Mon, 15 Oct 2018 14:58:52 +0100
Merge
1 /*
2 * Copyright (c) 1999, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 // no precompiled headers
26 #include "classfile/classLoader.hpp"
27 #include "classfile/systemDictionary.hpp"
28 #include "classfile/vmSymbols.hpp"
29 #include "code/icBuffer.hpp"
30 #include "code/vtableStubs.hpp"
31 #include "compiler/compileBroker.hpp"
32 #include "compiler/disassembler.hpp"
33 #include "interpreter/interpreter.hpp"
34 #include "jvm_linux.h"
35 #include "memory/allocation.inline.hpp"
36 #include "memory/filemap.hpp"
37 #include "mutex_linux.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "os_share_linux.hpp"
40 #include "osContainer_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/extendedPC.hpp"
46 #include "runtime/globals.hpp"
47 #include "runtime/interfaceSupport.hpp"
48 #include "runtime/init.hpp"
49 #include "runtime/java.hpp"
50 #include "runtime/javaCalls.hpp"
51 #include "runtime/mutexLocker.hpp"
52 #include "runtime/objectMonitor.hpp"
53 #include "runtime/orderAccess.inline.hpp"
54 #include "runtime/osThread.hpp"
55 #include "runtime/perfMemory.hpp"
56 #include "runtime/sharedRuntime.hpp"
57 #include "runtime/statSampler.hpp"
58 #include "runtime/stubRoutines.hpp"
59 #include "runtime/thread.inline.hpp"
60 #include "runtime/threadCritical.hpp"
61 #include "runtime/timer.hpp"
62 #include "services/attachListener.hpp"
63 #include "services/memTracker.hpp"
64 #include "services/runtimeService.hpp"
65 #include "utilities/decoder.hpp"
66 #include "utilities/defaultStream.hpp"
67 #include "utilities/events.hpp"
68 #include "utilities/elfFile.hpp"
69 #include "utilities/growableArray.hpp"
70 #include "utilities/vmError.hpp"
72 // put OS-includes here
73 # include <sys/types.h>
74 # include <sys/mman.h>
75 # include <sys/stat.h>
76 # include <sys/select.h>
77 # include <pthread.h>
78 # include <signal.h>
79 # include <errno.h>
80 # include <dlfcn.h>
81 # include <stdio.h>
82 # include <unistd.h>
83 # include <sys/resource.h>
84 # include <pthread.h>
85 # include <sys/stat.h>
86 # include <sys/time.h>
87 # include <sys/times.h>
88 # include <sys/utsname.h>
89 # include <sys/socket.h>
90 # include <sys/wait.h>
91 # include <pwd.h>
92 # include <poll.h>
93 # include <semaphore.h>
94 # include <fcntl.h>
95 # include <string.h>
96 # include <syscall.h>
97 # include <sys/sysinfo.h>
98 # include <gnu/libc-version.h>
99 # include <sys/ipc.h>
100 # include <sys/shm.h>
101 # include <link.h>
102 # include <stdint.h>
103 # include <inttypes.h>
104 # include <sys/ioctl.h>
106 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
108 #ifndef _GNU_SOURCE
109 #define _GNU_SOURCE
110 #include <sched.h>
111 #undef _GNU_SOURCE
112 #else
113 #include <sched.h>
114 #endif
116 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
117 // getrusage() is prepared to handle the associated failure.
