - 1;
1601 }
1602 ::strncpy(buf, s, n);
1603 buf[n] = '\0';
1604 return n;
1605 }
1607 intx os::current_thread_id() { return (intx)pthread_self(); }
1608 int os::current_process_id() {
1610 // Under the old linux thread library, linux gives each thread
1611 // its own process id. Because of this each thread will return
1612 // a different pid if this method were to return the result
1613 // of getpid(2). Linux provides no api that returns the pid
1614 // of the launcher thread for the vm. This implementation
1615 // returns a unique pid, the pid of the launcher thread
1616 // that starts the vm 'process'.
1618 // Under the NPTL, getpid() returns the same pid as the
1619 // launcher thread rather than a unique pid per thread.
1620 // Use gettid() if you want the old pre NPTL behaviour.
1622 // if you are looking for the result of a call to getpid() that
1623 // returns a unique pid for the calling thread, then look at the
1624 // OSThread::thread_id() method in osThread_linux.hpp file
1626 return (int)(_initial_pid ? _initial_pid : getpid());
1627 }
1629 // DLL functions
1631 const char* os::dll_file_extension() { return ".so"; }
1633 // This must be hard coded because it's the system's temporary
1634 // directory not the java application's temp directory, ala java.io.tmpdir.
1635 const char* os::get_temp_directory() { return "/tmp"; }
1637 static bool file_exists(const char* filename) {
1638 struct stat statbuf;
1639 if (filename == NULL || strlen(filename) == 0) {
1640 return false;
1641 }
1642 return os::stat(filename, &statbuf) == 0;
1643 }
1645 bool os::dll_build_name(char* buffer, size_t buflen,
1646 const char* pname, const char* fname) {
1647 bool retval = false;
1648 // Copied from libhpi
1649 const size_t pnamelen = pname ? strlen(pname) : 0;
1651 // Return error on buffer overflow.
1652 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1653 return retval;
1654 }
1656 if (pnamelen == 0) {
1657 snprintf(buffer, buflen, "lib%s.so", fname);
1658 retval = true;
1659 } else if (strchr(pname, *os::path_separator()) != NULL) {
1660 int n;
1661 char** pelements = split_path(pname, &n);
1662 if (pelements == NULL) {
1663 return false;
1664 }
1665 for (int i = 0 ; i < n ; i++) {
1666 // Really shouldn't be NULL, but check can't hurt
1667 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1668 continue; // skip the empty path values
1669 }
1670 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1671 if (file_exists(buffer)) {
1672 retval = true;
1673 break;
1674 }
1675 }
1676 // release the storage
1677 for (int i = 0 ; i < n ; i++) {
1678 if (pelements[i] != NULL) {
1679 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1680 }
1681 }
1682 if (pelements != NULL) {
1683 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1684 }
1685 } else {
1686 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1687 retval = true;
1688 }
1689 return retval;
1690 }
1692 // check if addr is inside libjvm.so
1693 bool os::address_is_in_vm(address addr) {
1694 static address libjvm_base_addr;
1695 Dl_info dlinfo;
1697 if (libjvm_base_addr == NULL) {
1698 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1699 libjvm_base_addr = (address)dlinfo.dli_fbase;
1700 }
1701 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1702 }
1704 if (dladdr((void *)addr, &dlinfo) != 0) {
1705 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1706 }
1708 return false;
1709 }
1711 bool os::dll_address_to_function_name(address addr, char *buf,
1712 int buflen, int *offset) {
1713 // buf is not optional, but offset is optional
1714 assert(buf != NULL, "sanity check");
1716 Dl_info dlinfo;
1718 if (dladdr((void*)addr, &dlinfo) != 0) {
1719 // see if we have a matching symbol
1720 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1721 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1722 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1723 }
1724 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1725 return true;
1726 }
1727 // no matching symbol so try for just file info
1728 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1729 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1730 buf, buflen, offset, dlinfo.dli_fname)) {
1731 return true;
1732 }
1733 }
1734 }
1736 buf[0] = '\0';
1737 if (offset != NULL) *offset = -1;
1738 return false;
1739 }
1741 struct _address_to_library_name {
1742 address addr; // input : memory address
1743 size_t buflen; // size of fname
1744 char* fname; // output: library name
1745 address base; // library base addr
1746 };
1748 static int address_to_library_name_callback(struct dl_phdr_info *info,
1749 size_t size, void *data) {
1750 int i;
1751 bool found = false;
1752 address libbase = NULL;
1753 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1755 // iterate through all loadable segments
1756 for (i = 0; i < info->dlpi_phnum; i++) {
1757 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1758 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1759 // base address of a library is the lowest address of its loaded
1760 // segments.
1761 if (libbase == NULL || libbase > segbase) {
1762 libbase = segbase;
1763 }
1764 // see if 'addr' is within current segment
1765 if (segbase <= d->addr &&
1766 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1767 found = true;
1768 }
1769 }
1770 }
1772 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1773 // so dll_address_to_library_name() can fall through to use dladdr() which
1774 // can figure out executable name from argv[0].
1775 if (found && info->dlpi_name && info->dlpi_name[0]) {
1776 d->base = libbase;
1777 if (d->fname) {
1778 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1779 }
1780 return 1;
1781 }
1782 return 0;
1783 }
1785 bool os::dll_address_to_library_name(address addr, char* buf,
1786 int buflen, int* offset) {
1787 // buf is not optional, but offset is optional
1788 assert(buf != NULL, "sanity check");
1790 Dl_info dlinfo;
1791 struct _address_to_library_name data;
1793 // There is a bug in old glibc dladdr() implementation that it could resolve
1794 // to wrong library name if the .so file has a base address != NULL. Here
1795 // we iterate through the program headers of all loaded libraries to find
1796 // out which library 'addr' really belongs to. This workaround can be
1797 // removed once the minimum requirement for glibc is moved to 2.3.x.
1798 data.addr = addr;
1799 data.fname = buf;
1800 data.buflen = buflen;
1801 data.base = NULL;
1802 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1804 if (rslt) {
1805 // buf already contains library name
1806 if (offset) *offset = addr - data.base;
1807 return true;
1808 }
1809 if (dladdr((void*)addr, &dlinfo) != 0) {
1810 if (dlinfo.dli_fname != NULL) {
1811 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1812 }
1813 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1814 *offset = addr - (address)dlinfo.dli_fbase;
1815 }
1816 return true;
1817 }
1819 buf[0] = '\0';
1820 if (offset) *offset = -1;
1821 return false;
1822 }
1824 // Loads .dll/.so and
1825 // in case of error it checks if .dll/.so was built for the
1826 // same architecture as Hotspot is running on
1829 // Remember the stack's state. The Linux dynamic linker will change
1830 // the stack to 'executable' at most once, so we must safepoint only once.
1831 bool os::Linux::_stack_is_executable = false;
1833 // VM operation that loads a library. This is necessary if stack protection
1834 // of the Java stacks can be lost during loading the library. If we
1835 // do not stop the Java threads, they can stack overflow before the stacks
1836 // are protected again.
1837 class VM_LinuxDllLoad: public VM_Operation {
1838 private:
1839 const char *_filename;
1840 char *_ebuf;
1841 int _ebuflen;
1842 void *_lib;
1843 public:
1844 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1845 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1846 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1847 void doit() {
1848 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1849 os::Linux::_stack_is_executable = true;
1850 }
1851 void* loaded_library() { return _lib; }
1852 };
1854 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1855 {
1856 void * result = NULL;
1857 bool load_attempted = false;
1859 // Check whether the library to load might change execution rights
1860 // of the stack. If they are changed, the protection of the stack
1861 // guard pages will be lost. We need a safepoint to fix this.
1862 //
1863 // See Linux man page execstack(8) for more info.
1864 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1865 ElfFile ef(filename);
1866 if (!ef.specifies_noexecstack()) {
1867 if (!is_init_completed()) {
1868 os::Linux::_stack_is_executable = true;
1869 // This is OK - No Java threads have been created yet, and hence no
1870 // stack guard pages to fix.
1871 //
1872 // This should happen only when you are building JDK7 using a very
1873 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1874 //
1875 // Dynamic loader will make all stacks executable after
1876 // this function returns, and will not do that again.
1877 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1878 } else {
1879 warning("You have loaded library %s which might have disabled stack guard. "
1880 "The VM will try to fix the stack guard now.\n"
1881 "It's highly recommended that you fix the library with "
1882 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1883 filename);
1885 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1886 JavaThread *jt = JavaThread::current();
1887 if (jt->thread_state() != _thread_in_native) {
1888 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1889 // that requires ExecStack. Cannot enter safe point. Let's give up.
1890 warning("Unable to fix stack guard. Giving up.");
1891 } else {
1892 if (!LoadExecStackDllInVMThread) {
1893 // This is for the case where the DLL has an static
1894 // constructor function that executes JNI code. We cannot
1895 // load such DLLs in the VMThread.
1896 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1897 }
1899 ThreadInVMfromNative tiv(jt);
1900 debug_only(VMNativeEntryWrapper vew;)
1902 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1903 VMThread::execute(&op);
1904 if (LoadExecStackDllInVMThread) {
1905 result = op.loaded_library();
1906 }
1907 load_attempted = true;
1908 }
1909 }
1910 }
1911 }
1913 if (!load_attempted) {
1914 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1915 }
1917 if (result != NULL) {
1918 // Successful loading
1919 return result;
1920 }
1922 Elf32_Ehdr elf_head;
1923 int diag_msg_max_length=ebuflen-strlen(ebuf);
1924 char* diag_msg_buf=ebuf+strlen(ebuf);
1926 if (diag_msg_max_length==0) {
1927 // No more space in ebuf for additional diagnostics message
1928 return NULL;
1929 }
1932 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1934 if (file_descriptor < 0) {
1935 // Can't open library, report dlerror() message
1936 return NULL;
1937 }
1939 bool failed_to_read_elf_head=
1940 (sizeof(elf_head)!=
1941 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1943 ::close(file_descriptor);
1944 if (failed_to_read_elf_head) {
1945 // file i/o error - report dlerror() msg
1946 return NULL;
1947 }
1949 typedef struct {
1950 Elf32_Half code; // Actual value as defined in elf.h
1951 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1952 char elf_class; // 32 or 64 bit
1953 char endianess; // MSB or LSB
1954 char* name; // String representation
1955 } arch_t;
1957 #ifndef EM_486
1958 #define EM_486 6 /* Intel 80486 */
1959 #endif
1961 static const arch_t arch_array[]={
1962 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1963 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1964 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1965 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1966 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1967 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1968 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1969 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1970 #if defined(VM_LITTLE_ENDIAN)
1971 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1972 #else
1973 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1974 #endif
1975 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1976 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1977 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1978 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1979 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1980 {EM_MIPS, EM_MIPS, ELFCLASS64, ELFDATA2LSB, (char*)"MIPS64 LE"},
1981 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1982 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1983 };
1985 #if (defined IA32)
1986 static Elf32_Half running_arch_code=EM_386;
1987 #elif (defined AMD64)
1988 static Elf32_Half running_arch_code=EM_X86_64;
1989 #elif (defined IA64)
1990 static Elf32_Half running_arch_code=EM_IA_64;
1991 #elif (defined __sparc) && (defined _LP64)
1992 static Elf32_Half running_arch_code=EM_SPARCV9;
1993 #elif (defined __sparc) && (!defined _LP64)
1994 static Elf32_Half running_arch_code=EM_SPARC;
1995 #elif (defined MIPS64)
1996 static Elf32_Half running_arch_code=EM_MIPS;
1997 #elif (defined __powerpc64__)
1998 static Elf32_Half running_arch_code=EM_PPC64;
1999 #elif (defined __powerpc__)
2000 static Elf32_Half running_arch_code=EM_PPC;
2001 #elif (defined ARM)
2002 static Elf32_Half running_arch_code=EM_ARM;
2003 #elif (defined S390)
2004 static Elf32_Half running_arch_code=EM_S390;
2005 #elif (defined ALPHA)
2006 static Elf32_Half running_arch_code=EM_ALPHA;
2007 #elif (defined MIPSEL)
2008 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
2009 #elif (defined PARISC)
2010 static Elf32_Half running_arch_code=EM_PARISC;
2011 #elif (defined MIPS)
2012 static Elf32_Half running_arch_code=EM_MIPS;
2013 #elif (defined M68K)
2014 static Elf32_Half running_arch_code=EM_68K;
2015 #else
2016 #error Method os::dll_load requires that one of following is defined:\
2017 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, __mips64, PARISC, M68K
2018 #endif
2020 // Identify compatability class for VM's architecture and library's architecture
2021 // Obtain string descriptions for architectures
2023 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2024 int running_arch_index=-1;
2026 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2027 if (running_arch_code == arch_array[i].code) {
2028 running_arch_index = i;
2029 }
2030 if (lib_arch.code == arch_array[i].code) {
2031 lib_arch.compat_class = arch_array[i].compat_class;
2032 lib_arch.name = arch_array[i].name;
2033 }
2034 }
2036 assert(running_arch_index != -1,
2037 "Didn't find running architecture code (running_arch_code) in arch_array");
2038 if (running_arch_index == -1) {
2039 // Even though running architecture detection failed
2040 // we may still continue with reporting dlerror() message
2041 return NULL;
2042 }
2044 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2045 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2046 return NULL;
2047 }
2049 #ifndef S390
2050 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2051 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2052 return NULL;
2053 }
2054 #endif // !S390
2056 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2057 if ( lib_arch.name!=NULL ) {
2058 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2059 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2060 lib_arch.name, arch_array[running_arch_index].name);
2061 } else {
2062 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2063 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2064 lib_arch.code,
2065 arch_array[running_arch_index].name);
2066 }
2067 }
2069 return NULL;
2070 }
2072 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2073 void * result = ::dlopen(filename, RTLD_LAZY);
2074 if (result == NULL) {
2075 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2076 ebuf[ebuflen-1] = '\0';
2077 }
2078 return result;
2079 }
2081 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2082 void * result = NULL;
2083 if (LoadExecStackDllInVMThread) {
2084 result = dlopen_helper(filename, ebuf, ebuflen);
2085 }
2087 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2088 // library that requires an executable stack, or which does not have this
2089 // stack attribute set, dlopen changes the stack attribute to executable. The
2090 // read protection of the guard pages gets lost.
2091 //
2092 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2093 // may have been queued at the same time.
2095 if (!_stack_is_executable) {
2096 JavaThread *jt = Threads::first();
2098 while (jt) {
2099 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2100 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2101 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2102 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2103 warning("Attempt to reguard stack yellow zone failed.");
2104 }
2105 }
2106 jt = jt->next();
2107 }
2108 }
2110 return result;
2111 }
2113 /*
2114 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2115 * chances are you might want to run the generated bits against glibc-2.0
2116 * libdl.so, so always use locking for any version of glibc.
2117 */
2118 void* os::dll_lookup(void* handle, const char* name) {
2119 pthread_mutex_lock(&dl_mutex);
2120 void* res = dlsym(handle, name);
2121 pthread_mutex_unlock(&dl_mutex);
2122 return res;
2123 }
2125 void* os::get_default_process_handle() {
2126 return (void*)::dlopen(NULL, RTLD_LAZY);
2127 }
2129 static bool _print_ascii_file(const char* filename, outputStream* st) {
2130 int fd = ::open(filename, O_RDONLY);
2131 if (fd == -1) {
2132 return false;
2133 }
2135 char buf[32];
2136 int bytes;
2137 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2138 st->print_raw(buf, bytes);
2139 }
2141 ::close(fd);
2143 return true;
2144 }
2146 void os::print_dll_info(outputStream *st) {
2147 st->print_cr("Dynamic libraries:");
2149 char fname[32];
2150 pid_t pid = os::Linux::gettid();
2152 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2154 if (!_print_ascii_file(fname, st)) {
2155 st->print("Can not get library information for pid = %d\n", pid);
2156 }
2157 }
2159 void os::print_os_info_brief(outputStream* st) {
2160 os::Linux::print_distro_info(st);
2162 os::Posix::print_uname_info(st);
2164 os::Linux::print_libversion_info(st);
2166 }
2168 void os::print_os_info(outputStream* st) {
2169 st->print("OS:");
2171 os::Linux::print_distro_info(st);
2173 os::Posix::print_uname_info(st);
2175 // Print warning if unsafe chroot environment detected
2176 if (unsafe_chroot_detected) {
2177 st->print("WARNING!! ");
2178 st->print_cr("%s", unstable_chroot_error);
2179 }
2181 os::Linux::print_libversion_info(st);
2183 os::Posix::print_rlimit_info(st);
2185 os::Posix::print_load_average(st);
2187 os::Linux::print_full_memory_info(st);
2189 os::Linux::print_container_info(st);
2190 }
2192 // Try to identify popular distros.
2193 // Most Linux distributions have a /etc/XXX-release file, which contains
2194 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2195 // file that also contains the OS version string. Some have more than one
2196 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2197 // /etc/redhat-release.), so the order is important.
2198 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2199 // their own specific XXX-release file as well as a redhat-release file.
2200 // Because of this the XXX-release file needs to be searched for before the
2201 // redhat-release file.
2202 // Since Red Hat has a lsb-release file that is not very descriptive the
2203 // search for redhat-release needs to be before lsb-release.
2204 // Since the lsb-release file is the new standard it needs to be searched
2205 // before the older style release files.
2206 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2207 // next to last resort. The os-release file is a new standard that contains
2208 // distribution information and the system-release file seems to be an old
2209 // standard that has been replaced by the lsb-release and os-release files.
2210 // Searching for the debian_version file is the last resort. It contains
2211 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2212 // "Debian " is printed before the contents of the debian_version file.