118 #ifndef RUSAGE_THREAD
119 #define RUSAGE_THREAD (1) /* only the calling thread */
120 #endif
122 #define MAX_PATH (2 * K)
124 #define MAX_SECS 100000000
126 // for timer info max values which include all bits
127 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
129 #define LARGEPAGES_BIT (1 << 6)
130 ////////////////////////////////////////////////////////////////////////////////
131 // global variables
132 julong os::Linux::_physical_memory = 0;
134 address os::Linux::_initial_thread_stack_bottom = NULL;
135 uintptr_t os::Linux::_initial_thread_stack_size = 0;
137 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
138 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
139 Mutex* os::Linux::_createThread_lock = NULL;
140 pthread_t os::Linux::_main_thread;
141 int os::Linux::_page_size = -1;
142 const int os::Linux::_vm_default_page_size = (8 * K);
143 bool os::Linux::_is_floating_stack = false;
144 bool os::Linux::_is_NPTL = false;
145 bool os::Linux::_supports_fast_thread_cpu_time = false;
146 const char * os::Linux::_glibc_version = NULL;
147 const char * os::Linux::_libpthread_version = NULL;
148 pthread_condattr_t os::Linux::_condattr[1];
150 static jlong initial_time_count=0;
152 static int clock_tics_per_sec = 100;
154 // For diagnostics to print a message once. see run_periodic_checks
155 static sigset_t check_signal_done;
156 static bool check_signals = true;
158 static pid_t _initial_pid = 0;
160 /* Signal number used to suspend/resume a thread */
162 /* do not use any signal number less than SIGSEGV, see 4355769 */
163 static int SR_signum = SIGUSR2;
164 sigset_t SR_sigset;
166 /* Used to protect dlsym() calls */
167 static pthread_mutex_t dl_mutex;
169 // Declarations
170 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
172 // utility functions
174 static int SR_initialize();
176 julong os::available_memory() {
177 return Linux::available_memory();
178 }
180 julong os::Linux::available_memory() {
181 // values in struct sysinfo are "unsigned long"
182 struct sysinfo si;
183 julong avail_mem;
185 if (OSContainer::is_containerized()) {
186 jlong mem_limit, mem_usage;
187 if ((mem_limit = OSContainer::memory_limit_in_bytes()) < 1) {
188 if (PrintContainerInfo) {
189 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
190 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
191 }
192 }
194 if (mem_limit > 0 && (mem_usage = OSContainer::memory_usage_in_bytes()) < 1) {
195 if (PrintContainerInfo) {
196 tty->print_cr("container memory usage failed: " JLONG_FORMAT ", using host value", mem_usage);
197 }
198 }
200 if (mem_limit > 0 && mem_usage > 0 ) {
201 avail_mem = mem_limit > mem_usage ? (julong)mem_limit - (julong)mem_usage : 0;
202 if (PrintContainerInfo) {
203 tty->print_cr("available container memory: " JULONG_FORMAT, avail_mem);
204 }
205 return avail_mem;
206 }
207 }
209 sysinfo(&si);
210 avail_mem = (julong)si.freeram * si.mem_unit;
211 if (Verbose) {
212 tty->print_cr("available memory: " JULONG_FORMAT, avail_mem);
213 }
214 return avail_mem;
215 }
217 julong os::physical_memory() {
218 jlong phys_mem = 0;
219 if (OSContainer::is_containerized()) {
220 jlong mem_limit;
221 if ((mem_limit = OSContainer::memory_limit_in_bytes()) > 0) {
222 if (PrintContainerInfo) {
223 tty->print_cr("total container memory: " JLONG_FORMAT, mem_limit);
224 }
225 return mem_limit;
226 }
228 if (PrintContainerInfo) {
229 tty->print_cr("container memory limit %s: " JLONG_FORMAT ", using host value",
230 mem_limit == OSCONTAINER_ERROR ? "failed" : "unlimited", mem_limit);
231 }
232 }
234 phys_mem = Linux::physical_memory();
235 if (Verbose) {
236 tty->print_cr("total system memory: " JLONG_FORMAT, phys_mem);
237 }
238 return phys_mem;
239 }
241 ////////////////////////////////////////////////////////////////////////////////
242 // environment support
244 bool os::getenv(const char* name, char* buf, int len) {
245 const char* val = ::getenv(name);
246 if (val != NULL && strlen(val) < (size_t)len) {
247 strcpy(buf, val);
248 return true;
249 }
250 if (len > 0) buf[0] = 0; // return a null string
251 return false;
252 }
255 // Return true if user is running as root.
257 bool os::have_special_privileges() {
258 static bool init = false;
259 static bool privileges = false;
260 if (!init) {
261 privileges = (getuid() != geteuid()) || (getgid() != getegid());
262 init = true;
263 }
264 return privileges;
265 }
268 #ifndef SYS_gettid
269 // i386: 224, ia64: 1105, amd64: 186, sparc 143
270 #ifdef __ia64__
271 #define SYS_gettid 1105
272 #else
273 #ifdef __i386__
274 #define SYS_gettid 224
275 #else
276 #ifdef __amd64__
277 #define SYS_gettid 186
278 #else
279 #ifdef __sparc__
280 #define SYS_gettid 143
281 #else
282 #error define gettid for the arch
283 #endif
284 #endif
285 #endif
286 #endif
287 #endif
289 // Cpu architecture string
290 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
292 // pid_t gettid()
293 //
294 // Returns the kernel thread id of the currently running thread. Kernel
295 // thread id is used to access /proc.
296 //
297 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
298 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
299 //
300 pid_t os::Linux::gettid() {
301 int rslt = syscall(SYS_gettid);
302 if (rslt == -1) {
303 // old kernel, no NPTL support
304 return getpid();
305 } else {
306 return (pid_t)rslt;
307 }
308 }
310 // Most versions of linux have a bug where the number of processors are
311 // determined by looking at the /proc file system. In a chroot environment,
312 // the system call returns 1. This causes the VM to act as if it is
313 // a single processor and elide locking (see is_MP() call).
314 static bool unsafe_chroot_detected = false;
315 static const char *unstable_chroot_error = "/proc file system not found.\n"
316 "Java may be unstable running multithreaded in a chroot "
317 "environment on Linux when /proc filesystem is not mounted.";
319 void os::Linux::initialize_system_info() {
320 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
321 if (processor_count() == 1) {
322 pid_t pid = os::Linux::gettid();
323 char fname[32];
324 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
325 FILE *fp = fopen(fname, "r");
326 if (fp == NULL) {
327 unsafe_chroot_detected = true;
328 } else {
329 fclose(fp);
330 }
331 }
332 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
333 assert(processor_count() > 0, "linux error");
334 }
336 void os::init_system_properties_values() {
337 // The next steps are taken in the product version:
338 //
339 // Obtain the JAVA_HOME value from the location of libjvm.so.
340 // This library should be located at:
341 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
342 //
343 // If "/jre/lib/" appears at the right place in the path, then we
344 // assume libjvm.so is installed in a JDK and we use this path.
345 //
346 // Otherwise exit with message: "Could not create the Java virtual machine."
347 //
348 // The following extra steps are taken in the debugging version:
349 //
350 // If "/jre/lib/" does NOT appear at the right place in the path
351 // instead of exit check for $JAVA_HOME environment variable.
352 //
353 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
354 // then we append a fake suffix "hotspot/libjvm.so" to this path so
355 // it looks like libjvm.so is installed there
356 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
357 //
358 // Otherwise exit.
359 //
360 // Important note: if the location of libjvm.so changes this
361 // code needs to be changed accordingly.
363 // See ld(1):
364 // The linker uses the following search paths to locate required
365 // shared libraries:
366 // 1: ...
367 // ...
368 // 7: The default directories, normally /lib and /usr/lib.
369 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
370 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
371 #else
372 #define DEFAULT_LIBPATH "/lib:/usr/lib"
373 #endif
375 // Base path of extensions installed on the system.
376 #define SYS_EXT_DIR "/usr/java/packages"
377 #define EXTENSIONS_DIR "/lib/ext"
378 #define ENDORSED_DIR "/lib/endorsed"
380 // Buffer that fits several sprintfs.
381 // Note that the space for the colon and the trailing null are provided
382 // by the nulls included by the sizeof operator.
383 const size_t bufsize =
384 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
385 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
386 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
387 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
389 // sysclasspath, java_home, dll_dir
390 {
391 char *pslash;
392 os::jvm_path(buf, bufsize);
394 // Found the full path to libjvm.so.
395 // Now cut the path to <java_home>/jre if we can.
396 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
397 pslash = strrchr(buf, '/');
398 if (pslash != NULL) {
399 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
400 }
401 Arguments::set_dll_dir(buf);
403 if (pslash != NULL) {
404 pslash = strrchr(buf, '/');
405 if (pslash != NULL) {
406 *pslash = '\0'; // Get rid of /<arch>.
407 pslash = strrchr(buf, '/');
408 if (pslash != NULL) {
409 *pslash = '\0'; // Get rid of /lib.
410 }
411 }
412 }
413 Arguments::set_java_home(buf);
414 set_boot_path('/', ':');
415 }
417 // Where to look for native libraries.
418 //
419 // Note: Due to a legacy implementation, most of the library path
420 // is set in the launcher. This was to accomodate linking restrictions
421 // on legacy Linux implementations (which are no longer supported).
422 // Eventually, all the library path setting will be done here.
423 //
424 // However, to prevent the proliferation of improperly built native
425 // libraries, the new path component /usr/java/packages is added here.
426 // Eventually, all the library path setting will be done here.
427 {
428 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
429 // should always exist (until the legacy problem cited above is
430 // addressed).