2213 void os::Linux::print_distro_info(outputStream* st) {
2214 if (!_print_ascii_file("/etc/oracle-release", st) &&
2215 !_print_ascii_file("/etc/mandriva-release", st) &&
2216 !_print_ascii_file("/etc/mandrake-release", st) &&
2217 !_print_ascii_file("/etc/sun-release", st) &&
2218 !_print_ascii_file("/etc/redhat-release", st) &&
2219 !_print_ascii_file("/etc/lsb-release", st) &&
2220 !_print_ascii_file("/etc/SuSE-release", st) &&
2221 !_print_ascii_file("/etc/turbolinux-release", st) &&
2222 !_print_ascii_file("/etc/gentoo-release", st) &&
2223 !_print_ascii_file("/etc/ltib-release", st) &&
2224 !_print_ascii_file("/etc/angstrom-version", st) &&
2225 !_print_ascii_file("/etc/system-release", st) &&
2226 !_print_ascii_file("/etc/os-release", st)) {
2228 if (file_exists("/etc/debian_version")) {
2229 st->print("Debian ");
2230 _print_ascii_file("/etc/debian_version", st);
2231 } else {
2232 st->print("Linux");
2233 }
2234 }
2235 st->cr();
2236 }
2238 void os::Linux::print_libversion_info(outputStream* st) {
2239 // libc, pthread
2240 st->print("libc:");
2241 st->print("%s ", os::Linux::glibc_version());
2242 st->print("%s ", os::Linux::libpthread_version());
2243 if (os::Linux::is_LinuxThreads()) {
2244 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2245 }
2246 st->cr();
2247 }
2249 void os::Linux::print_full_memory_info(outputStream* st) {
2250 st->print("\n/proc/meminfo:\n");
2251 _print_ascii_file("/proc/meminfo", st);
2252 st->cr();
2253 }
2255 void os::Linux::print_container_info(outputStream* st) {
2256 if (!OSContainer::is_containerized()) {
2257 return;
2258 }
2260 st->print("container (cgroup) information:\n");
2262 const char *p_ct = OSContainer::container_type();
2263 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2265 char *p = OSContainer::cpu_cpuset_cpus();
2266 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2267 free(p);
2269 p = OSContainer::cpu_cpuset_memory_nodes();
2270 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2271 free(p);
2273 int i = OSContainer::active_processor_count();
2274 if (i > 0) {
2275 st->print("active_processor_count: %d\n", i);
2276 } else {
2277 st->print("active_processor_count: failed\n");
2278 }
2280 i = OSContainer::cpu_quota();
2281 st->print("cpu_quota: %d\n", i);
2283 i = OSContainer::cpu_period();
2284 st->print("cpu_period: %d\n", i);
2286 i = OSContainer::cpu_shares();
2287 st->print("cpu_shares: %d\n", i);
2289 jlong j = OSContainer::memory_limit_in_bytes();
2290 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2292 j = OSContainer::memory_and_swap_limit_in_bytes();
2293 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2295 j = OSContainer::memory_soft_limit_in_bytes();
2296 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2298 j = OSContainer::OSContainer::memory_usage_in_bytes();
2299 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2301 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2302 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2303 st->cr();
2304 }
2306 void os::print_memory_info(outputStream* st) {
2308 st->print("Memory:");
2309 st->print(" %dk page", os::vm_page_size()>>10);
2311 // values in struct sysinfo are "unsigned long"
2312 struct sysinfo si;
2313 sysinfo(&si);
2315 st->print(", physical " UINT64_FORMAT "k",
2316 os::physical_memory() >> 10);
2317 st->print("(" UINT64_FORMAT "k free)",
2318 os::available_memory() >> 10);
2319 st->print(", swap " UINT64_FORMAT "k",
2320 ((jlong)si.totalswap * si.mem_unit) >> 10);
2321 st->print("(" UINT64_FORMAT "k free)",
2322 ((jlong)si.freeswap * si.mem_unit) >> 10);
2323 st->cr();
2324 }
2326 void os::pd_print_cpu_info(outputStream* st) {
2327 st->print("\n/proc/cpuinfo:\n");
2328 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2329 st->print(" <Not Available>");
2330 }
2331 st->cr();
2332 }
2334 void os::print_siginfo(outputStream* st, void* siginfo) {
2335 const siginfo_t* si = (const siginfo_t*)siginfo;
2337 os::Posix::print_siginfo_brief(st, si);
2338 #if INCLUDE_CDS
2339 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2340 UseSharedSpaces) {
2341 FileMapInfo* mapinfo = FileMapInfo::current_info();
2342 if (mapinfo->is_in_shared_space(si->si_addr)) {
2343 st->print("\n\nError accessing class data sharing archive." \
2344 " Mapped file inaccessible during execution, " \
2345 " possible disk/network problem.");
2346 }
2347 }
2348 #endif
2349 st->cr();
2350 }
2353 static void print_signal_handler(outputStream* st, int sig,
2354 char* buf, size_t buflen);
2356 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2357 st->print_cr("Signal Handlers:");
2358 print_signal_handler(st, SIGSEGV, buf, buflen);
2359 print_signal_handler(st, SIGBUS , buf, buflen);
2360 print_signal_handler(st, SIGFPE , buf, buflen);
2361 print_signal_handler(st, SIGPIPE, buf, buflen);
2362 print_signal_handler(st, SIGXFSZ, buf, buflen);
2363 print_signal_handler(st, SIGILL , buf, buflen);
2364 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2365 print_signal_handler(st, SR_signum, buf, buflen);
2366 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2367 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2368 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2369 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2370 #if defined(PPC64)
2371 print_signal_handler(st, SIGTRAP, buf, buflen);
2372 #endif
2373 }
2375 static char saved_jvm_path[MAXPATHLEN] = {0};
2377 // Find the full path to the current module, libjvm.so
2378 void os::jvm_path(char *buf, jint buflen) {
2379 // Error checking.
2380 if (buflen < MAXPATHLEN) {
2381 assert(false, "must use a large-enough buffer");
2382 buf[0] = '\0';
2383 return;
2384 }
2385 // Lazy resolve the path to current module.
2386 if (saved_jvm_path[0] != 0) {
2387 strcpy(buf, saved_jvm_path);
2388 return;
2389 }
2391 char dli_fname[MAXPATHLEN];
2392 bool ret = dll_address_to_library_name(
2393 CAST_FROM_FN_PTR(address, os::jvm_path),
2394 dli_fname, sizeof(dli_fname), NULL);
2395 assert(ret, "cannot locate libjvm");
2396 char *rp = NULL;
2397 if (ret && dli_fname[0] != '\0') {
2398 rp = realpath(dli_fname, buf);
2399 }
2400 if (rp == NULL)
2401 return;
2403 if (Arguments::created_by_gamma_launcher()) {
2404 // Support for the gamma launcher. Typical value for buf is
2405 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2406 // the right place in the string, then assume we are installed in a JDK and
2407 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2408 // up the path so it looks like libjvm.so is installed there (append a
2409 // fake suffix hotspot/libjvm.so).
2410 const char *p = buf + strlen(buf) - 1;
2411 for (int count = 0; p > buf && count < 5; ++count) {
2412 for (--p; p > buf && *p != '/'; --p)
2413 /* empty */ ;
2414 }
2416 if (strncmp(p, "/jre/lib/", 9) != 0) {
2417 // Look for JAVA_HOME in the environment.
2418 char* java_home_var = ::getenv("JAVA_HOME");
2419 if (java_home_var != NULL && java_home_var[0] != 0) {
2420 char* jrelib_p;
2421 int len;
2423 // Check the current module name "libjvm.so".
2424 p = strrchr(buf, '/');
2425 assert(strstr(p, "/libjvm") == p, "invalid library name");
2427 rp = realpath(java_home_var, buf);
2428 if (rp == NULL)
2429 return;
2431 // determine if this is a legacy image or modules image
2432 // modules image doesn't have "jre" subdirectory
2433 len = strlen(buf);
2434 assert(len < buflen, "Ran out of buffer room");
2435 jrelib_p = buf + len;
2436 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2437 if (0 != access(buf, F_OK)) {
2438 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2439 }
2441 if (0 == access(buf, F_OK)) {
2442 // Use current module name "libjvm.so"
2443 len = strlen(buf);
2444 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2445 } else {
2446 // Go back to path of .so
2447 rp = realpath(dli_fname, buf);
2448 if (rp == NULL)
2449 return;
2450 }
2451 }
2452 }
2453 }
2455 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2456 }
2458 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2459 // no prefix required, not even "_"
2460 }
2462 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2463 // no suffix required
2464 }
2466 ////////////////////////////////////////////////////////////////////////////////
2467 // sun.misc.Signal support
2469 static volatile jint sigint_count = 0;
2471 static void
2472 UserHandler(int sig, void *siginfo, void *context) {
2473 // 4511530 - sem_post is serialized and handled by the manager thread. When
2474 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2475 // don't want to flood the manager thread with sem_post requests.
2476 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2477 return;
2479 // Ctrl-C is pressed during error reporting, likely because the error
2480 // handler fails to abort. Let VM die immediately.
2481 if (sig == SIGINT && is_error_reported()) {
2482 os::die();
2483 }
2485 os::signal_notify(sig);
2486 }
2488 void* os::user_handler() {
2489 return CAST_FROM_FN_PTR(void*, UserHandler);
2490 }
2492 class Semaphore : public StackObj {
2493 public:
2494 Semaphore();
2495 ~Semaphore();
2496 void signal();
2497 void wait();
2498 bool trywait();
2499 bool timedwait(unsigned int sec, int nsec);
2500 private:
2501 sem_t _semaphore;
2502 };
2504 Semaphore::Semaphore() {
2505 sem_init(&_semaphore, 0, 0);
2506 }
2508 Semaphore::~Semaphore() {
2509 sem_destroy(&_semaphore);
2510 }
2512 void Semaphore::signal() {
2513 sem_post(&_semaphore);
2514 }
2516 void Semaphore::wait() {
2517 sem_wait(&_semaphore);
2518 }
2520 bool Semaphore::trywait() {
2521 return sem_trywait(&_semaphore) == 0;
2522 }
2524 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2526 struct timespec ts;
2527 // Semaphore's are always associated with CLOCK_REALTIME
2528 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2529 // see unpackTime for discussion on overflow checking
2530 if (sec >= MAX_SECS) {
2531 ts.tv_sec += MAX_SECS;
2532 ts.tv_nsec = 0;
2533 } else {
2534 ts.tv_sec += sec;
2535 ts.tv_nsec += nsec;
2536 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2537 ts.tv_nsec -= NANOSECS_PER_SEC;
2538 ++ts.tv_sec; // note: this must be <= max_secs
2539 }
2540 }
2542 while (1) {
2543 int result = sem_timedwait(&_semaphore, &ts);
2544 if (result == 0) {
2545 return true;
2546 } else if (errno == EINTR) {
2547 continue;
2548 } else if (errno == ETIMEDOUT) {
2549 return false;
2550 } else {
2551 return false;
2552 }
2553 }
2554 }
2556 extern "C" {
2557 typedef void (*sa_handler_t)(int);
2558 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2559 }
2561 void* os::signal(int signal_number, void* handler) {
2562 struct sigaction sigAct, oldSigAct;
2564 sigfillset(&(sigAct.sa_mask));
2565 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2566 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2568 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2569 // -1 means registration failed
2570 return (void *)-1;
2571 }
2573 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2574 }
2576 void os::signal_raise(int signal_number) {
2577 ::raise(signal_number);
2578 }
2580 /*
2581 * The following code is moved from os.cpp for making this
2582 * code platform specific, which it is by its very nature.
2583 */
2585 // Will be modified when max signal is changed to be dynamic
2586 int os::sigexitnum_pd() {
2587 return NSIG;
2588 }
2590 // a counter for each possible signal value
2591 static volatile jint pending_signals[NSIG+1] = { 0 };
2593 // Linux(POSIX) specific hand shaking semaphore.
2594 static sem_t sig_sem;
2595 static Semaphore sr_semaphore;
2597 void os::signal_init_pd() {
2598 // Initialize signal structures
2599 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2601 // Initialize signal semaphore
2602 ::sem_init(&sig_sem, 0, 0);
2603 }
2605 void os::signal_notify(int sig) {
2606 Atomic::inc(&pending_signals[sig]);
2607 ::sem_post(&sig_sem);
2608 }
2610 static int check_pending_signals(bool wait) {
2611 Atomic::store(0, &sigint_count);
2612 for (;;) {
2613 for (int i = 0; i < NSIG + 1; i++) {
2614 jint n = pending_signals[i];
2615 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2616 return i;
2617 }
2618 }
2619 if (!wait) {
2620 return -1;
2621 }
2622 JavaThread *thread = JavaThread::current();
2623 ThreadBlockInVM tbivm(thread);
2625 bool threadIsSuspended;
2626 do {
2627 thread->set_suspend_equivalent();
2628 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2629 ::sem_wait(&sig_sem);
2631 // were we externally suspended while we were waiting?
2632 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2633 if (threadIsSuspended) {
2634 //
2635 // The semaphore has been incremented, but while we were waiting
2636 // another thread suspended us. We don't want to continue running
2637 // while suspended because that would surprise the thread that
2638 // suspended us.
2639 //
2640 ::sem_post(&sig_sem);
2642 thread->java_suspend_self();
2643 }
2644 } while (threadIsSuspended);
2645 }
2646 }
2648 int os::signal_lookup() {
2649 return check_pending_signals(false);
2650 }
2652 int os::signal_wait() {
2653 return check_pending_signals(true);
2654 }
2656 ////////////////////////////////////////////////////////////////////////////////
2657 // Virtual Memory
2659 int os::vm_page_size() {
2660 // Seems redundant as all get out
2661 assert(os::Linux::page_size() != -1, "must call os::init");
2662 return os::Linux::page_size();
2663 }
2665 // Solaris allocates memory by pages.
2666 int os::vm_allocation_granularity() {
2667 assert(os::Linux::page_size() != -1, "must call os::init");
2668 return os::Linux::page_size();
2669 }
2671 // Rationale behind this function:
2672 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2673 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2674 // samples for JITted code. Here we create private executable mapping over the code cache
2675 // and then we can use standard (well, almost, as mapping can change) way to provide
2676 // info for the reporting script by storing timestamp and location of symbol
2677 void linux_wrap_code(char* base, size_t size) {
2678 static volatile jint cnt = 0;
2680 if (!UseOprofile) {
2681 return;
2682 }
2684 char buf[PATH_MAX+1];
2685 int num = Atomic::add(1, &cnt);
2687 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2688 os::get_temp_directory(), os::current_process_id(), num);
2689 unlink(buf);
2691 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2693 if (fd != -1) {
2694 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2695 if (rv != (off_t)-1) {
2696 if (::write(fd, "", 1) == 1) {
2697 mmap(base, size,
2698 PROT_READ|PROT_WRITE|PROT_EXEC,
2699 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2700 }
2701 }
2702 ::close(fd);
2703 unlink(buf);
2704 }
2705 }
2707 static bool recoverable_mmap_error(int err) {
2708 // See if the error is one we can let the caller handle. This
2709 // list of errno values comes from JBS-6843484. I can't find a
2710 // Linux man page that documents this specific set of errno
2711 // values so while this list currently matches Solaris, it may
2712 // change as we gain experience with this failure mode.
2713 switch (err) {
2714 case EBADF:
2715 case EINVAL:
2716 case ENOTSUP:
2717 // let the caller deal with these errors
2718 return true;
2720 default:
2721 // Any remaining errors on this OS can cause our reserved mapping
2722 // to be lost. That can cause confusion where different data
2723 // structures think they have the same memory mapped. The worst
2724 // scenario is if both the VM and a library think they have the
2725 // same memory mapped.
2726 return false;
2727 }
2728 }
2730 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2731 int err) {
2732 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2733 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2734 strerror(err), err);
2735 }
2737 static void warn_fail_commit_memory(char* addr, size_t size,
2738 size_t alignment_hint, bool exec,
2739 int err) {
2740 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2741 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2742 alignment_hint, exec, strerror(err), err);
2743 }
2745 // NOTE: Linux kernel does not really reserve the pages for us.
2746 // All it does is to check if there are enough free pages
2747 // left at the time of mmap(). This could be a potential
2748 // problem.
2749 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2750 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2751 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2752 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2753 if (res != (uintptr_t) MAP_FAILED) {
2754 if (UseNUMAInterleaving) {
2755 numa_make_global(addr, size);
2756 }
2757 return 0;
2758 }
2760 int err = errno; // save errno from mmap() call above
2762 if (!recoverable_mmap_error(err)) {
2763 warn_fail_commit_memory(addr, size, exec, err);
2764 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2765 }
2767 return err;
2768 }
2770 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2771 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2772 }
2774 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2775 const char* mesg) {
2776 assert(mesg != NULL, "mesg must be specified");
2777 int err = os::Linux::commit_memory_impl(addr, size, exec);
2778 if (err != 0) {
2779 // the caller wants all commit errors to exit with the specified mesg:
2780 warn_fail_commit_memory(addr, size, exec, err);
2781 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2782 }
2783 }
2785 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2786 #ifndef MAP_HUGETLB
2787 #define MAP_HUGETLB 0x40000
2788 #endif
2790 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2791 #ifndef MADV_HUGEPAGE
2792 #define MADV_HUGEPAGE 14
2793 #endif
2795 int os::Linux::commit_memory_impl(char* addr, size_t size,
2796 size_t alignment_hint, bool exec) {
2797 int err = os::Linux::commit_memory_impl(addr, size, exec);
2798 if (err == 0) {
2799 realign_memory(addr, size, alignment_hint);
2800 }
2801 return err;
2802 }
2804 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2805 bool exec) {
2806 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2807 }
2809 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2810 size_t alignment_hint, bool exec,
2811 const char* mesg) {
2812 assert(mesg != NULL, "mesg must be specified");
2813 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2814 if (err != 0) {
2815 // the caller wants all commit errors to exit with the specified mesg:
2816 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2817 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2818 }
2819 }
2821 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2822 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2823 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2824 // be supported or the memory may already be backed by huge pages.
2825 ::madvise(addr, bytes, MADV_HUGEPAGE);
2826 }
2827 }
2829 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2830 // This method works by doing an mmap over an existing mmaping and effectively discarding
2831 // the existing pages. However it won't work for SHM-based large pages that cannot be
2832 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2833 // small pages on top of the SHM segment. This method always works for small pages, so we
2834 // allow that in any case.
2835 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2836 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2837 }
2838 }
2840 void os::numa_make_global(char *addr, size_t bytes) {
2841 Linux::numa_interleave_memory(addr, bytes);
2842 }
2844 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2845 // bind policy to MPOL_PREFERRED for the current thread.
2846 #define USE_MPOL_PREFERRED 0
2848 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2849 // To make NUMA and large pages more robust when both enabled, we need to ease
2850 // the requirements on where the memory should be allocated. MPOL_BIND is the
2851 // default policy and it will force memory to be allocated on the specified
2852 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2853 // the specified node, but will not force it. Using this policy will prevent
2854 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2855 // free large pages.
2856 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2857 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2858 }
2860 bool os::numa_topology_changed() { return false; }
2862 size_t os::numa_get_groups_num() {
2863 // Return just the number of nodes in which it's possible to allocate memory
2864 // (in numa terminology, configured nodes).