431 const char *v = ::getenv("LD_LIBRARY_PATH");
432 const char *v_colon = ":";
433 if (v == NULL) { v = ""; v_colon = ""; }
434 // That's +1 for the colon and +1 for the trailing '\0'.
435 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
436 strlen(v) + 1 +
437 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
438 mtInternal);
439 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
440 Arguments::set_library_path(ld_library_path);
441 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
442 }
444 // Extensions directories.
445 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
446 Arguments::set_ext_dirs(buf);
448 // Endorsed standards default directory.
449 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
450 Arguments::set_endorsed_dirs(buf);
452 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
454 #undef DEFAULT_LIBPATH
455 #undef SYS_EXT_DIR
456 #undef EXTENSIONS_DIR
457 #undef ENDORSED_DIR
458 }
460 ////////////////////////////////////////////////////////////////////////////////
461 // breakpoint support
463 void os::breakpoint() {
464 BREAKPOINT;
465 }
467 extern "C" void breakpoint() {
468 // use debugger to set breakpoint here
469 }
471 ////////////////////////////////////////////////////////////////////////////////
472 // signal support
474 debug_only(static bool signal_sets_initialized = false);
475 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
477 bool os::Linux::is_sig_ignored(int sig) {
478 struct sigaction oact;
479 sigaction(sig, (struct sigaction*)NULL, &oact);
480 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
481 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
482 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
483 return true;
484 else
485 return false;
486 }
488 void os::Linux::signal_sets_init() {
489 // Should also have an assertion stating we are still single-threaded.
490 assert(!signal_sets_initialized, "Already initialized");
491 // Fill in signals that are necessarily unblocked for all threads in
492 // the VM. Currently, we unblock the following signals:
493 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
494 // by -Xrs (=ReduceSignalUsage));
495 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
496 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
497 // the dispositions or masks wrt these signals.
498 // Programs embedding the VM that want to use the above signals for their
499 // own purposes must, at this time, use the "-Xrs" option to prevent
500 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
501 // (See bug 4345157, and other related bugs).
502 // In reality, though, unblocking these signals is really a nop, since
503 // these signals are not blocked by default.
504 sigemptyset(&unblocked_sigs);
505 sigemptyset(&allowdebug_blocked_sigs);
506 sigaddset(&unblocked_sigs, SIGILL);
507 sigaddset(&unblocked_sigs, SIGSEGV);
508 sigaddset(&unblocked_sigs, SIGBUS);
509 sigaddset(&unblocked_sigs, SIGFPE);
510 #if defined(PPC64)
511 sigaddset(&unblocked_sigs, SIGTRAP);
512 #endif
513 sigaddset(&unblocked_sigs, SR_signum);
515 if (!ReduceSignalUsage) {
516 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
517 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
518 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
519 }
520 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
521 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
522 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
523 }
524 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
525 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
526 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
527 }
528 }
529 // Fill in signals that are blocked by all but the VM thread.
530 sigemptyset(&vm_sigs);
531 if (!ReduceSignalUsage)
532 sigaddset(&vm_sigs, BREAK_SIGNAL);
533 debug_only(signal_sets_initialized = true);
535 }
537 // These are signals that are unblocked while a thread is running Java.
538 // (For some reason, they get blocked by default.)
539 sigset_t* os::Linux::unblocked_signals() {
540 assert(signal_sets_initialized, "Not initialized");
541 return &unblocked_sigs;
542 }
544 // These are the signals that are blocked while a (non-VM) thread is
545 // running Java. Only the VM thread handles these signals.
546 sigset_t* os::Linux::vm_signals() {
547 assert(signal_sets_initialized, "Not initialized");
548 return &vm_sigs;
549 }
551 // These are signals that are blocked during cond_wait to allow debugger in
552 sigset_t* os::Linux::allowdebug_blocked_signals() {
553 assert(signal_sets_initialized, "Not initialized");
554 return &allowdebug_blocked_sigs;
555 }
557 void os::Linux::hotspot_sigmask(Thread* thread) {
559 //Save caller's signal mask before setting VM signal mask
560 sigset_t caller_sigmask;
561 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
563 OSThread* osthread = thread->osthread();
564 osthread->set_caller_sigmask(caller_sigmask);
566 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
568 if (!ReduceSignalUsage) {
569 if (thread->is_VM_thread()) {
570 // Only the VM thread handles BREAK_SIGNAL ...