2865 return Linux::numa_num_configured_nodes();
2866 }
2868 int os::numa_get_group_id() {
2869 int cpu_id = Linux::sched_getcpu();
2870 if (cpu_id != -1) {
2871 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2872 if (lgrp_id != -1) {
2873 return lgrp_id;
2874 }
2875 }
2876 return 0;
2877 }
2879 int os::Linux::get_existing_num_nodes() {
2880 size_t node;
2881 size_t highest_node_number = Linux::numa_max_node();
2882 int num_nodes = 0;
2884 // Get the total number of nodes in the system including nodes without memory.
2885 for (node = 0; node <= highest_node_number; node++) {
2886 if (isnode_in_existing_nodes(node)) {
2887 num_nodes++;
2888 }
2889 }
2890 return num_nodes;
2891 }
2893 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2894 size_t highest_node_number = Linux::numa_max_node();
2895 size_t i = 0;
2897 // Map all node ids in which is possible to allocate memory. Also nodes are
2898 // not always consecutively available, i.e. available from 0 to the highest
2899 // node number.
2900 for (size_t node = 0; node <= highest_node_number; node++) {
2901 if (Linux::isnode_in_configured_nodes(node)) {
2902 ids[i++] = node;
2903 }
2904 }
2905 return i;
2906 }
2908 bool os::get_page_info(char *start, page_info* info) {
2909 return false;
2910 }
2912 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2913 return end;
2914 }
2917 int os::Linux::sched_getcpu_syscall(void) {
2918 unsigned int cpu = 0;
2919 int retval = -1;
2921 #if defined(IA32)
2922 # ifndef SYS_getcpu
2923 # define SYS_getcpu 318
2924 # endif
2925 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2926 #elif defined(AMD64)
2927 // Unfortunately we have to bring all these macros here from vsyscall.h
2928 // to be able to compile on old linuxes.
2929 # define __NR_vgetcpu 2
2930 # define VSYSCALL_START (-10UL << 20)
2931 # define VSYSCALL_SIZE 1024
2932 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2933 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2934 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2935 retval = vgetcpu(&cpu, NULL, NULL);
2936 #endif
2938 return (retval == -1) ? retval : cpu;
2939 }
2941 // Something to do with the numa-aware allocator needs these symbols
2942 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2943 extern "C" JNIEXPORT void numa_error(char *where) { }
2944 extern "C" JNIEXPORT int fork1() { return fork(); }
2946 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2947 // load symbol from base version instead.
2948 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2949 void *f = dlvsym(handle, name, "libnuma_1.1");
2950 if (f == NULL) {
2951 f = dlsym(handle, name);
2952 }
2953 return f;
2954 }
2956 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2957 // Return NULL if the symbol is not defined in this particular version.
2958 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2959 return dlvsym(handle, name, "libnuma_1.2");
2960 }
2962 bool os::Linux::libnuma_init() {
2963 // sched_getcpu() should be in libc.
2964 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2965 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2967 // If it's not, try a direct syscall.
2968 if (sched_getcpu() == -1)
2969 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2971 if (sched_getcpu() != -1) { // Does it work?
2972 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2973 if (handle != NULL) {
2974 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2975 libnuma_dlsym(handle, "numa_node_to_cpus")));
2976 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2977 libnuma_dlsym(handle, "numa_max_node")));
2978 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2979 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2980 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2981 libnuma_dlsym(handle, "numa_available")));
2982 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2983 libnuma_dlsym(handle, "numa_tonode_memory")));
2984 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2985 libnuma_dlsym(handle, "numa_interleave_memory")));
2986 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2987 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2988 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2989 libnuma_dlsym(handle, "numa_set_bind_policy")));
2990 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2991 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2992 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2993 libnuma_dlsym(handle, "numa_distance")));
2995 if (numa_available() != -1) {
2996 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2997 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2998 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2999 // Create an index -> node mapping, since nodes are not always consecutive
3000 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3001 rebuild_nindex_to_node_map();
3002 // Create a cpu -> node mapping
3003 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
3004 rebuild_cpu_to_node_map();
3005 return true;
3006 }
3007 }
3008 }
3009 return false;
3010 }
3012 void os::Linux::rebuild_nindex_to_node_map() {
3013 int highest_node_number = Linux::numa_max_node();
3015 nindex_to_node()->clear();
3016 for (int node = 0; node <= highest_node_number; node++) {
3017 if (Linux::isnode_in_existing_nodes(node)) {
3018 nindex_to_node()->append(node);
3019 }
3020 }
3021 }
3023 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3024 // The table is later used in get_node_by_cpu().
3025 void os::Linux::rebuild_cpu_to_node_map() {
3026 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3027 // in libnuma (possible values are starting from 16,
3028 // and continuing up with every other power of 2, but less
3029 // than the maximum number of CPUs supported by kernel), and
3030 // is a subject to change (in libnuma version 2 the requirements
3031 // are more reasonable) we'll just hardcode the number they use
3032 // in the library.
3033 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3035 size_t cpu_num = processor_count();
3036 size_t cpu_map_size = NCPUS / BitsPerCLong;
3037 size_t cpu_map_valid_size =
3038 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3040 cpu_to_node()->clear();
3041 cpu_to_node()->at_grow(cpu_num - 1);
3043 size_t node_num = get_existing_num_nodes();
3045 int distance = 0;
3046 int closest_distance = INT_MAX;
3047 int closest_node = 0;
3048 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3049 for (size_t i = 0; i < node_num; i++) {
3050 // Check if node is configured (not a memory-less node). If it is not, find
3051 // the closest configured node.
3052 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3053 closest_distance = INT_MAX;
3054 // Check distance from all remaining nodes in the system. Ignore distance
3055 // from itself and from another non-configured node.
3056 for (size_t m = 0; m < node_num; m++) {
3057 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3058 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3059 // If a closest node is found, update. There is always at least one
3060 // configured node in the system so there is always at least one node
3061 // close.
3062 if (distance != 0 && distance < closest_distance) {
3063 closest_distance = distance;
3064 closest_node = nindex_to_node()->at(m);
3065 }
3066 }
3067 }
3068 } else {
3069 // Current node is already a configured node.
3070 closest_node = nindex_to_node()->at(i);
3071 }
3073 // Get cpus from the original node and map them to the closest node. If node
3074 // is a configured node (not a memory-less node), then original node and
3075 // closest node are the same.
3076 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3077 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3078 if (cpu_map[j] != 0) {
3079 for (size_t k = 0; k < BitsPerCLong; k++) {
3080 if (cpu_map[j] & (1UL << k)) {
3081 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3082 }
3083 }
3084 }
3085 }
3086 }
3087 }
3088 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
3089 }
3091 int os::Linux::get_node_by_cpu(int cpu_id) {
3092 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3093 return cpu_to_node()->at(cpu_id);
3094 }
3095 return -1;
3096 }
3098 GrowableArray<int>* os::Linux::_cpu_to_node;
3099 GrowableArray<int>* os::Linux::_nindex_to_node;
3100 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3101 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3102 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3103 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3104 os::Linux::numa_available_func_t os::Linux::_numa_available;
3105 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3106 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3107 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3108 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3109 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3110 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3111 unsigned long* os::Linux::_numa_all_nodes;
3112 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3113 struct bitmask* os::Linux::_numa_nodes_ptr;
3115 bool os::pd_uncommit_memory(char* addr, size_t size) {
3116 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3117 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3118 return res != (uintptr_t) MAP_FAILED;
3119 }
3121 static
3122 address get_stack_commited_bottom(address bottom, size_t size) {
3123 address nbot = bottom;
3124 address ntop = bottom + size;
3126 size_t page_sz = os::vm_page_size();
3127 unsigned pages = size / page_sz;
3129 unsigned char vec[1];
3130 unsigned imin = 1, imax = pages + 1, imid;
3131 int mincore_return_value = 0;
3133 assert(imin <= imax, "Unexpected page size");
3135 while (imin < imax) {
3136 imid = (imax + imin) / 2;
3137 nbot = ntop - (imid * page_sz);
3139 // Use a trick with mincore to check whether the page is mapped or not.
3140 // mincore sets vec to 1 if page resides in memory and to 0 if page
3141 // is swapped output but if page we are asking for is unmapped
3142 // it returns -1,ENOMEM
3143 mincore_return_value = mincore(nbot, page_sz, vec);
3145 if (mincore_return_value == -1) {
3146 // Page is not mapped go up
3147 // to find first mapped page
3148 if (errno != EAGAIN) {
3149 assert(errno == ENOMEM, "Unexpected mincore errno");
3150 imax = imid;
3151 }
3152 } else {
3153 // Page is mapped go down
3154 // to find first not mapped page
3155 imin = imid + 1;
3156 }
3157 }
3159 nbot = nbot + page_sz;
3161 // Adjust stack bottom one page up if last checked page is not mapped
3162 if (mincore_return_value == -1) {
3163 nbot = nbot + page_sz;
3164 }
3166 return nbot;
3167 }
3170 // Linux uses a growable mapping for the stack, and if the mapping for
3171 // the stack guard pages is not removed when we detach a thread the
3172 // stack cannot grow beyond the pages where the stack guard was
3173 // mapped. If at some point later in the process the stack expands to
3174 // that point, the Linux kernel cannot expand the stack any further
3175 // because the guard pages are in the way, and a segfault occurs.
3176 //
3177 // However, it's essential not to split the stack region by unmapping
3178 // a region (leaving a hole) that's already part of the stack mapping,
3179 // so if the stack mapping has already grown beyond the guard pages at
3180 // the time we create them, we have to truncate the stack mapping.
3181 // So, we need to know the extent of the stack mapping when
3182 // create_stack_guard_pages() is called.
3184 // We only need this for stacks that are growable: at the time of
3185 // writing thread stacks don't use growable mappings (i.e. those
3186 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3187 // only applies to the main thread.
3189 // If the (growable) stack mapping already extends beyond the point
3190 // where we're going to put our guard pages, truncate the mapping at
3191 // that point by munmap()ping it. This ensures that when we later
3192 // munmap() the guard pages we don't leave a hole in the stack
3193 // mapping. This only affects the main/primordial thread
3195 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3197 if (os::is_primordial_thread()) {
3198 // As we manually grow stack up to bottom inside create_attached_thread(),
3199 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3200 // we don't need to do anything special.
3201 // Check it first, before calling heavy function.
3202 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3203 unsigned char vec[1];
3205 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3206 // Fallback to slow path on all errors, including EAGAIN
3207 stack_extent = (uintptr_t) get_stack_commited_bottom(
3208 os::Linux::initial_thread_stack_bottom(),
3209 (size_t)addr - stack_extent);
3210 }
3212 if (stack_extent < (uintptr_t)addr) {
3213 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3214 }
3215 }
3217 return os::commit_memory(addr, size, !ExecMem);
3218 }
3220 // If this is a growable mapping, remove the guard pages entirely by
3221 // munmap()ping them. If not, just call uncommit_memory(). This only
3222 // affects the main/primordial thread, but guard against future OS changes.
3223 // It's safe to always unmap guard pages for primordial thread because we
3224 // always place it right after end of the mapped region.
3226 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3227 uintptr_t stack_extent, stack_base;
3229 if (os::is_primordial_thread()) {
3230 return ::munmap(addr, size) == 0;
3231 }
3233 return os::uncommit_memory(addr, size);
3234 }
3236 static address _highest_vm_reserved_address = NULL;
3238 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3239 // at 'requested_addr'. If there are existing memory mappings at the same
3240 // location, however, they will be overwritten. If 'fixed' is false,
3241 // 'requested_addr' is only treated as a hint, the return value may or
3242 // may not start from the requested address. Unlike Linux mmap(), this
3243 // function returns NULL to indicate failure.
3244 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3245 char * addr;
3246 int flags;
3248 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3249 if (fixed) {
3250 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3251 flags |= MAP_FIXED;
3252 }
3254 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3255 // touch an uncommitted page. Otherwise, the read/write might
3256 // succeed if we have enough swap space to back the physical page.
3257 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3258 flags, -1, 0);
3260 if (addr != MAP_FAILED) {
3261 // anon_mmap() should only get called during VM initialization,
3262 // don't need lock (actually we can skip locking even it can be called
3263 // from multiple threads, because _highest_vm_reserved_address is just a
3264 // hint about the upper limit of non-stack memory regions.)
3265 if ((address)addr + bytes > _highest_vm_reserved_address) {
3266 _highest_vm_reserved_address = (address)addr + bytes;
3267 }
3268 }
3270 return addr == MAP_FAILED ? NULL : addr;
3271 }
3273 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3274 // (req_addr != NULL) or with a given alignment.
3275 // - bytes shall be a multiple of alignment.
3276 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3277 // - alignment sets the alignment at which memory shall be allocated.
3278 // It must be a multiple of allocation granularity.
3279 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3280 // req_addr or NULL.
3281 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3283 size_t extra_size = bytes;
3284 if (req_addr == NULL && alignment > 0) {
3285 extra_size += alignment;
3286 }
3288 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3289 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3290 -1, 0);
3291 if (start == MAP_FAILED) {
3292 start = NULL;
3293 } else {
3294 if (req_addr != NULL) {
3295 if (start != req_addr) {
3296 ::munmap(start, extra_size);
3297 start = NULL;
3298 }
3299 } else {
3300 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3301 char* const end_aligned = start_aligned + bytes;
3302 char* const end = start + extra_size;
3303 if (start_aligned > start) {
3304 ::munmap(start, start_aligned - start);
3305 }
3306 if (end_aligned < end) {
3307 ::munmap(end_aligned, end - end_aligned);
3308 }
3309 start = start_aligned;
3310 }
3311 }
3312 return start;
3313 }
3315 // Don't update _highest_vm_reserved_address, because there might be memory
3316 // regions above addr + size. If so, releasing a memory region only creates
3317 // a hole in the address space, it doesn't help prevent heap-stack collision.
3318 //
3319 static int anon_munmap(char * addr, size_t size) {
3320 return ::munmap(addr, size) == 0;
3321 }
3323 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3324 size_t alignment_hint) {
3325 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3326 }
3328 bool os::pd_release_memory(char* addr, size_t size) {
3329 return anon_munmap(addr, size);
3330 }
3332 static address highest_vm_reserved_address() {
3333 return _highest_vm_reserved_address;
3334 }
3336 static bool linux_mprotect(char* addr, size_t size, int prot) {
3337 // Linux wants the mprotect address argument to be page aligned.
3338 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3340 // According to SUSv3, mprotect() should only be used with mappings
3341 // established by mmap(), and mmap() always maps whole pages. Unaligned
3342 // 'addr' likely indicates problem in the VM (e.g. trying to change
3343 // protection of malloc'ed or statically allocated memory). Check the
3344 // caller if you hit this assert.
3345 assert(addr == bottom, "sanity check");
3347 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3348 return ::mprotect(bottom, size, prot) == 0;
3349 }
3351 // Set protections specified
3352 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3353 bool is_committed) {
3354 unsigned int p = 0;
3355 switch (prot) {
3356 case MEM_PROT_NONE: p = PROT_NONE; break;
3357 case MEM_PROT_READ: p = PROT_READ; break;
3358 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3359 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3360 default:
3361 ShouldNotReachHere();
3362 }
3363 // is_committed is unused.
3364 return linux_mprotect(addr, bytes, p);
3365 }
3367 bool os::guard_memory(char* addr, size_t size) {
3368 return linux_mprotect(addr, size, PROT_NONE);
3369 }
3371 bool os::unguard_memory(char* addr, size_t size) {
3372 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3373 }
3375 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3376 bool result = false;
3377 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3378 MAP_ANONYMOUS|MAP_PRIVATE,
3379 -1, 0);
3380 if (p != MAP_FAILED) {
3381 void *aligned_p = align_ptr_up(p, page_size);
3383 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3385 munmap(p, page_size * 2);
3386 }
3388 if (warn && !result) {
3389 warning("TransparentHugePages is not supported by the operating system.");
3390 }
3392 return result;
3393 }
3395 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3396 bool result = false;
3397 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3398 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3399 -1, 0);
3401 if (p != MAP_FAILED) {
3402 // We don't know if this really is a huge page or not.
3403 FILE *fp = fopen("/proc/self/maps", "r");
3404 if (fp) {
3405 while (!feof(fp)) {
3406 char chars[257];
3407 long x = 0;
3408 if (fgets(chars, sizeof(chars), fp)) {
3409 if (sscanf(chars, "%lx-%*x", &x) == 1
3410 && x == (long)p) {
3411 if (strstr (chars, "hugepage")) {
3412 result = true;
3413 break;
3414 }
3415 }
3416 }
3417 }
3418 fclose(fp);
3419 }
3420 munmap(p, page_size);
3421 }
3423 if (warn && !result) {
3424 warning("HugeTLBFS is not supported by the operating system.");
3425 }
3427 return result;
3428 }
3430 /*
3431 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3432 *
3433 * From the coredump_filter documentation:
3434 *
3435 * - (bit 0) anonymous private memory
3436 * - (bit 1) anonymous shared memory
3437 * - (bit 2) file-backed private memory
3438 * - (bit 3) file-backed shared memory
3439 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3440 * effective only if the bit 2 is cleared)
3441 * - (bit 5) hugetlb private memory
3442 * - (bit 6) hugetlb shared memory
3443 */
3444 static void set_coredump_filter(void) {
3445 FILE *f;
3446 long cdm;
3448 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3449 return;
3450 }
3452 if (fscanf(f, "%lx", &cdm) != 1) {
3453 fclose(f);
3454 return;
3455 }
3457 rewind(f);
3459 if ((cdm & LARGEPAGES_BIT) == 0) {
3460 cdm |= LARGEPAGES_BIT;
3461 fprintf(f, "%#lx", cdm);
3462 }
3464 fclose(f);
3465 }
3467 // Large page support
3469 static size_t _large_page_size = 0;
3471 size_t os::Linux::find_large_page_size() {
3472 size_t large_page_size = 0;
3474 // large_page_size on Linux is used to round up heap size. x86 uses either
3475 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3476 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3477 // page as large as 256M.
3478 //
3479 // Here we try to figure out page size by parsing /proc/meminfo and looking
3480 // for a line with the following format:
3481 // Hugepagesize: 2048 kB
3482 //
3483 // If we can't determine the value (e.g. /proc is not mounted, or the text
3484 // format has been changed), we'll use the largest page size supported by
3485 // the processor.