571 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
572 } else {
573 // ... all other threads block BREAK_SIGNAL
574 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
575 }
576 }
577 }
579 //////////////////////////////////////////////////////////////////////////////
580 // detecting pthread library
582 void os::Linux::libpthread_init() {
583 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
584 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
585 // generic name for earlier versions.
586 // Define macros here so we can build HotSpot on old systems.
587 # ifndef _CS_GNU_LIBC_VERSION
588 # define _CS_GNU_LIBC_VERSION 2
589 # endif
590 # ifndef _CS_GNU_LIBPTHREAD_VERSION
591 # define _CS_GNU_LIBPTHREAD_VERSION 3
592 # endif
594 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
595 if (n > 0) {
596 char *str = (char *)malloc(n, mtInternal);
597 confstr(_CS_GNU_LIBC_VERSION, str, n);
598 os::Linux::set_glibc_version(str);
599 } else {
600 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
601 static char _gnu_libc_version[32];
602 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
603 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
604 os::Linux::set_glibc_version(_gnu_libc_version);
605 }
607 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
608 if (n > 0) {
609 char *str = (char *)malloc(n, mtInternal);
610 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
611 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
612 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
613 // is the case. LinuxThreads has a hard limit on max number of threads.
614 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
615 // On the other hand, NPTL does not have such a limit, sysconf()
616 // will return -1 and errno is not changed. Check if it is really NPTL.
617 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
618 strstr(str, "NPTL") &&
619 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
620 free(str);
621 os::Linux::set_libpthread_version("linuxthreads");
622 } else {
623 os::Linux::set_libpthread_version(str);
624 }
625 } else {
626 // glibc before 2.3.2 only has LinuxThreads.
627 os::Linux::set_libpthread_version("linuxthreads");
628 }
630 if (strstr(libpthread_version(), "NPTL")) {
631 os::Linux::set_is_NPTL();
632 } else {
633 os::Linux::set_is_LinuxThreads();
634 }
636 // LinuxThreads have two flavors: floating-stack mode, which allows variable
637 // stack size; and fixed-stack mode. NPTL is always floating-stack.
638 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
639 os::Linux::set_is_floating_stack();
640 }
641 }
643 /////////////////////////////////////////////////////////////////////////////
644 // thread stack
646 // Force Linux kernel to expand current thread stack. If "bottom" is close
647 // to the stack guard, caller should block all signals.
648 //
649 // MAP_GROWSDOWN:
650 // A special mmap() flag that is used to implement thread stacks. It tells
651 // kernel that the memory region should extend downwards when needed. This
652 // allows early versions of LinuxThreads to only mmap the first few pages
653 // when creating a new thread. Linux kernel will automatically expand thread
654 // stack as needed (on page faults).
655 //
656 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
657 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
658 // region, it's hard to tell if the fault is due to a legitimate stack
659 // access or because of reading/writing non-exist memory (e.g. buffer
660 // overrun). As a rule, if the fault happens below current stack pointer,
661 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
662 // application (see Linux kernel fault.c).
663 //
664 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
665 // stack overflow detection.
666 //
667 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
668 // not use this flag. However, the stack of initial thread is not created
669 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
670 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
671 // and then attach the thread to JVM.
672 //
673 // To get around the problem and allow stack banging on Linux, we need to
674 // manually expand thread stack after receiving the SIGSEGV.
675 //
676 // There are two ways to expand thread stack to address "bottom", we used
677 // both of them in JVM before 1.5:
678 // 1. adjust stack pointer first so that it is below "bottom", and then
679 // touch "bottom"
680 // 2. mmap() the page in question
681 //
682 // Now alternate signal stack is gone, it's harder to use 2. For instance,
683 // if current sp is already near the lower end of page 101, and we need to
684 // call mmap() to map page 100, it is possible that part of the mmap() frame
685 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
686 // That will destroy the mmap() frame and cause VM to crash.
687 //
688 // The following code works by adjusting sp first, then accessing the "bottom"
689 // page to force a page fault. Linux kernel will then automatically expand the
690 // stack mapping.
691 //
692 // _expand_stack_to() assumes its frame size is less than page size, which
693 // should always be true if the function is not inlined.
695 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
696 #define NOINLINE
697 #else
698 #define NOINLINE __attribute__ ((noinline))
699 #endif
701 static void _expand_stack_to(address bottom) NOINLINE;
703 static void _expand_stack_to(address bottom) {
704 address sp;
705 size_t size;
706 volatile char *p;
708 // Adjust bottom to point to the largest address within the same page, it
709 // gives us a one-page buffer if alloca() allocates slightly more memory.