3487 #ifndef ZERO
3488 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3489 ARM_ONLY(2 * M) PPC_ONLY(4 * M) MIPS64_ONLY(4 * M); //In MIPS _large_page_size is seted 4*M.
3490 #endif // ZERO
3492 FILE *fp = fopen("/proc/meminfo", "r");
3493 if (fp) {
3494 while (!feof(fp)) {
3495 int x = 0;
3496 char buf[16];
3497 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3498 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3499 large_page_size = x * K;
3500 break;
3501 }
3502 } else {
3503 // skip to next line
3504 for (;;) {
3505 int ch = fgetc(fp);
3506 if (ch == EOF || ch == (int)'\n') break;
3507 }
3508 }
3509 }
3510 fclose(fp);
3511 }
3513 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3514 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3515 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3516 proper_unit_for_byte_size(large_page_size));
3517 }
3519 return large_page_size;
3520 }
3522 size_t os::Linux::setup_large_page_size() {
3523 _large_page_size = Linux::find_large_page_size();
3524 const size_t default_page_size = (size_t)Linux::page_size();
3525 if (_large_page_size > default_page_size) {
3526 _page_sizes[0] = _large_page_size;
3527 _page_sizes[1] = default_page_size;
3528 _page_sizes[2] = 0;
3529 }
3531 return _large_page_size;
3532 }
3534 bool os::Linux::setup_large_page_type(size_t page_size) {
3535 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3536 FLAG_IS_DEFAULT(UseSHM) &&
3537 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3539 // The type of large pages has not been specified by the user.
3541 // Try UseHugeTLBFS and then UseSHM.
3542 UseHugeTLBFS = UseSHM = true;
3544 // Don't try UseTransparentHugePages since there are known
3545 // performance issues with it turned on. This might change in the future.
3546 UseTransparentHugePages = false;
3547 }
3549 if (UseTransparentHugePages) {
3550 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3551 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3552 UseHugeTLBFS = false;
3553 UseSHM = false;
3554 return true;
3555 }
3556 UseTransparentHugePages = false;
3557 }
3559 if (UseHugeTLBFS) {
3560 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3561 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3562 UseSHM = false;
3563 return true;
3564 }
3565 UseHugeTLBFS = false;
3566 }
3568 return UseSHM;
3569 }
3571 void os::large_page_init() {
3572 if (!UseLargePages &&
3573 !UseTransparentHugePages &&
3574 !UseHugeTLBFS &&
3575 !UseSHM) {
3576 // Not using large pages.
3577 return;
3578 }
3580 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3581 // The user explicitly turned off large pages.
3582 // Ignore the rest of the large pages flags.
3583 UseTransparentHugePages = false;
3584 UseHugeTLBFS = false;
3585 UseSHM = false;
3586 return;
3587 }
3589 size_t large_page_size = Linux::setup_large_page_size();
3590 UseLargePages = Linux::setup_large_page_type(large_page_size);
3592 set_coredump_filter();
3593 }
3595 #ifndef SHM_HUGETLB
3596 #define SHM_HUGETLB 04000
3597 #endif
3599 #define shm_warning_format(format, ...) \
3600 do { \
3601 if (UseLargePages && \
3602 (!FLAG_IS_DEFAULT(UseLargePages) || \
3603 !FLAG_IS_DEFAULT(UseSHM) || \
3604 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3605 warning(format, __VA_ARGS__); \
3606 } \
3607 } while (0)
3609 #define shm_warning(str) shm_warning_format("%s", str)
3611 #define shm_warning_with_errno(str) \
3612 do { \
3613 int err = errno; \
3614 shm_warning_format(str " (error = %d)", err); \
3615 } while (0)
3617 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3618 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3620 if (!is_size_aligned(alignment, SHMLBA)) {
3621 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3622 return NULL;
3623 }
3625 // To ensure that we get 'alignment' aligned memory from shmat,
3626 // we pre-reserve aligned virtual memory and then attach to that.
3628 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3629 if (pre_reserved_addr == NULL) {
3630 // Couldn't pre-reserve aligned memory.
3631 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3632 return NULL;
3633 }
3635 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3636 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3638 if ((intptr_t)addr == -1) {
3639 int err = errno;
3640 shm_warning_with_errno("Failed to attach shared memory.");
3642 assert(err != EACCES, "Unexpected error");
3643 assert(err != EIDRM, "Unexpected error");
3644 assert(err != EINVAL, "Unexpected error");
3646 // Since we don't know if the kernel unmapped the pre-reserved memory area
3647 // we can't unmap it, since that would potentially unmap memory that was
3648 // mapped from other threads.
3649 return NULL;
3650 }
3652 return addr;
3653 }
3655 static char* shmat_at_address(int shmid, char* req_addr) {
3656 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3657 assert(false, "Requested address needs to be SHMLBA aligned");
3658 return NULL;
3659 }
3661 char* addr = (char*)shmat(shmid, req_addr, 0);
3663 if ((intptr_t)addr == -1) {
3664 shm_warning_with_errno("Failed to attach shared memory.");
3665 return NULL;
3666 }
3668 return addr;
3669 }
3671 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3672 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3673 if (req_addr != NULL) {
3674 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3675 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3676 return shmat_at_address(shmid, req_addr);
3677 }
3679 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3680 // return large page size aligned memory addresses when req_addr == NULL.
3681 // However, if the alignment is larger than the large page size, we have
3682 // to manually ensure that the memory returned is 'alignment' aligned.
3683 if (alignment > os::large_page_size()) {
3684 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3685 return shmat_with_alignment(shmid, bytes, alignment);
3686 } else {
3687 return shmat_at_address(shmid, NULL);
3688 }
3689 }
3691 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3692 // "exec" is passed in but not used. Creating the shared image for
3693 // the code cache doesn't have an SHM_X executable permission to check.
3694 assert(UseLargePages && UseSHM, "only for SHM large pages");
3695 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3696 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3698 if (!is_size_aligned(bytes, os::large_page_size())) {
3699 return NULL; // Fallback to small pages.
3700 }
3702 // Create a large shared memory region to attach to based on size.
3703 // Currently, size is the total size of the heap.
3704 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3705 if (shmid == -1) {
3706 // Possible reasons for shmget failure:
3707 // 1. shmmax is too small for Java heap.
3708 // > check shmmax value: cat /proc/sys/kernel/shmmax
3709 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3710 // 2. not enough large page memory.
3711 // > check available large pages: cat /proc/meminfo
3712 // > increase amount of large pages:
3713 // echo new_value > /proc/sys/vm/nr_hugepages
3714 // Note 1: different Linux may use different name for this property,
3715 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3716 // Note 2: it's possible there's enough physical memory available but
3717 // they are so fragmented after a long run that they can't
3718 // coalesce into large pages. Try to reserve large pages when
3719 // the system is still "fresh".
3720 shm_warning_with_errno("Failed to reserve shared memory.");
3721 return NULL;
3722 }
3724 // Attach to the region.
3725 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3727 // Remove shmid. If shmat() is successful, the actual shared memory segment
3728 // will be deleted when it's detached by shmdt() or when the process
3729 // terminates. If shmat() is not successful this will remove the shared
3730 // segment immediately.
3731 shmctl(shmid, IPC_RMID, NULL);
3733 return addr;
3734 }
3736 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3737 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3739 bool warn_on_failure = UseLargePages &&
3740 (!FLAG_IS_DEFAULT(UseLargePages) ||
3741 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3742 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3744 if (warn_on_failure) {
3745 char msg[128];
3746 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3747 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3748 warning("%s", msg);
3749 }
3750 }
3752 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3753 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3754 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3755 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3757 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3758 char* addr = (char*)::mmap(req_addr, bytes, prot,
3759 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3760 -1, 0);
3762 if (addr == MAP_FAILED) {
3763 warn_on_large_pages_failure(req_addr, bytes, errno);
3764 return NULL;
3765 }
3767 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3769 return addr;
3770 }
3772 // Reserve memory using mmap(MAP_HUGETLB).
3773 // - bytes shall be a multiple of alignment.
3774 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3775 // - alignment sets the alignment at which memory shall be allocated.
3776 // It must be a multiple of allocation granularity.
3777 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3778 // req_addr or NULL.
3779 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3780 size_t large_page_size = os::large_page_size();
3781 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3783 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3784 assert(is_size_aligned(bytes, alignment), "Must be");
3786 // First reserve - but not commit - the address range in small pages.
3787 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3789 if (start == NULL) {
3790 return NULL;
3791 }
3793 assert(is_ptr_aligned(start, alignment), "Must be");
3795 char* end = start + bytes;
3797 // Find the regions of the allocated chunk that can be promoted to large pages.
3798 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3799 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3801 size_t lp_bytes = lp_end - lp_start;
3803 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3805 if (lp_bytes == 0) {
3806 // The mapped region doesn't even span the start and the end of a large page.
3807 // Fall back to allocate a non-special area.
3808 ::munmap(start, end - start);
3809 return NULL;
3810 }
3812 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3814 void* result;
3816 // Commit small-paged leading area.
3817 if (start != lp_start) {
3818 result = ::mmap(start, lp_start - start, prot,
3819 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3820 -1, 0);
3821 if (result == MAP_FAILED) {
3822 ::munmap(lp_start, end - lp_start);
3823 return NULL;
3824 }
3825 }
3827 // Commit large-paged area.
3828 result = ::mmap(lp_start, lp_bytes, prot,
3829 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3830 -1, 0);
3831 if (result == MAP_FAILED) {
3832 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3833 // If the mmap above fails, the large pages region will be unmapped and we
3834 // have regions before and after with small pages. Release these regions.
3835 //
3836 // | mapped | unmapped | mapped |
3837 // ^ ^ ^ ^
3838 // start lp_start lp_end end
3839 //
3840 ::munmap(start, lp_start - start);
3841 ::munmap(lp_end, end - lp_end);
3842 return NULL;
3843 }
3845 // Commit small-paged trailing area.
3846 if (lp_end != end) {
3847 result = ::mmap(lp_end, end - lp_end, prot,
3848 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3849 -1, 0);
3850 if (result == MAP_FAILED) {
3851 ::munmap(start, lp_end - start);
3852 return NULL;
3853 }
3854 }
3856 return start;
3857 }
3859 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3860 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3861 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3862 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3863 assert(is_power_of_2(os::large_page_size()), "Must be");
3864 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3866 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3867 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3868 } else {
3869 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3870 }
3871 }
3873 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3874 assert(UseLargePages, "only for large pages");
3876 char* addr;
3877 if (UseSHM) {
3878 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3879 } else {
3880 assert(UseHugeTLBFS, "must be");
3881 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3882 }
3884 if (addr != NULL) {
3885 if (UseNUMAInterleaving) {
3886 numa_make_global(addr, bytes);
3887 }
3889 // The memory is committed
3890 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3891 }
3893 return addr;
3894 }
3896 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3897 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3898 return shmdt(base) == 0;
3899 }
3901 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3902 return pd_release_memory(base, bytes);
3903 }
3905 bool os::release_memory_special(char* base, size_t bytes) {
3906 bool res;
3907 if (MemTracker::tracking_level() > NMT_minimal) {
3908 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3909 res = os::Linux::release_memory_special_impl(base, bytes);
3910 if (res) {
3911 tkr.record((address)base, bytes);
3912 }
3914 } else {
3915 res = os::Linux::release_memory_special_impl(base, bytes);
3916 }
3917 return res;
3918 }
3920 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3921 assert(UseLargePages, "only for large pages");
3922 bool res;
3924 if (UseSHM) {
3925 res = os::Linux::release_memory_special_shm(base, bytes);
3926 } else {
3927 assert(UseHugeTLBFS, "must be");
3928 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3929 }
3930 return res;
3931 }
3933 size_t os::large_page_size() {
3934 return _large_page_size;
3935 }
3937 // With SysV SHM the entire memory region must be allocated as shared
3938 // memory.
3939 // HugeTLBFS allows application to commit large page memory on demand.
3940 // However, when committing memory with HugeTLBFS fails, the region
3941 // that was supposed to be committed will lose the old reservation
3942 // and allow other threads to steal that memory region. Because of this
3943 // behavior we can't commit HugeTLBFS memory.
3944 bool os::can_commit_large_page_memory() {
3945 return UseTransparentHugePages;
3946 }
3948 bool os::can_execute_large_page_memory() {
3949 return UseTransparentHugePages || UseHugeTLBFS;
3950 }
3952 // Reserve memory at an arbitrary address, only if that area is
3953 // available (and not reserved for something else).
3955 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3956 const int max_tries = 10;
3957 char* base[max_tries];
3958 size_t size[max_tries];
3959 const size_t gap = 0x000000;
3961 // Assert only that the size is a multiple of the page size, since
3962 // that's all that mmap requires, and since that's all we really know
3963 // about at this low abstraction level. If we need higher alignment,
3964 // we can either pass an alignment to this method or verify alignment
3965 // in one of the methods further up the call chain. See bug 5044738.
3966 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3968 // Repeatedly allocate blocks until the block is allocated at the
3969 // right spot. Give up after max_tries. Note that reserve_memory() will
3970 // automatically update _highest_vm_reserved_address if the call is
3971 // successful. The variable tracks the highest memory address every reserved
3972 // by JVM. It is used to detect heap-stack collision if running with
3973 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3974 // space than needed, it could confuse the collision detecting code. To
3975 // solve the problem, save current _highest_vm_reserved_address and
3976 // calculate the correct value before return.
3977 address old_highest = _highest_vm_reserved_address;
3979 // Linux mmap allows caller to pass an address as hint; give it a try first,
3980 // if kernel honors the hint then we can return immediately.
3981 char * addr = anon_mmap(requested_addr, bytes, false);
3982 if (addr == requested_addr) {
3983 return requested_addr;
3984 }
3986 if (addr != NULL) {
3987 // mmap() is successful but it fails to reserve at the requested address
3988 anon_munmap(addr, bytes);
3989 }
3991 int i;
3992 for (i = 0; i < max_tries; ++i) {
3993 base[i] = reserve_memory(bytes);
3995 if (base[i] != NULL) {
3996 // Is this the block we wanted?
3997 if (base[i] == requested_addr) {
3998 size[i] = bytes;
3999 break;
4000 }
4002 // Does this overlap the block we wanted? Give back the overlapped
4003 // parts and try again.
4005 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
4006 if (top_overlap >= 0 && top_overlap < bytes) {
4007 unmap_memory(base[i], top_overlap);
4008 base[i] += top_overlap;
4009 size[i] = bytes - top_overlap;
4010 } else {
4011 size_t bottom_overlap = base[i] + bytes - requested_addr;
4012 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
4013 unmap_memory(requested_addr, bottom_overlap);
4014 size[i] = bytes - bottom_overlap;
4015 } else {
4016 size[i] = bytes;
4017 }
4018 }
4019 }
4020 }
4022 // Give back the unused reserved pieces.
4024 for (int j = 0; j < i; ++j) {
4025 if (base[j] != NULL) {
4026 unmap_memory(base[j], size[j]);
4027 }
4028 }
4030 if (i < max_tries) {
4031 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
4032 return requested_addr;
4033 } else {
4034 _highest_vm_reserved_address = old_highest;
4035 return NULL;
4036 }
4037 }
4039 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4040 return ::read(fd, buf, nBytes);
4041 }
4043 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
4044 // Solaris uses poll(), linux uses park().
4045 // Poll() is likely a better choice, assuming that Thread.interrupt()
4046 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
4047 // SIGSEGV, see 4355769.
4049 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
4050 assert(thread == Thread::current(), "thread consistency check");
4052 ParkEvent * const slp = thread->_SleepEvent ;
4053 slp->reset() ;
4054 OrderAccess::fence() ;
4056 if (interruptible) {
4057 jlong prevtime = javaTimeNanos();
4059 for (;;) {
4060 if (os::is_interrupted(thread, true)) {
4061 return OS_INTRPT;
4062 }
4064 jlong newtime = javaTimeNanos();
4066 if (newtime - prevtime < 0) {
4067 // time moving backwards, should only happen if no monotonic clock
4068 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4069 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4070 } else {
4071 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4072 }
4074 if(millis <= 0) {
4075 return OS_OK;
4076 }
4078 prevtime = newtime;
4080 {
4081 assert(thread->is_Java_thread(), "sanity check");
4082 JavaThread *jt = (JavaThread *) thread;
4083 ThreadBlockInVM tbivm(jt);
4084 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
4086 jt->set_suspend_equivalent();
4087 // cleared by handle_special_suspend_equivalent_condition() or
4088 // java_suspend_self() via check_and_wait_while_suspended()
4090 slp->park(millis);
4092 // were we externally suspended while we were waiting?
4093 jt->check_and_wait_while_suspended();
4094 }
4095 }
4096 } else {
4097 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4098 jlong prevtime = javaTimeNanos();
4100 for (;;) {
4101 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
4102 // the 1st iteration ...
4103 jlong newtime = javaTimeNanos();
4105 if (newtime - prevtime < 0) {
4106 // time moving backwards, should only happen if no monotonic clock
4107 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4108 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4109 } else {
4110 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4111 }
4113 if(millis <= 0) break ;
4115 prevtime = newtime;
4116 slp->park(millis);
4117 }
4118 return OS_OK ;
4119 }
4120 }
4122 //
4123 // Short sleep, direct OS call.
4124 //
4125 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4126 // sched_yield(2) will actually give up the CPU:
4127 //
4128 // * Alone on this pariticular CPU, keeps running.
4129 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4130 // (pre 2.6.39).
4131 //
4132 // So calling this with 0 is an alternative.
4133 //
4134 void os::naked_short_sleep(jlong ms) {
4135 struct timespec req;
4137 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4138 req.tv_sec = 0;
4139 if (ms > 0) {
4140 req.tv_nsec = (ms % 1000) * 1000000;
4141 }
4142 else {
4143 req.tv_nsec = 1;
4144 }
4146 nanosleep(&req, NULL);
4148 return;
4149 }
4151 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4152 void os::infinite_sleep() {
4153 while (true) { // sleep forever ...
4154 ::sleep(100); // ... 100 seconds at a time
4155 }
4156 }
4158 // Used to convert frequent JVM_Yield() to nops
4159 bool os::dont_yield() {
4160 return DontYieldALot;
4161 }
4163 void os::yield() {
4164 sched_yield();
4165 }
4167 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
4169 void os::yield_all(int attempts) {
4170 // Yields to all threads, including threads with lower priorities
4171 // Threads on Linux are all with same priority. The Solaris style
4172 // os::yield_all() with nanosleep(1ms) is not necessary.
4173 sched_yield();
4174 }
4176 // Called from the tight loops to possibly influence time-sharing heuristics
4177 void os::loop_breaker(int attempts) {
4178 os::yield_all(attempts);
4179 }
4181 ////////////////////////////////////////////////////////////////////////////////
4182 // thread priority support
4184 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4185 // only supports dynamic priority, static priority must be zero. For real-time
4186 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4187 // However, for large multi-threaded applications, SCHED_RR is not only slower
4188 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4189 // of 5 runs - Sep 2005).
4190 //
4191 // The following code actually changes the niceness of kernel-thread/LWP. It
4192 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4193 // not the entire user process, and user level threads are 1:1 mapped to kernel
4194 // threads. It has always been the case, but could change in the future. For
4195 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4196 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4198 int os::java_to_os_priority[CriticalPriority + 1] = {
4199 19, // 0 Entry should never be used
4201 4, // 1 MinPriority
4202 3, // 2
4203 2, // 3
4205 1, // 4
4206 0, // 5 NormPriority
4207 -1, // 6
4209 -2, // 7
4210 -3, // 8
4211 -4, // 9 NearMaxPriority
4213 -5, // 10 MaxPriority
4215 -5 // 11 CriticalPriority
4216 };
4218 static int prio_init() {
4219 if (ThreadPriorityPolicy == 1) {
4220 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4221 // if effective uid is not root. Perhaps, a more elegant way of doing
4222 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4223 if (geteuid() != 0) {
4224 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4225 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4226 }
4227 ThreadPriorityPolicy = 0;
4228 }
4229 }
4230 if (UseCriticalJavaThreadPriority) {
4231 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4232 }
4233 return 0;
4234 }
4236 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4237 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4239 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4240 return (ret == 0) ? OS_OK : OS_ERR;
4241 }
4243 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4244 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4245 *priority_ptr = java_to_os_priority[NormPriority];
4246 return OS_OK;
4247 }
4249 errno = 0;
4250 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4251 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4252 }
4254 // Hint to the underlying OS that a task switch would not be good.
4255 // Void return because it's a hint and can fail.
4256 void os::hint_no_preempt() {}
4258 ////////////////////////////////////////////////////////////////////////////////
4259 // suspend/resume support
4261 // the low-level signal-based suspend/resume support is a remnant from the
4262 // old VM-suspension that used to be for java-suspension, safepoints etc,
4263 // within hotspot. Now there is a single use-case for this:
4264 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4265 // that runs in the watcher thread.
4266 // The remaining code is greatly simplified from the more general suspension
4267 // code that used to be used.
4268 //
4269 // The protocol is quite simple:
4270 // - suspend:
4271 // - sends a signal to the target thread
4272 // - polls the suspend state of the osthread using a yield loop
4273 // - target thread signal handler (SR_handler) sets suspend state
4274 // and blocks in sigsuspend until continued
4275 // - resume:
4276 // - sets target osthread state to continue
4277 // - sends signal to end the sigsuspend loop in the SR_handler
4278 //
4279 // Note that the SR_lock plays no role in this suspend/resume protocol.
4280 //
4282 static void resume_clear_context(OSThread *osthread) {
4283 osthread->set_ucontext(NULL);
4284 osthread->set_siginfo(NULL);
4285 }
4287 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4288 osthread->set_ucontext(context);
4289 osthread->set_siginfo(siginfo);
4290 }
4292 //
4293 // Handler function invoked when a thread's execution is suspended or
4294 // resumed. We have to be careful that only async-safe functions are
4295 // called here (Note: most pthread functions are not async safe and
4296 // should be avoided.)
4297 //
4298 // Note: sigwait() is a more natural fit than sigsuspend() from an
4299 // interface point of view, but sigwait() prevents the signal hander
4300 // from being run. libpthread would get very confused by not having
4301 // its signal handlers run and prevents sigwait()'s use with the
4302 // mutex granting granting signal.
4303 //
4304 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4305 //
4306 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4307 // Save and restore errno to avoid confusing native code with EINTR
4308 // after sigsuspend.
4309 int old_errno = errno;
4311 Thread* thread = Thread::current();
4312 OSThread* osthread = thread->osthread();
4313 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4315 os::SuspendResume::State current = osthread->sr.state();
4316 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4317 suspend_save_context(osthread, siginfo, context);
4319 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4320 os::SuspendResume::State state = osthread->sr.suspended();
4321 if (state == os::SuspendResume::SR_SUSPENDED) {
4322 sigset_t suspend_set; // signals for sigsuspend()
4324 // get current set of blocked signals and unblock resume signal
4325 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4326 sigdelset(&suspend_set, SR_signum);
4328 sr_semaphore.signal();
4329 // wait here until we are resumed
4330 while (1) {
4331 sigsuspend(&suspend_set);
4333 os::SuspendResume::State result = osthread->sr.running();
4334 if (result == os::SuspendResume::SR_RUNNING) {
4335 sr_semaphore.signal();
4336 break;
4337 }
4338 }
4340 } else if (state == os::SuspendResume::SR_RUNNING) {
4341 // request was cancelled, continue
4342 } else {
4343 ShouldNotReachHere();
4344 }
4346 resume_clear_context(osthread);
4347 } else if (current == os::SuspendResume::SR_RUNNING) {
4348 // request was cancelled, continue
4349 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4350 // ignore
4351 } else {
4352 // ignore
4353 }
4355 errno = old_errno;
4356 }
4359 static int SR_initialize() {
4360 struct sigaction act;
4361 char *s;
4362 /* Get signal number to use for suspend/resume */
4363 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4364 int sig = ::strtol(s, 0, 10);
4365 if (sig > 0 || sig < _NSIG) {
4366 SR_signum = sig;
4367 }
4368 }
4370 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4371 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4373 sigemptyset(&SR_sigset);
4374 sigaddset(&SR_sigset, SR_signum);
4376 /* Set up signal handler for suspend/resume */
4377 act.sa_flags = SA_RESTART|SA_SIGINFO;
4378 act.sa_handler = (void (*)(int)) SR_handler;
4380 // SR_signum is blocked by default.
4381 // 4528190 - We also need to block pthread restart signal (32 on all
4382 // supported Linux platforms). Note that LinuxThreads need to block
4383 // this signal for all threads to work properly. So we don't have
4384 // to use hard-coded signal number when setting up the mask.
4385 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4387 if (sigaction(SR_signum, &act, 0) == -1) {
4388 return -1;
4389 }
4391 // Save signal flag
4392 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4393 return 0;
4394 }
4396 static int sr_notify(OSThread* osthread) {
4397 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4398 assert_status(status == 0, status, "pthread_kill");
4399 return status;
4400 }
4402 // "Randomly" selected value for how long we want to spin
4403 // before bailing out on suspending a thread, also how often
4404 // we send a signal to a thread we want to resume
4405 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4406 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4408 // returns true on success and false on error - really an error is fatal
4409 // but this seems the normal response to library errors
4410 static bool do_suspend(OSThread* osthread) {
4411 assert(osthread->sr.is_running(), "thread should be running");
4412 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4414 // mark as suspended and send signal
4415 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4416 // failed to switch, state wasn't running?
4417 ShouldNotReachHere();
4418 return false;
4419 }
4421 if (sr_notify(osthread) != 0) {
4422 ShouldNotReachHere();
4423 }
4425 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4426 while (true) {
4427 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4428 break;
4429 } else {
4430 // timeout
4431 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4432 if (cancelled == os::SuspendResume::SR_RUNNING) {
4433 return false;
4434 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4435 // make sure that we consume the signal on the semaphore as well
4436 sr_semaphore.wait();
4437 break;
4438 } else {
4439 ShouldNotReachHere();
4440 return false;
4441 }
4442 }
4443 }
4445 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4446 return true;
4447 }
4449 static void do_resume(OSThread* osthread) {
4450 assert(osthread->sr.is_suspended(), "thread should be suspended");
4451 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4453 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4454 // failed to switch to WAKEUP_REQUEST
4455 ShouldNotReachHere();
4456 return;
4457 }
4459 while (true) {
4460 if (sr_notify(osthread) == 0) {
4461 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4462 if (osthread->sr.is_running()) {
4463 return;
4464 }
4465 }
4466 } else {
4467 ShouldNotReachHere();
4468 }
4469 }
4471 guarantee(osthread->sr.is_running(), "Must be running!");
4472 }
4474 ////////////////////////////////////////////////////////////////////////////////
4475 // interrupt support
4477 void os::interrupt(Thread* thread) {
4478 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4479 "possibility of dangling Thread pointer");
4481 OSThread* osthread = thread->osthread();
4483 if (!osthread->interrupted()) {
4484 osthread->set_interrupted(true);
4485 // More than one thread can get here with the same value of osthread,
4486 // resulting in multiple notifications. We do, however, want the store
4487 // to interrupted() to be visible to other threads before we execute unpark().
4488 OrderAccess::fence();
4489 ParkEvent * const slp = thread->_SleepEvent ;
4490 if (slp != NULL) slp->unpark() ;
4491 }
4493 // For JSR166. Unpark even if interrupt status already was set
4494 if (thread->is_Java_thread())
4495 ((JavaThread*)thread)->parker()->unpark();
4497 ParkEvent * ev = thread->_ParkEvent ;
4498 if (ev != NULL) ev->unpark() ;
4500 }
4502 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4503 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4504 "possibility of dangling Thread pointer");
4506 OSThread* osthread = thread->osthread();
4508 bool interrupted = osthread->interrupted();
4510 if (interrupted && clear_interrupted) {
4511 osthread->set_interrupted(false);
4512 // consider thread->_SleepEvent->reset() ... optional optimization
4513 }
4515 return interrupted;
4516 }
4518 ///////////////////////////////////////////////////////////////////////////////////
4519 // signal handling (except suspend/resume)
4521 // This routine may be used by user applications as a "hook" to catch signals.
4522 // The user-defined signal handler must pass unrecognized signals to this
4523 // routine, and if it returns true (non-zero), then the signal handler must
4524 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4525 // routine will never retun false (zero), but instead will execute a VM panic
4526 // routine kill the process.
4527 //
4528 // If this routine returns false, it is OK to call it again. This allows
4529 // the user-defined signal handler to perform checks either before or after
4530 // the VM performs its own checks. Naturally, the user code would be making
4531 // a serious error if it tried to handle an exception (such as a null check
4532 // or breakpoint) that the VM was generating for its own correct operation.
4533 //
4534 // This routine may recognize any of the following kinds of signals:
4535 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4536 // It should be consulted by handlers for any of those signals.
4537 //
4538 // The caller of this routine must pass in the three arguments supplied
4539 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4540 // field of the structure passed to sigaction(). This routine assumes that
4541 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4542 //
4543 // Note that the VM will print warnings if it detects conflicting signal
4544 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4545 //
4546 extern "C" JNIEXPORT int
4547 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4548 void* ucontext, int abort_if_unrecognized);
4550 void signalHandler(int sig, siginfo_t* info, void* uc) {
4551 assert(info != NULL && uc != NULL, "it must be old kernel");
4552 int orig_errno = errno; // Preserve errno value over signal handler.
4553 JVM_handle_linux_signal(sig, info, uc, true);
4554 errno = orig_errno;
4555 }
4558 // This boolean allows users to forward their own non-matching signals
4559 // to JVM_handle_linux_signal, harmlessly.
4560 bool os::Linux::signal_handlers_are_installed = false;
4562 // For signal-chaining
4563 struct sigaction os::Linux::sigact[MAXSIGNUM];
4564 unsigned int os::Linux::sigs = 0;
4565 bool os::Linux::libjsig_is_loaded = false;
4566 typedef struct sigaction *(*get_signal_t)(int);
4567 get_signal_t os::Linux::get_signal_action = NULL;
4569 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4570 struct sigaction *actp = NULL;
4572 if (libjsig_is_loaded) {
4573 // Retrieve the old signal handler from libjsig
4574 actp = (*get_signal_action)(sig);
4575 }
4576 if (actp == NULL) {
4577 // Retrieve the preinstalled signal handler from jvm
4578 actp = get_preinstalled_handler(sig);
4579 }
4581 return actp;
4582 }
4584 static bool call_chained_handler(struct sigaction *actp, int sig,
4585 siginfo_t *siginfo, void *context) {
4586 // Call the old signal handler
4587 if (actp->sa_handler == SIG_DFL) {
4588 // It's more reasonable to let jvm treat it as an unexpected exception
4589 // instead of taking the default action.
4590 return false;
4591 } else if (actp->sa_handler != SIG_IGN) {
4592 if ((actp->sa_flags & SA_NODEFER) == 0) {
4593 // automaticlly block the signal
4594 sigaddset(&(actp->sa_mask), sig);
4595 }
4597 sa_handler_t hand = NULL;
4598 sa_sigaction_t sa = NULL;
4599 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4600 // retrieve the chained handler
4601 if (siginfo_flag_set) {
4602 sa = actp->sa_sigaction;
4603 } else {
4604 hand = actp->sa_handler;
4605 }
4607 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4608 actp->sa_handler = SIG_DFL;
4609 }
4611 // try to honor the signal mask
4612 sigset_t oset;
4613 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4615 // call into the chained handler
4616 if (siginfo_flag_set) {
4617 (*sa)(sig, siginfo, context);
4618 } else {
4619 (*hand)(sig);
4620 }
4622 // restore the signal mask
4623 pthread_sigmask(SIG_SETMASK, &oset, 0);
4624 }
4625 // Tell jvm's signal handler the signal is taken care of.
4626 return true;
4627 }
4629 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4630 bool chained = false;
4631 // signal-chaining
4632 if (UseSignalChaining) {
4633 struct sigaction *actp = get_chained_signal_action(sig);
4634 if (actp != NULL) {
4635 chained = call_chained_handler(actp, sig, siginfo, context);
4636 }
4637 }
4638 return chained;
4639 }
4641 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4642 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4643 return &sigact[sig];
4644 }
4645 return NULL;
4646 }
4648 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4649 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4650 sigact[sig] = oldAct;
4651 sigs |= (unsigned int)1 << sig;
4652 }
4654 // for diagnostic
4655 int os::Linux::sigflags[MAXSIGNUM];
4657 int os::Linux::get_our_sigflags(int sig) {
4658 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4659 return sigflags[sig];
4660 }
4662 void os::Linux::set_our_sigflags(int sig, int flags) {
4663 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4664 sigflags[sig] = flags;
4665 }
4667 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4668 // Check for overwrite.
4669 struct sigaction oldAct;
4670 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4672 void* oldhand = oldAct.sa_sigaction
4673 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4674 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4675 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4676 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4677 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4678 if (AllowUserSignalHandlers || !set_installed) {
4679 // Do not overwrite; user takes responsibility to forward to us.
4680 return;
4681 } else if (UseSignalChaining) {
4682 // save the old handler in jvm
4683 save_preinstalled_handler(sig, oldAct);
4684 // libjsig also interposes the sigaction() call below and saves the
4685 // old sigaction on it own.
4686 } else {
4687 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4688 "%#lx for signal %d.", (long)oldhand, sig));
4689 }
4690 }
4692 struct sigaction sigAct;
4693 sigfillset(&(sigAct.sa_mask));
4694 sigAct.sa_handler = SIG_DFL;
4695 if (!set_installed) {
4696 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4697 } else {
4698 sigAct.sa_sigaction = signalHandler;
4699 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4700 }
4701 // Save flags, which are set by ours
4702 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4703 sigflags[sig] = sigAct.sa_flags;
4705 int ret = sigaction(sig, &sigAct, &oldAct);
4706 assert(ret == 0, "check");
4708 void* oldhand2 = oldAct.sa_sigaction
4709 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4710 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4711 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4712 }
4714 // install signal handlers for signals that HotSpot needs to
4715 // handle in order to support Java-level exception handling.
4717 void os::Linux::install_signal_handlers() {
4718 if (!signal_handlers_are_installed) {
4719 signal_handlers_are_installed = true;
4721 // signal-chaining
4722 typedef void (*signal_setting_t)();
4723 signal_setting_t begin_signal_setting = NULL;
4724 signal_setting_t end_signal_setting = NULL;
4725 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4726 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4727 if (begin_signal_setting != NULL) {
4728 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4729 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4730 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4731 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4732 libjsig_is_loaded = true;
4733 assert(UseSignalChaining, "should enable signal-chaining");
4734 }
4735 if (libjsig_is_loaded) {
4736 // Tell libjsig jvm is setting signal handlers
4737 (*begin_signal_setting)();
4738 }
4740 set_signal_handler(SIGSEGV, true);
4741 set_signal_handler(SIGPIPE, true);
4742 set_signal_handler(SIGBUS, true);
4743 set_signal_handler(SIGILL, true);
4744 set_signal_handler(SIGFPE, true);
4745 #if defined(PPC64)
4746 set_signal_handler(SIGTRAP, true);
4747 #endif
4748 set_signal_handler(SIGXFSZ, true);
4750 if (libjsig_is_loaded) {
4751 // Tell libjsig jvm finishes setting signal handlers
4752 (*end_signal_setting)();
4753 }
4755 // We don't activate signal checker if libjsig is in place, we trust ourselves
4756 // and if UserSignalHandler is installed all bets are off.
4757 // Log that signal checking is off only if -verbose:jni is specified.
4758 if (CheckJNICalls) {
4759 if (libjsig_is_loaded) {
4760 if (PrintJNIResolving) {
4761 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4762 }
4763 check_signals = false;
4764 }
4765 if (AllowUserSignalHandlers) {
4766 if (PrintJNIResolving) {
4767 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4768 }
4769 check_signals = false;
4770 }
4771 }
4772 }
4773 }
4775 // This is the fastest way to get thread cpu time on Linux.
4776 // Returns cpu time (user+sys) for any thread, not only for current.
4777 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4778 // It might work on 2.6.10+ with a special kernel/glibc patch.
4779 // For reference, please, see IEEE Std 1003.1-2004:
4780 // http://www.unix.org/single_unix_specification
4782 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4783 struct timespec tp;
4784 int rc = os::Linux::clock_gettime(clockid, &tp);
4785 assert(rc == 0, "clock_gettime is expected to return 0 code");
4787 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4788 }
4790 /////
4791 // glibc on Linux platform uses non-documented flag
4792 // to indicate, that some special sort of signal
4793 // trampoline is used.
4794 // We will never set this flag, and we should
4795 // ignore this flag in our diagnostic
4796 #ifdef SIGNIFICANT_SIGNAL_MASK
4797 #undef SIGNIFICANT_SIGNAL_MASK
4798 #endif
4799 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4801 static const char* get_signal_handler_name(address handler,
4802 char* buf, int buflen) {
4803 int offset = 0;
4804 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4805 if (found) {
4806 // skip directory names
4807 const char *p1, *p2;
4808 p1 = buf;
4809 size_t len = strlen(os::file_separator());
4810 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4811 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4812 } else {
4813 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4814 }
4815 return buf;
4816 }
4818 static void print_signal_handler(outputStream* st, int sig,
4819 char* buf, size_t buflen) {
4820 struct sigaction sa;
4822 sigaction(sig, NULL, &sa);
4824 // See comment for SIGNIFICANT_SIGNAL_MASK define
4825 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4827 st->print("%s: ", os::exception_name(sig, buf, buflen));
4829 address handler = (sa.sa_flags & SA_SIGINFO)
4830 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4831 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4833 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4834 st->print("SIG_DFL");
4835 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4836 st->print("SIG_IGN");
4837 } else {
4838 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4839 }
4841 st->print(", sa_mask[0]=");
4842 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4844 address rh = VMError::get_resetted_sighandler(sig);
4845 // May be, handler was resetted by VMError?
4846 if(rh != NULL) {
4847 handler = rh;
4848 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4849 }
4851 st->print(", sa_flags=");
4852 os::Posix::print_sa_flags(st, sa.sa_flags);
4854 // Check: is it our handler?
4855 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4856 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4857 // It is our signal handler
4858 // check for flags, reset system-used one!
4859 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4860 st->print(
4861 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4862 os::Linux::get_our_sigflags(sig));
4863 }
4864 }
4865 st->cr();
4866 }
4869 #define DO_SIGNAL_CHECK(sig) \
4870 if (!sigismember(&check_signal_done, sig)) \
4871 os::Linux::check_signal_handler(sig)
4873 // This method is a periodic task to check for misbehaving JNI applications
4874 // under CheckJNI, we can add any periodic checks here
4876 void os::run_periodic_checks() {
4878 if (check_signals == false) return;
4880 // SEGV and BUS if overridden could potentially prevent
4881 // generation of hs*.log in the event of a crash, debugging
4882 // such a case can be very challenging, so we absolutely
4883 // check the following for a good measure:
4884 DO_SIGNAL_CHECK(SIGSEGV);
4885 DO_SIGNAL_CHECK(SIGILL);
4886 DO_SIGNAL_CHECK(SIGFPE);
4887 DO_SIGNAL_CHECK(SIGBUS);
4888 DO_SIGNAL_CHECK(SIGPIPE);
4889 DO_SIGNAL_CHECK(SIGXFSZ);
4890 #if defined(PPC64)
4891 DO_SIGNAL_CHECK(SIGTRAP);
4892 #endif
4894 // ReduceSignalUsage allows the user to override these handlers
4895 // see comments at the very top and jvm_solaris.h
4896 if (!ReduceSignalUsage) {
4897 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4898 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4899 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4900 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4901 }
4903 DO_SIGNAL_CHECK(SR_signum);
4904 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4905 }
4907 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4909 static os_sigaction_t os_sigaction = NULL;
4911 void os::Linux::check_signal_handler(int sig) {
4912 char buf[O_BUFLEN];
4913 address jvmHandler = NULL;
4916 struct sigaction act;
4917 if (os_sigaction == NULL) {
4918 // only trust the default sigaction, in case it has been interposed
4919 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4920 if (os_sigaction == NULL) return;
4921 }
4923 os_sigaction(sig, (struct sigaction*)NULL, &act);
4926 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4928 address thisHandler = (act.sa_flags & SA_SIGINFO)
4929 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4930 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4933 switch(sig) {
4934 case SIGSEGV:
4935 case SIGBUS:
4936 case SIGFPE:
4937 case SIGPIPE:
4938 case SIGILL:
4939 case SIGXFSZ:
4940 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4941 break;
4943 case SHUTDOWN1_SIGNAL:
4944 case SHUTDOWN2_SIGNAL:
4945 case SHUTDOWN3_SIGNAL:
4946 case BREAK_SIGNAL:
4947 jvmHandler = (address)user_handler();
4948 break;
4950 case INTERRUPT_SIGNAL:
4951 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4952 break;
4954 default:
4955 if (sig == SR_signum) {
4956 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4957 } else {
4958 return;
4959 }
4960 break;
4961 }
4963 if (thisHandler != jvmHandler) {
4964 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4965 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4966 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4967 // No need to check this sig any longer
4968 sigaddset(&check_signal_done, sig);
4969 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4970 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4971 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4972 exception_name(sig, buf, O_BUFLEN));
4973 }
4974 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4975 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4976 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4977 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4978 // No need to check this sig any longer
4979 sigaddset(&check_signal_done, sig);
4980 }
4982 // Dump all the signal
4983 if (sigismember(&check_signal_done, sig)) {
4984 print_signal_handlers(tty, buf, O_BUFLEN);
4985 }
4986 }
4988 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4990 extern bool signal_name(int signo, char* buf, size_t len);
4992 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4993 if (0 < exception_code && exception_code <= SIGRTMAX) {
4994 // signal
4995 if (!signal_name(exception_code, buf, size)) {
4996 jio_snprintf(buf, size, "SIG%d", exception_code);
4997 }
4998 return buf;
4999 } else {
5000 return NULL;
5001 }
5002 }
5004 // this is called _before_ most of the global arguments have been parsed
5005 void os::init(void) {
5006 char dummy; /* used to get a guess on initial stack address */
5008 // With LinuxThreads the JavaMain thread pid (primordial thread)
5009 // is different than the pid of the java launcher thread.
5010 // So, on Linux, the launcher thread pid is passed to the VM
5011 // via the sun.java.launcher.pid property.
5012 // Use this property instead of getpid() if it was correctly passed.
5013 // See bug 6351349.
5014 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
5016 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
5018 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5020 init_random(1234567);
5022 ThreadCritical::initialize();
5024 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5025 if (Linux::page_size() == -1) {
5026 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
5027 strerror(errno)));
5028 }
5029 init_page_sizes((size_t) Linux::page_size());
5031 Linux::initialize_system_info();
5033 // _main_thread points to the thread that created/loaded the JVM.
5034 Linux::_main_thread = pthread_self();
5036 Linux::clock_init();
5037 initial_time_count = javaTimeNanos();
5039 // pthread_condattr initialization for monotonic clock
5040 int status;
5041 pthread_condattr_t* _condattr = os::Linux::condAttr();
5042 if ((status = pthread_condattr_init(_condattr)) != 0) {
5043 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
5044 }
5045 // Only set the clock if CLOCK_MONOTONIC is available
5046 if (Linux::supports_monotonic_clock()) {
5047 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
5048 if (status == EINVAL) {
5049 warning("Unable to use monotonic clock with relative timed-waits" \
5050 " - changes to the time-of-day clock may have adverse affects");
5051 } else {
5052 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
5053 }
5054 }
5055 }
5056 // else it defaults to CLOCK_REALTIME
5058 pthread_mutex_init(&dl_mutex, NULL);
5060 // If the pagesize of the VM is greater than 8K determine the appropriate
5061 // number of initial guard pages. The user can change this with the
5062 // command line arguments, if needed.
5063 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
5064 StackYellowPages = 1;
5065 StackRedPages = 1;
5066 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
5067 }
5068 }
5070 // To install functions for atexit system call
5071 extern "C" {
5072 static void perfMemory_exit_helper() {
5073 perfMemory_exit();
5074 }
5075 }
5077 void os::pd_init_container_support() {
5078 OSContainer::init();
5079 }
5081 // this is called _after_ the global arguments have been parsed
5082 jint os::init_2(void)
5083 {
5084 Linux::fast_thread_clock_init();
5086 // Allocate a single page and mark it as readable for safepoint polling
5087 #ifdef OPT_SAFEPOINT
5088 void * p = (void *)(0x10000);
5089 address polling_page = (address) ::mmap(p, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5090 #else
5091 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5092 #endif
5093 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
5095 os::set_polling_page( polling_page );
5097 #ifndef PRODUCT
5098 if(Verbose && PrintMiscellaneous)
5099 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5100 #endif
5102 if (!UseMembar) {
5103 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5104 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
5105 os::set_memory_serialize_page( mem_serialize_page );
5107 #ifndef PRODUCT
5108 if(Verbose && PrintMiscellaneous)
5109 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5110 #endif
5111 }
5113 // initialize suspend/resume support - must do this before signal_sets_init()
5114 if (SR_initialize() != 0) {
5115 perror("SR_initialize failed");
5116 return JNI_ERR;
5117 }
5119 Linux::signal_sets_init();
5120 Linux::install_signal_handlers();
5122 // Check minimum allowable stack size for thread creation and to initialize
5123 // the java system classes, including StackOverflowError - depends on page
5124 // size. Add a page for compiler2 recursion in main thread.
5125 // Add in 2*BytesPerWord times page size to account for VM stack during
5126 // class initialization depending on 32 or 64 bit VM.
5128 /* 2014/1/2 Liao: JDK8 requires larger -Xss option.
5129 * TongWeb cannot run with -Xss192K.
5130 * We are not sure whether this causes errors, so simply print a warning. */
5131 size_t min_stack_allowed_jdk6 = os::Linux::min_stack_allowed;
5132 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
5133 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
5134 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
5136 size_t threadStackSizeInBytes = ThreadStackSize * K;
5137 if (threadStackSizeInBytes != 0 &&
5138 threadStackSizeInBytes < min_stack_allowed_jdk6) {
5139 tty->print_cr("\nThe stack size specified is too small, "
5140 "Specify at least %dk",
5141 os::Linux::min_stack_allowed/ K);
5142 return JNI_ERR;
5143 }
5145 // Make the stack size a multiple of the page size so that
5146 // the yellow/red zones can be guarded.
5147 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5148 vm_page_size()));
5150 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5152 #if defined(IA32)
5153 workaround_expand_exec_shield_cs_limit();
5154 #endif
5156 Linux::libpthread_init();
5157 if (PrintMiscellaneous && (Verbose || WizardMode)) {
5158 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
5159 Linux::glibc_version(), Linux::libpthread_version(),
5160 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
5161 }
5163 if (UseNUMA) {
5164 if (!Linux::libnuma_init()) {
5165 UseNUMA = false;
5166 } else {
5167 if ((Linux::numa_max_node() < 1)) {
5168 // There's only one node(they start from 0), disable NUMA.
5169 UseNUMA = false;
5170 }
5171 }
5172 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5173 // we can make the adaptive lgrp chunk resizing work. If the user specified
5174 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
5175 // disable adaptive resizing.
5176 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5177 if (FLAG_IS_DEFAULT(UseNUMA)) {
5178 UseNUMA = false;
5179 } else {
5180 if (FLAG_IS_DEFAULT(UseLargePages) &&
5181 FLAG_IS_DEFAULT(UseSHM) &&
5182 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
5183 UseLargePages = false;
5184 } else {
5185 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
5186 UseAdaptiveSizePolicy = false;
5187 UseAdaptiveNUMAChunkSizing = false;
5188 }
5189 }
5190 }
5191 if (!UseNUMA && ForceNUMA) {
5192 UseNUMA = true;
5193 }
5194 }
5196 if (MaxFDLimit) {
5197 // set the number of file descriptors to max. print out error
5198 // if getrlimit/setrlimit fails but continue regardless.
5199 struct rlimit nbr_files;
5200 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5201 if (status != 0) {
5202 if (PrintMiscellaneous && (Verbose || WizardMode))
5203 perror("os::init_2 getrlimit failed");
5204 } else {
5205 nbr_files.rlim_cur = nbr_files.rlim_max;
5206 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5207 if (status != 0) {
5208 if (PrintMiscellaneous && (Verbose || WizardMode))
5209 perror("os::init_2 setrlimit failed");
5210 }
5211 }
5212 }
5214 // Initialize lock used to serialize thread creation (see os::create_thread)
5215 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5217 // at-exit methods are called in the reverse order of their registration.
5218 // atexit functions are called on return from main or as a result of a
5219 // call to exit(3C). There can be only 32 of these functions registered
5220 // and atexit() does not set errno.
5222 if (PerfAllowAtExitRegistration) {
5223 // only register atexit functions if PerfAllowAtExitRegistration is set.
5224 // atexit functions can be delayed until process exit time, which
5225 // can be problematic for embedded VM situations. Embedded VMs should
5226 // call DestroyJavaVM() to assure that VM resources are released.
5228 // note: perfMemory_exit_helper atexit function may be removed in
5229 // the future if the appropriate cleanup code can be added to the
5230 // VM_Exit VMOperation's doit method.
5231 if (atexit(perfMemory_exit_helper) != 0) {
5232 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5233 }
5234 }
5236 // initialize thread priority policy
5237 prio_init();
5239 return JNI_OK;
5240 }
5242 // Mark the polling page as unreadable
5243 void os::make_polling_page_unreadable(void) {
5244 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5245 fatal("Could not disable polling page");
5246 };
5248 // Mark the polling page as readable
5249 void os::make_polling_page_readable(void) {
5250 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5251 fatal("Could not enable polling page");
5252 }
5253 };
5255 static int os_cpu_count(const cpu_set_t* cpus) {
5256 int count = 0;
5257 // only look up to the number of configured processors
5258 for (int i = 0; i < os::processor_count(); i++) {
5259 if (CPU_ISSET(i, cpus)) {
5260 count++;
5261 }
5262 }
5263 return count;
5264 }
5266 // Get the current number of available processors for this process.
5267 // This value can change at any time during a process's lifetime.
5268 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5269 // If anything goes wrong we fallback to returning the number of online
5270 // processors - which can be greater than the number available to the process.
5271 int os::Linux::active_processor_count() {
5272 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5273 int cpus_size = sizeof(cpu_set_t);
5274 int cpu_count = 0;
5276 // pid 0 means the current thread - which we have to assume represents the process
5277 if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
5278 cpu_count = os_cpu_count(&cpus);
5279 if (PrintActiveCpus) {
5280 tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5281 }
5282 }
5283 else {
5284 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5285 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5286 "which may exceed available processors", strerror(errno), cpu_count);
5287 }
5289 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5290 return cpu_count;
5291 }
5293 // Determine the active processor count from one of
5294 // three different sources:
5295 //
5296 // 1. User option -XX:ActiveProcessorCount
5297 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5298 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5299 //
5300 // Option 1, if specified, will always override.
5301 // If the cgroup subsystem is active and configured, we
5302 // will return the min of the cgroup and option 2 results.
5303 // This is required since tools, such as numactl, that
5304 // alter cpu affinity do not update cgroup subsystem
5305 // cpuset configuration files.
5306 int os::active_processor_count() {
5307 // User has overridden the number of active processors
5308 if (ActiveProcessorCount > 0) {
5309 if (PrintActiveCpus) {
5310 tty->print_cr("active_processor_count: "
5311 "active processor count set by user : %d",
5312 ActiveProcessorCount);
5313 }
5314 return ActiveProcessorCount;
5315 }
5317 int active_cpus;
5318 if (OSContainer::is_containerized()) {
5319 active_cpus = OSContainer::active_processor_count();
5320 if (PrintActiveCpus) {
5321 tty->print_cr("active_processor_count: determined by OSContainer: %d",
5322 active_cpus);
5323 }
5324 } else {
5325 active_cpus = os::Linux::active_processor_count();
5326 }
5328 return active_cpus;
5329 }
5331 void os::set_native_thread_name(const char *name) {
5332 // Not yet implemented.
5333 return;
5334 }
5336 bool os::distribute_processes(uint length, uint* distribution) {
5337 // Not yet implemented.
5338 return false;
5339 }
5341 bool os::bind_to_processor(uint processor_id) {
5342 // Not yet implemented.
5343 return false;
5344 }
5346 ///
5348 void os::SuspendedThreadTask::internal_do_task() {
5349 if (do_suspend(_thread->osthread())) {
5350 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5351 do_task(context);
5352 do_resume(_thread->osthread());
5353 }
5354 }
5356 class PcFetcher : public os::SuspendedThreadTask {
5357 public:
5358 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5359 ExtendedPC result();
5360 protected:
5361 void do_task(const os::SuspendedThreadTaskContext& context);
5362 private:
5363 ExtendedPC _epc;
5364 };
5366 ExtendedPC PcFetcher::result() {
5367 guarantee(is_done(), "task is not done yet.");
5368 return _epc;
5369 }
5371 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5372 Thread* thread = context.thread();
5373 OSThread* osthread = thread->osthread();
5374 if (osthread->ucontext() != NULL) {
5375 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5376 } else {
5377 // NULL context is unexpected, double-check this is the VMThread
5378 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5379 }
5380 }
5382 // Suspends the target using the signal mechanism and then grabs the PC before
5383 // resuming the target. Used by the flat-profiler only
5384 ExtendedPC os::get_thread_pc(Thread* thread) {
5385 // Make sure that it is called by the watcher for the VMThread
5386 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5387 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5389 PcFetcher fetcher(thread);
5390 fetcher.run();
5391 return fetcher.result();
5392 }
5394 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5395 {
5396 if (is_NPTL()) {
5397 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5398 } else {
5399 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5400 // word back to default 64bit precision if condvar is signaled. Java
5401 // wants 53bit precision. Save and restore current value.
5402 int fpu = get_fpu_control_word();
5403 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5404 set_fpu_control_word(fpu);
5405 return status;
5406 }
5407 }
5409 ////////////////////////////////////////////////////////////////////////////////
5410 // debug support
5412 bool os::find(address addr, outputStream* st) {
5413 Dl_info dlinfo;
5414 memset(&dlinfo, 0, sizeof(dlinfo));
5415 if (dladdr(addr, &dlinfo) != 0) {
5416 st->print(PTR_FORMAT ": ", addr);
5417 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5418 st->print("%s+%#x", dlinfo.dli_sname,
5419 addr - (intptr_t)dlinfo.dli_saddr);
5420 } else if (dlinfo.dli_fbase != NULL) {
5421 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5422 } else {
5423 st->print("<absolute address>");
5424 }
5425 if (dlinfo.dli_fname != NULL) {
5426 st->print(" in %s", dlinfo.dli_fname);
5427 }
5428 if (dlinfo.dli_fbase != NULL) {
5429 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5430 }
5431 st->cr();
5433 if (Verbose) {
5434 // decode some bytes around the PC
5435 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5436 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5437 address lowest = (address) dlinfo.dli_sname;
5438 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5439 if (begin < lowest) begin = lowest;
5440 Dl_info dlinfo2;
5441 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5442 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5443 end = (address) dlinfo2.dli_saddr;
5444 Disassembler::decode(begin, end, st);
5445 }
5446 return true;
5447 }
5448 return false;
5449 }
5451 ////////////////////////////////////////////////////////////////////////////////
5452 // misc
5454 // This does not do anything on Linux. This is basically a hook for being
5455 // able to use structured exception handling (thread-local exception filters)
5456 // on, e.g., Win32.
5457 void
5458 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5459 JavaCallArguments* args, Thread* thread) {
5460 f(value, method, args, thread);
5461 }
5463 void os::print_statistics() {
5464 }
5466 int os::message_box(const char* title, const char* message) {
5467 int i;
5468 fdStream err(defaultStream::error_fd());
5469 for (i = 0; i < 78; i++) err.print_raw("=");
5470 err.cr();
5471 err.print_raw_cr(title);
5472 for (i = 0; i < 78; i++) err.print_raw("-");
5473 err.cr();
5474 err.print_raw_cr(message);
5475 for (i = 0; i < 78; i++) err.print_raw("=");
5476 err.cr();
5478 char buf[16];
5479 // Prevent process from exiting upon "read error" without consuming all CPU
5480 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5482 return buf[0] == 'y' || buf[0] == 'Y';
5483 }
5485 int os::stat(const char *path, struct stat *sbuf) {
5486 char pathbuf[MAX_PATH];
5487 if (strlen(path) > MAX_PATH - 1) {
5488 errno = ENAMETOOLONG;
5489 return -1;
5490 }
5491 os::native_path(strcpy(pathbuf, path));
5492 return ::stat(pathbuf, sbuf);
5493 }
5495 bool os::check_heap(bool force) {
5496 return true;
5497 }
5499 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5500 return ::vsnprintf(buf, count, format, args);
5501 }
5503 // Is a (classpath) directory empty?
5504 bool os::dir_is_empty(const char* path) {
5505 DIR *dir = NULL;
5506 struct dirent *ptr;
5508 dir = opendir(path);
5509 if (dir == NULL) return true;
5511 /* Scan the directory */
5512 bool result = true;
5513 char buf[sizeof(struct dirent) + MAX_PATH];
5514 while (result && (ptr = ::readdir(dir)) != NULL) {
5515 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5516 result = false;
5517 }
5518 }
5519 closedir(dir);
5520 return result;
5521 }
5523 // This code originates from JDK's sysOpen and open64_w
5524 // from src/solaris/hpi/src/system_md.c
5526 #ifndef O_DELETE
5527 #define O_DELETE 0x10000
5528 #endif
5530 // Open a file. Unlink the file immediately after open returns
5531 // if the specified oflag has the O_DELETE flag set.
5532 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5534 int os::open(const char *path, int oflag, int mode) {
5536 if (strlen(path) > MAX_PATH - 1) {
5537 errno = ENAMETOOLONG;
5538 return -1;
5539 }
5540 int fd;
5541 int o_delete = (oflag & O_DELETE);
5542 oflag = oflag & ~O_DELETE;
5544 fd = ::open64(path, oflag, mode);
5545 if (fd == -1) return -1;
5547 //If the open succeeded, the file might still be a directory
5548 {
5549 struct stat64 buf64;
5550 int ret = ::fstat64(fd, &buf64);
5551 int st_mode = buf64.st_mode;
5553 if (ret != -1) {
5554 if ((st_mode & S_IFMT) == S_IFDIR) {
5555 errno = EISDIR;
5556 ::close(fd);
5557 return -1;
5558 }
5559 } else {
5560 ::close(fd);
5561 return -1;
5562 }
5563 }
5565 /*
5566 * All file descriptors that are opened in the JVM and not
5567 * specifically destined for a subprocess should have the
5568 * close-on-exec flag set. If we don't set it, then careless 3rd
5569 * party native code might fork and exec without closing all
5570 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5571 * UNIXProcess.c), and this in turn might:
5572 *
5573 * - cause end-of-file to fail to be detected on some file
5574 * descriptors, resulting in mysterious hangs, or
5575 *
5576 * - might cause an fopen in the subprocess to fail on a system
5577 * suffering from bug 1085341.
5578 *
5579 * (Yes, the default setting of the close-on-exec flag is a Unix
5580 * design flaw)
5581 *
5582 * See:
5583 * 1085341: 32-bit stdio routines should support file descriptors >255
5584 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5585 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5586 */
5587 #ifdef FD_CLOEXEC
5588 {
5589 int flags = ::fcntl(fd, F_GETFD);
5590 if (flags != -1)
5591 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5592 }
5593 #endif
5595 if (o_delete != 0) {
5596 ::unlink(path);
5597 }
5598 return fd;
5599 }
5602 // create binary file, rewriting existing file if required
5603 int os::create_binary_file(const char* path, bool rewrite_existing) {
5604 int oflags = O_WRONLY | O_CREAT;
5605 if (!rewrite_existing) {
5606 oflags |= O_EXCL;
5607 }
5608 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5609 }
5611 // return current position of file pointer
5612 jlong os::current_file_offset(int fd) {
5613 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5614 }
5616 // move file pointer to the specified offset
5617 jlong os::seek_to_file_offset(int fd, jlong offset) {
5618 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5619 }
5621 // This code originates from JDK's sysAvailable
5622 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5624 int os::available(int fd, jlong *bytes) {
5625 jlong cur, end;
5626 int mode;
5627 struct stat64 buf64;
5629 if (::fstat64(fd, &buf64) >= 0) {
5630 mode = buf64.st_mode;
5631 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5632 /*
5633 * XXX: is the following call interruptible? If so, this might
5634 * need to go through the INTERRUPT_IO() wrapper as for other
5635 * blocking, interruptible calls in this file.
5636 */
5637 int n;
5638 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5639 *bytes = n;
5640 return 1;
5641 }
5642 }
5643 }
5644 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5645 return 0;
5646 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5647 return 0;
5648 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5649 return 0;
5650 }
5651 *bytes = end - cur;
5652 return 1;
5653 }
5655 int os::socket_available(int fd, jint *pbytes) {
5656 // Linux doc says EINTR not returned, unlike Solaris
5657 int ret = ::ioctl(fd, FIONREAD, pbytes);
5659 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5660 // is expected to return 0 on failure and 1 on success to the jdk.
5661 return (ret < 0) ? 0 : 1;
5662 }
5664 // Map a block of memory.
5665 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5666 char *addr, size_t bytes, bool read_only,
5667 bool allow_exec) {
5668 int prot;
5669 int flags = MAP_PRIVATE;
5671 if (read_only) {
5672 prot = PROT_READ;
5673 } else {
5674 prot = PROT_READ | PROT_WRITE;
5675 }
5677 if (allow_exec) {
5678 prot |= PROT_EXEC;
5679 }
5681 if (addr != NULL) {
5682 flags |= MAP_FIXED;
5683 }
5685 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5686 fd, file_offset);
5687 if (mapped_address == MAP_FAILED) {
5688 return NULL;
5689 }
5690 return mapped_address;
5691 }
5694 // Remap a block of memory.
5695 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5696 char *addr, size_t bytes, bool read_only,
5697 bool allow_exec) {
5698 // same as map_memory() on this OS
5699 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5700 allow_exec);
5701 }
5704 // Unmap a block of memory.
5705 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5706 return munmap(addr, bytes) == 0;
5707 }
5709 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5711 static clockid_t thread_cpu_clockid(Thread* thread) {
5712 pthread_t tid = thread->osthread()->pthread_id();
5713 clockid_t clockid;
5715 // Get thread clockid
5716 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5717 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5718 return clockid;
5719 }
5721 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5722 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5723 // of a thread.
5724 //
5725 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5726 // the fast estimate available on the platform.
5728 jlong os::current_thread_cpu_time() {
5729 if (os::Linux::supports_fast_thread_cpu_time()) {
5730 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5731 } else {
5732 // return user + sys since the cost is the same
5733 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5734 }
5735 }
5737 jlong os::thread_cpu_time(Thread* thread) {
5738 // consistent with what current_thread_cpu_time() returns
5739 if (os::Linux::supports_fast_thread_cpu_time()) {
5740 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5741 } else {
5742 return slow_thread_cpu_time(thread, true /* user + sys */);
5743 }
5744 }
5746 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5747 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5748 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5749 } else {
5750 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5751 }
5752 }
5754 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5755 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5756 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5757 } else {
5758 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5759 }
5760 }
5762 //
5763 // -1 on error.
5764 //
5766 PRAGMA_DIAG_PUSH
5767 PRAGMA_FORMAT_NONLITERAL_IGNORED
5768 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5769 static bool proc_task_unchecked = true;
5770 static const char *proc_stat_path = "/proc/%d/stat";
5771 pid_t tid = thread->osthread()->thread_id();
5772 char *s;
5773 char stat[2048];
5774 int statlen;
5775 char proc_name[64];
5776 int count;
5777 long sys_time, user_time;
5778 char cdummy;
5779 int idummy;
5780 long ldummy;
5781 FILE *fp;
5783 // The /proc/<tid>/stat aggregates per-process usage on
5784 // new Linux kernels 2.6+ where NPTL is supported.
5785 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5786 // See bug 6328462.
5787 // There possibly can be cases where there is no directory
5788 // /proc/self/task, so we check its availability.
5789 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5790 // This is executed only once
5791 proc_task_unchecked = false;
5792 fp = fopen("/proc/self/task", "r");
5793 if (fp != NULL) {
5794 proc_stat_path = "/proc/self/task/%d/stat";
5795 fclose(fp);
5796 }
5797 }
5799 sprintf(proc_name, proc_stat_path, tid);
5800 fp = fopen(proc_name, "r");
5801 if ( fp == NULL ) return -1;
5802 statlen = fread(stat, 1, 2047, fp);
5803 stat[statlen] = '\0';
5804 fclose(fp);
5806 // Skip pid and the command string. Note that we could be dealing with
5807 // weird command names, e.g. user could decide to rename java launcher
5808 // to "java 1.4.2 :)", then the stat file would look like
5809 // 1234 (java 1.4.2 :)) R ... ...
5810 // We don't really need to know the command string, just find the last
5811 // occurrence of ")" and then start parsing from there. See bug 4726580.
5812 s = strrchr(stat, ')');
5813 if (s == NULL ) return -1;
5815 // Skip blank chars
5816 do s++; while (isspace(*s));
5818 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5819 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5820 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5821 &user_time, &sys_time);
5822 if ( count != 13 ) return -1;
5823 if (user_sys_cpu_time) {
5824 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5825 } else {
5826 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5827 }
5828 }
5829 PRAGMA_DIAG_POP
5831 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5832 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5833 info_ptr->may_skip_backward = false; // elapsed time not wall time
5834 info_ptr->may_skip_forward = false; // elapsed time not wall time
5835 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5836 }
5838 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5839 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5840 info_ptr->may_skip_backward = false; // elapsed time not wall time
5841 info_ptr->may_skip_forward = false; // elapsed time not wall time
5842 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5843 }
5845 bool os::is_thread_cpu_time_supported() {
5846 return true;
5847 }
5849 // System loadavg support. Returns -1 if load average cannot be obtained.
5850 // Linux doesn't yet have a (official) notion of processor sets,
5851 // so just return the system wide load average.
5852 int os::loadavg(double loadavg[], int nelem) {
5853 return ::getloadavg(loadavg, nelem);
5854 }
5856 void os::pause() {
5857 char filename[MAX_PATH];
5858 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5859 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5860 } else {
5861 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5862 }
5864 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5865 if (fd != -1) {
5866 struct stat buf;
5867 ::close(fd);
5868 while (::stat(filename, &buf) == 0) {
5869 (void)::poll(NULL, 0, 100);
5870 }
5871 } else {
5872 jio_fprintf(stderr,
5873 "Could not open pause file '%s', continuing immediately.\n", filename);
5874 }
5875 }
5878 // Refer to the comments in os_solaris.cpp park-unpark.
5879 //
5880 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5881 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5882 // For specifics regarding the bug see GLIBC BUGID 261237 :
5883 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5884 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5885 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5886 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5887 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5888 // and monitorenter when we're using 1-0 locking. All those operations may result in
5889 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5890 // of libpthread avoids the problem, but isn't practical.
5891 //
5892 // Possible remedies:
5893 //
5894 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5895 // This is palliative and probabilistic, however. If the thread is preempted
5896 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5897 // than the minimum period may have passed, and the abstime may be stale (in the
5898 // past) resultin in a hang. Using this technique reduces the odds of a hang
5899 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5900 //
5901 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5902 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5903 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5904 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5905 // thread.
5906 //
5907 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5908 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5909 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5910 // This also works well. In fact it avoids kernel-level scalability impediments
5911 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5912 // timers in a graceful fashion.
5913 //
5914 // 4. When the abstime value is in the past it appears that control returns
5915 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5916 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5917 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5918 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5919 // It may be possible to avoid reinitialization by checking the return
5920 // value from pthread_cond_timedwait(). In addition to reinitializing the
5921 // condvar we must establish the invariant that cond_signal() is only called
5922 // within critical sections protected by the adjunct mutex. This prevents
5923 // cond_signal() from "seeing" a condvar that's in the midst of being
5924 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5925 // desirable signal-after-unlock optimization that avoids futile context switching.
5926 //
5927 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5928 // structure when a condvar is used or initialized. cond_destroy() would
5929 // release the helper structure. Our reinitialize-after-timedwait fix
5930 // put excessive stress on malloc/free and locks protecting the c-heap.
5931 //
5932 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5933 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5934 // and only enabling the work-around for vulnerable environments.
5936 // utility to compute the abstime argument to timedwait:
5937 // millis is the relative timeout time
5938 // abstime will be the absolute timeout time
5939 // TODO: replace compute_abstime() with unpackTime()
5941 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5942 if (millis < 0) millis = 0;
5944 jlong seconds = millis / 1000;
5945 millis %= 1000;
5946 if (seconds > 50000000) { // see man cond_timedwait(3T)
5947 seconds = 50000000;
5948 }
5950 if (os::Linux::supports_monotonic_clock()) {
5951 struct timespec now;
5952 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5953 assert_status(status == 0, status, "clock_gettime");
5954 abstime->tv_sec = now.tv_sec + seconds;
5955 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5956 if (nanos >= NANOSECS_PER_SEC) {
5957 abstime->tv_sec += 1;
5958 nanos -= NANOSECS_PER_SEC;
5959 }
5960 abstime->tv_nsec = nanos;
5961 } else {
5962 struct timeval now;
5963 int status = gettimeofday(&now, NULL);
5964 assert(status == 0, "gettimeofday");
5965 abstime->tv_sec = now.tv_sec + seconds;
5966 long usec = now.tv_usec + millis * 1000;
5967 if (usec >= 1000000) {
5968 abstime->tv_sec += 1;
5969 usec -= 1000000;
5970 }
5971 abstime->tv_nsec = usec * 1000;
5972 }
5973 return abstime;
5974 }
5977 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5978 // Conceptually TryPark() should be equivalent to park(0).
5980 int os::PlatformEvent::TryPark() {
5981 for (;;) {
5982 const int v = _Event ;
5983 guarantee ((v == 0) || (v == 1), "invariant") ;
5984 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5985 }
5986 }
5988 void os::PlatformEvent::park() { // AKA "down()"
5989 // Invariant: Only the thread associated with the Event/PlatformEvent
5990 // may call park().
5991 // TODO: assert that _Assoc != NULL or _Assoc == Self
5992 int v ;
5993 for (;;) {
5994 v = _Event ;
5995 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5996 }
5997 guarantee (v >= 0, "invariant") ;
5998 if (v == 0) {
5999 // Do this the hard way by blocking ...
6000 int status = pthread_mutex_lock(_mutex);
6001 assert_status(status == 0, status, "mutex_lock");
6002 guarantee (_nParked == 0, "invariant") ;
6003 ++ _nParked ;
6004 while (_Event < 0) {
6005 status = pthread_cond_wait(_cond, _mutex);
6006 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
6007 // Treat this the same as if the wait was interrupted
6008 if (status == ETIME) { status = EINTR; }
6009 assert_status(status == 0 || status == EINTR, status, "cond_wait");
6010 }
6011 -- _nParked ;
6013 _Event = 0 ;
6014 status = pthread_mutex_unlock(_mutex);
6015 assert_status(status == 0, status, "mutex_unlock");
6016 // Paranoia to ensure our locked and lock-free paths interact
6017 // correctly with each other.
6018 OrderAccess::fence();
6019 }
6020 guarantee (_Event >= 0, "invariant") ;
6021 }
6023 int os::PlatformEvent::park(jlong millis) {
6024 guarantee (_nParked == 0, "invariant") ;
6026 int v ;
6027 for (;;) {
6028 v = _Event ;
6029 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6030 }
6031 guarantee (v >= 0, "invariant") ;
6032 if (v != 0) return OS_OK ;
6034 // We do this the hard way, by blocking the thread.
6035 // Consider enforcing a minimum timeout value.
6036 struct timespec abst;
6037 compute_abstime(&abst, millis);
6039 int ret = OS_TIMEOUT;
6040 int status = pthread_mutex_lock(_mutex);
6041 assert_status(status == 0, status, "mutex_lock");
6042 guarantee (_nParked == 0, "invariant") ;
6043 ++_nParked ;
6045 // Object.wait(timo) will return because of
6046 // (a) notification
6047 // (b) timeout
6048 // (c) thread.interrupt
6049 //
6050 // Thread.interrupt and object.notify{All} both call Event::set.
6051 // That is, we treat thread.interrupt as a special case of notification.
6052 // The underlying Solaris implementation, cond_timedwait, admits
6053 // spurious/premature wakeups, but the JLS/JVM spec prevents the
6054 // JVM from making those visible to Java code. As such, we must
6055 // filter out spurious wakeups. We assume all ETIME returns are valid.
6056 //
6057 // TODO: properly differentiate simultaneous notify+interrupt.
6058 // In that case, we should propagate the notify to another waiter.
6060 while (_Event < 0) {
6061 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
6062 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6063 pthread_cond_destroy (_cond);
6064 pthread_cond_init (_cond, os::Linux::condAttr()) ;
6065 }
6066 assert_status(status == 0 || status == EINTR ||
6067 status == ETIME || status == ETIMEDOUT,
6068 status, "cond_timedwait");
6069 if (!FilterSpuriousWakeups) break ; // previous semantics
6070 if (status == ETIME || status == ETIMEDOUT) break ;
6071 // We consume and ignore EINTR and spurious wakeups.
6072 }
6073 --_nParked ;
6074 if (_Event >= 0) {
6075 ret = OS_OK;
6076 }
6077 _Event = 0 ;
6078 status = pthread_mutex_unlock(_mutex);
6079 assert_status(status == 0, status, "mutex_unlock");
6080 assert (_nParked == 0, "invariant") ;
6081 // Paranoia to ensure our locked and lock-free paths interact
6082 // correctly with each other.
6083 OrderAccess::fence();
6084 return ret;
6085 }
6087 void os::PlatformEvent::unpark() {
6088 // Transitions for _Event:
6089 // 0 :=> 1
6090 // 1 :=> 1
6091 // -1 :=> either 0 or 1; must signal target thread
6092 // That is, we can safely transition _Event from -1 to either
6093 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
6094 // unpark() calls.
6095 // See also: "Semaphores in Plan 9" by Mullender & Cox
6096 //
6097 // Note: Forcing a transition from "-1" to "1" on an unpark() means
6098 // that it will take two back-to-back park() calls for the owning
6099 // thread to block. This has the benefit of forcing a spurious return
6100 // from the first park() call after an unpark() call which will help
6101 // shake out uses of park() and unpark() without condition variables.
6103 if (Atomic::xchg(1, &_Event) >= 0) return;
6105 // Wait for the thread associated with the event to vacate
6106 int status = pthread_mutex_lock(_mutex);
6107 assert_status(status == 0, status, "mutex_lock");
6108 int AnyWaiters = _nParked;
6109 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6110 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
6111 AnyWaiters = 0;
6112 pthread_cond_signal(_cond);
6113 }
6114 status = pthread_mutex_unlock(_mutex);
6115 assert_status(status == 0, status, "mutex_unlock");
6116 if (AnyWaiters != 0) {
6117 status = pthread_cond_signal(_cond);
6118 assert_status(status == 0, status, "cond_signal");
6119 }
6121 // Note that we signal() _after dropping the lock for "immortal" Events.
6122 // This is safe and avoids a common class of futile wakeups. In rare
6123 // circumstances this can cause a thread to return prematurely from
6124 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
6125 // simply re-test the condition and re-park itself.
6126 }
6129 // JSR166
6130 // -------------------------------------------------------
6132 /*
6133 * The solaris and linux implementations of park/unpark are fairly
6134 * conservative for now, but can be improved. They currently use a
6135 * mutex/condvar pair, plus a a count.
6136 * Park decrements count if > 0, else does a condvar wait. Unpark
6137 * sets count to 1 and signals condvar. Only one thread ever waits
6138 * on the condvar. Contention seen when trying to park implies that someone
6139 * is unparking you, so don't wait. And spurious returns are fine, so there
6140 * is no need to track notifications.
6141 */
6143 /*
6144 * This code is common to linux and solaris and will be moved to a
6145 * common place in dolphin.
6146 *
6147 * The passed in time value is either a relative time in nanoseconds
6148 * or an absolute time in milliseconds. Either way it has to be unpacked
6149 * into suitable seconds and nanoseconds components and stored in the
6150 * given timespec structure.
6151 * Given time is a 64-bit value and the time_t used in the timespec is only
6152 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6153 * overflow if times way in the future are given. Further on Solaris versions
6154 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6155 * number of seconds, in abstime, is less than current_time + 100,000,000.
6156 * As it will be 28 years before "now + 100000000" will overflow we can
6157 * ignore overflow and just impose a hard-limit on seconds using the value
6158 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6159 * years from "now".
6160 */
6162 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6163 assert (time > 0, "convertTime");
6164 time_t max_secs = 0;
6166 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
6167 struct timeval now;
6168 int status = gettimeofday(&now, NULL);
6169 assert(status == 0, "gettimeofday");
6171 max_secs = now.tv_sec + MAX_SECS;
6173 if (isAbsolute) {
6174 jlong secs = time / 1000;
6175 if (secs > max_secs) {
6176 absTime->tv_sec = max_secs;
6177 } else {
6178 absTime->tv_sec = secs;
6179 }
6180 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6181 } else {
6182 jlong secs = time / NANOSECS_PER_SEC;
6183 if (secs >= MAX_SECS) {
6184 absTime->tv_sec = max_secs;
6185 absTime->tv_nsec = 0;
6186 } else {
6187 absTime->tv_sec = now.tv_sec + secs;
6188 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6189 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6190 absTime->tv_nsec -= NANOSECS_PER_SEC;
6191 ++absTime->tv_sec; // note: this must be <= max_secs
6192 }
6193 }
6194 }
6195 } else {
6196 // must be relative using monotonic clock
6197 struct timespec now;
6198 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6199 assert_status(status == 0, status, "clock_gettime");
6200 max_secs = now.tv_sec + MAX_SECS;
6201 jlong secs = time / NANOSECS_PER_SEC;
6202 if (secs >= MAX_SECS) {
6203 absTime->tv_sec = max_secs;
6204 absTime->tv_nsec = 0;
6205 } else {
6206 absTime->tv_sec = now.tv_sec + secs;
6207 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
6208 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6209 absTime->tv_nsec -= NANOSECS_PER_SEC;
6210 ++absTime->tv_sec; // note: this must be <= max_secs
6211 }
6212 }
6213 }
6214 assert(absTime->tv_sec >= 0, "tv_sec < 0");
6215 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6216 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6217 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6218 }
6220 void Parker::park(bool isAbsolute, jlong time) {
6221 // Ideally we'd do something useful while spinning, such
6222 // as calling unpackTime().
6224 // Optional fast-path check:
6225 // Return immediately if a permit is available.
6226 // We depend on Atomic::xchg() having full barrier semantics
6227 // since we are doing a lock-free update to _counter.
6228 if (Atomic::xchg(0, &_counter) > 0) return;
6230 Thread* thread = Thread::current();
6231 assert(thread->is_Java_thread(), "Must be JavaThread");
6232 JavaThread *jt = (JavaThread *)thread;
6234 // Optional optimization -- avoid state transitions if there's an interrupt pending.
6235 // Check interrupt before trying to wait
6236 if (Thread::is_interrupted(thread, false)) {
6237 return;
6238 }
6240 // Next, demultiplex/decode time arguments
6241 timespec absTime;
6242 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6243 return;
6244 }
6245 if (time > 0) {
6246 unpackTime(&absTime, isAbsolute, time);
6247 }
6250 // Enter safepoint region
6251 // Beware of deadlocks such as 6317397.
6252 // The per-thread Parker:: mutex is a classic leaf-lock.
6253 // In particular a thread must never block on the Threads_lock while
6254 // holding the Parker:: mutex. If safepoints are pending both the
6255 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6256 ThreadBlockInVM tbivm(jt);
6258 // Don't wait if cannot get lock since interference arises from
6259 // unblocking. Also. check interrupt before trying wait
6260 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6261 return;
6262 }
6264 int status ;
6265 if (_counter > 0) { // no wait needed
6266 _counter = 0;
6267 status = pthread_mutex_unlock(_mutex);
6268 assert (status == 0, "invariant") ;
6269 // Paranoia to ensure our locked and lock-free paths interact
6270 // correctly with each other and Java-level accesses.
6271 OrderAccess::fence();
6272 return;
6273 }
6275 #ifdef ASSERT
6276 // Don't catch signals while blocked; let the running threads have the signals.
6277 // (This allows a debugger to break into the running thread.)
6278 sigset_t oldsigs;
6279 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6280 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6281 #endif
6283 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6284 jt->set_suspend_equivalent();
6285 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6287 assert(_cur_index == -1, "invariant");
6288 if (time == 0) {
6289 _cur_index = REL_INDEX; // arbitrary choice when not timed
6290 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6291 } else {
6292 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6293 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6294 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6295 pthread_cond_destroy (&_cond[_cur_index]) ;
6296 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6297 }
6298 }
6299 _cur_index = -1;
6300 assert_status(status == 0 || status == EINTR ||
6301 status == ETIME || status == ETIMEDOUT,
6302 status, "cond_timedwait");
6304 #ifdef ASSERT
6305 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6306 #endif
6308 _counter = 0 ;
6309 status = pthread_mutex_unlock(_mutex) ;
6310 assert_status(status == 0, status, "invariant") ;
6311 // Paranoia to ensure our locked and lock-free paths interact
6312 // correctly with each other and Java-level accesses.
6313 OrderAccess::fence();
6315 // If externally suspended while waiting, re-suspend
6316 if (jt->handle_special_suspend_equivalent_condition()) {
6317 jt->java_suspend_self();
6318 }
6319 }
6321 void Parker::unpark() {
6322 int s, status ;
6323 status = pthread_mutex_lock(_mutex);
6324 assert (status == 0, "invariant") ;
6325 s = _counter;
6326 _counter = 1;
6327 if (s < 1) {
6328 // thread might be parked
6329 if (_cur_index != -1) {
6330 // thread is definitely parked
6331 if (WorkAroundNPTLTimedWaitHang) {
6332 status = pthread_cond_signal (&_cond[_cur_index]);
6333 assert (status == 0, "invariant");
6334 status = pthread_mutex_unlock(_mutex);
6335 assert (status == 0, "invariant");
6336 } else {
6337 // must capture correct index before unlocking
6338 int index = _cur_index;
6339 status = pthread_mutex_unlock(_mutex);
6340 assert (status == 0, "invariant");
6341 status = pthread_cond_signal (&_cond[index]);
6342 assert (status == 0, "invariant");
6343 }
6344 } else {
6345 pthread_mutex_unlock(_mutex);
6346 assert (status == 0, "invariant") ;
6347 }
6348 } else {
6349 pthread_mutex_unlock(_mutex);
6350 assert (status == 0, "invariant") ;
6351 }
6352 }
6355 extern char** environ;
6357 // Run the specified command in a separate process. Return its exit value,
6358 // or -1 on failure (e.g. can't fork a new process).
6359 // Unlike system(), this function can be called from signal handler. It
6360 // doesn't block SIGINT et al.
6361 int os::fork_and_exec(char* cmd, bool use_vfork_if_available) {
6362 const char * argv[4] = {"sh", "-c", cmd, NULL};
6364 pid_t pid ;
6366 if (use_vfork_if_available) {
6367 pid = vfork();
6368 } else {
6369 pid = fork();
6370 }
6372 if (pid < 0) {
6373 // fork failed
6374 return -1;
6376 } else if (pid == 0) {
6377 // child process
6379 execve("/bin/sh", (char* const*)argv, environ);
6381 // execve failed
6382 _exit(-1);
6384 } else {
6385 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6386 // care about the actual exit code, for now.
6388 int status;
6390 // Wait for the child process to exit. This returns immediately if
6391 // the child has already exited. */
6392 while (waitpid(pid, &status, 0) < 0) {
6393 switch (errno) {
6394 case ECHILD: return 0;
6395 case EINTR: break;
6396 default: return -1;
6397 }
6398 }
6400 if (WIFEXITED(status)) {
6401 // The child exited normally; get its exit code.
6402 return WEXITSTATUS(status);
6403 } else if (WIFSIGNALED(status)) {
6404 // The child exited because of a signal
6405 // The best value to return is 0x80 + signal number,
6406 // because that is what all Unix shells do, and because
6407 // it allows callers to distinguish between process exit and
6408 // process death by signal.
6409 return 0x80 + WTERMSIG(status);
6410 } else {
6411 // Unknown exit code; pass it through
6412 return status;
6413 }
6414 }
6415 }
6417 // is_headless_jre()
6418 //
6419 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6420 // in order to report if we are running in a headless jre
6421 //
6422 // Since JDK8 xawt/libmawt.so was moved into the same directory
6423 // as libawt.so, and renamed libawt_xawt.so
6424 //
6425 bool os::is_headless_jre() {
6426 struct stat statbuf;
6427 char buf[MAXPATHLEN];
6428 char libmawtpath[MAXPATHLEN];
6429 const char *xawtstr = "/xawt/libmawt.so";
6430 const char *new_xawtstr = "/libawt_xawt.so";
6431 char *p;
6433 // Get path to libjvm.so
6434 os::jvm_path(buf, sizeof(buf));
6436 // Get rid of libjvm.so
6437 p = strrchr(buf, '/');
6438 if (p == NULL) return false;
6439 else *p = '\0';
6441 // Get rid of client or server
6442 p = strrchr(buf, '/');
6443 if (p == NULL) return false;
6444 else *p = '\0';
6446 // check xawt/libmawt.so
6447 strcpy(libmawtpath, buf);
6448 strcat(libmawtpath, xawtstr);
6449 if (::stat(libmawtpath, &statbuf) == 0) return false;
6451 // check libawt_xawt.so
6452 strcpy(libmawtpath, buf);
6453 strcat(libmawtpath, new_xawtstr);
6454 if (::stat(libmawtpath, &statbuf) == 0) return false;
6456 return true;
6457 }
6459 // Get the default path to the core file
6460 // Returns the length of the string
6461 int os::get_core_path(char* buffer, size_t bufferSize) {
6462 const char* p = get_current_directory(buffer, bufferSize);
6464 if (p == NULL) {
6465 assert(p != NULL, "failed to get current directory");
6466 return 0;
6467 }
6469 return strlen(buffer);
6470 }
6472 /////////////// Unit tests ///////////////
6474 #ifndef PRODUCT
6476 #define test_log(...) \
6477 do {\
6478 if (VerboseInternalVMTests) { \
6479 tty->print_cr(__VA_ARGS__); \
6480 tty->flush(); \
6481 }\
6482 } while (false)
6484 class TestReserveMemorySpecial : AllStatic {
6485 public:
6486 static void small_page_write(void* addr, size_t size) {
6487 size_t page_size = os::vm_page_size();
6489 char* end = (char*)addr + size;
6490 for (char* p = (char*)addr; p < end; p += page_size) {
6491 *p = 1;
6492 }
6493 }
6495 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6496 if (!UseHugeTLBFS) {
6497 return;
6498 }
6500 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6502 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6504 if (addr != NULL) {
6505 small_page_write(addr, size);
6507 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6508 }
6509 }
6511 static void test_reserve_memory_special_huge_tlbfs_only() {
6512 if (!UseHugeTLBFS) {
6513 return;
6514 }
6516 size_t lp = os::large_page_size();
6518 for (size_t size = lp; size <= lp * 10; size += lp) {
6519 test_reserve_memory_special_huge_tlbfs_only(size);
6520 }
6521 }
6523 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6524 size_t lp = os::large_page_size();
6525 size_t ag = os::vm_allocation_granularity();
6527 // sizes to test
6528 const size_t sizes[] = {
6529 lp, lp + ag, lp + lp / 2, lp * 2,
6530 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6531 lp * 10, lp * 10 + lp / 2
6532 };
6533 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6535 // For each size/alignment combination, we test three scenarios:
6536 // 1) with req_addr == NULL
6537 // 2) with a non-null req_addr at which we expect to successfully allocate
6538 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6539 // expect the allocation to either fail or to ignore req_addr
6541 // Pre-allocate two areas; they shall be as large as the largest allocation
6542 // and aligned to the largest alignment we will be testing.
6543 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6544 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6545 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6546 -1, 0);
6547 assert(mapping1 != MAP_FAILED, "should work");
6549 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6550 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6551 -1, 0);
6552 assert(mapping2 != MAP_FAILED, "should work");
6554 // Unmap the first mapping, but leave the second mapping intact: the first
6555 // mapping will serve as a value for a "good" req_addr (case 2). The second
6556 // mapping, still intact, as "bad" req_addr (case 3).
6557 ::munmap(mapping1, mapping_size);
6559 // Case 1
6560 test_log("%s, req_addr NULL:", __FUNCTION__);
6561 test_log("size align result");
6563 for (int i = 0; i < num_sizes; i++) {
6564 const size_t size = sizes[i];
6565 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6566 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6567 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6568 size, alignment, p, (p != NULL ? "" : "(failed)"));
6569 if (p != NULL) {
6570 assert(is_ptr_aligned(p, alignment), "must be");
6571 small_page_write(p, size);
6572 os::Linux::release_memory_special_huge_tlbfs(p, size);
6573 }
6574 }
6575 }
6577 // Case 2
6578 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6579 test_log("size align req_addr result");
6581 for (int i = 0; i < num_sizes; i++) {
6582 const size_t size = sizes[i];
6583 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6584 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6585 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6586 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6587 size, alignment, req_addr, p,
6588 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6589 if (p != NULL) {
6590 assert(p == req_addr, "must be");
6591 small_page_write(p, size);
6592 os::Linux::release_memory_special_huge_tlbfs(p, size);
6593 }
6594 }
6595 }
6597 // Case 3
6598 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6599 test_log("size align req_addr result");
6601 for (int i = 0; i < num_sizes; i++) {
6602 const size_t size = sizes[i];
6603 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6604 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6605 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6606 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6607 size, alignment, req_addr, p,
6608 ((p != NULL ? "" : "(failed)")));
6609 // as the area around req_addr contains already existing mappings, the API should always
6610 // return NULL (as per contract, it cannot return another address)
6611 assert(p == NULL, "must be");
6612 }
6613 }
6615 ::munmap(mapping2, mapping_size);
6617 }
6619 static void test_reserve_memory_special_huge_tlbfs() {
6620 if (!UseHugeTLBFS) {
6621 return;
6622 }
6624 test_reserve_memory_special_huge_tlbfs_only();
6625 test_reserve_memory_special_huge_tlbfs_mixed();
6626 }
6628 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6629 if (!UseSHM) {
6630 return;
6631 }
6633 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6635 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6637 if (addr != NULL) {
6638 assert(is_ptr_aligned(addr, alignment), "Check");
6639 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6641 small_page_write(addr, size);
6643 os::Linux::release_memory_special_shm(addr, size);
6644 }
6645 }
6647 static void test_reserve_memory_special_shm() {
6648 size_t lp = os::large_page_size();
6649 size_t ag = os::vm_allocation_granularity();
6651 for (size_t size = ag; size < lp * 3; size += ag) {
6652 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6653 test_reserve_memory_special_shm(size, alignment);
6654 }
6655 }
6656 }
6658 static void test() {
6659 test_reserve_memory_special_huge_tlbfs();
6660 test_reserve_memory_special_shm();
6661 }
6662 };
6664 void TestReserveMemorySpecial_test() {
6665 TestReserveMemorySpecial::test();
6666 }
6668 #endif