710 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
711 bottom += os::Linux::page_size() - 1;
713 // sp might be slightly above current stack pointer; if that's the case, we
714 // will alloca() a little more space than necessary, which is OK. Don't use
715 // os::current_stack_pointer(), as its result can be slightly below current
716 // stack pointer, causing us to not alloca enough to reach "bottom".
717 sp = (address)&sp;
719 if (sp > bottom) {
720 size = sp - bottom;
721 p = (volatile char *)alloca(size);
722 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
723 p[0] = '\0';
724 }
725 }
727 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
728 assert(t!=NULL, "just checking");
729 assert(t->osthread()->expanding_stack(), "expand should be set");
730 assert(t->stack_base() != NULL, "stack_base was not initialized");
732 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
733 sigset_t mask_all, old_sigset;
734 sigfillset(&mask_all);
735 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
736 _expand_stack_to(addr);
737 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
738 return true;
739 }
740 return false;
741 }
743 //////////////////////////////////////////////////////////////////////////////
744 // create new thread
746 static address highest_vm_reserved_address();
748 // check if it's safe to start a new thread
749 static bool _thread_safety_check(Thread* thread) {
750 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
751 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
752 // Heap is mmap'ed at lower end of memory space. Thread stacks are
753 // allocated (MAP_FIXED) from high address space. Every thread stack
754 // occupies a fixed size slot (usually 2Mbytes, but user can change
755 // it to other values if they rebuild LinuxThreads).
756 //
757 // Problem with MAP_FIXED is that mmap() can still succeed even part of
758 // the memory region has already been mmap'ed. That means if we have too
759 // many threads and/or very large heap, eventually thread stack will
760 // collide with heap.
761 //
762 // Here we try to prevent heap/stack collision by comparing current
763 // stack bottom with the highest address that has been mmap'ed by JVM
764 // plus a safety margin for memory maps created by native code.
765 //
766 // This feature can be disabled by setting ThreadSafetyMargin to 0
767 //
768 if (ThreadSafetyMargin > 0) {
769 address stack_bottom = os::current_stack_base() - os::current_stack_size();
771 // not safe if our stack extends below the safety margin
772 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
773 } else {
774 return true;
775 }
776 } else {
777 // Floating stack LinuxThreads or NPTL:
778 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
779 // there's not enough space left, pthread_create() will fail. If we come
780 // here, that means enough space has been reserved for stack.
781 return true;
782 }
783 }
785 // Thread start routine for all newly created threads
786 static void *java_start(Thread *thread) {
787 // Try to randomize the cache line index of hot stack frames.
788 // This helps when threads of the same stack traces evict each other's
789 // cache lines. The threads can be either from the same JVM instance, or
790 // from different JVM instances. The benefit is especially true for
791 // processors with hyperthreading technology.
792 static int counter = 0;
793 int pid = os::current_process_id();
794 alloca(((pid ^ counter++) & 7) * 128);
796 ThreadLocalStorage::set_thread(thread);
798 OSThread* osthread = thread->osthread();
799 Monitor* sync = osthread->startThread_lock();
801 // non floating stack LinuxThreads needs extra check, see above
802 if (!_thread_safety_check(thread)) {
803 // notify parent thread
804 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
805 osthread->set_state(ZOMBIE);
806 sync->notify_all();
807 return NULL;
808 }
810 // thread_id is kernel thread id (similar to Solaris LWP id)
811 osthread->set_thread_id(os::Linux::gettid());
813 if (UseNUMA) {
814 int lgrp_id = os::numa_get_group_id();
815 if (lgrp_id != -1) {
816 thread->set_lgrp_id(lgrp_id);
817 }
818 }
819 // initialize signal mask for this thread
820 os::Linux::hotspot_sigmask(thread);
822 // initialize floating point control register
823 os::Linux::init_thread_fpu_state();
825 // handshaking with parent thread
826 {
827 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
829 // notify parent thread
830 osthread->set_state(INITIALIZED);
831 sync->notify_all();
833 // wait until os::start_thread()
834 while (osthread->get_state() == INITIALIZED) {
835 sync->wait(Mutex::_no_safepoint_check_flag);
836 }
837 }
839 // call one more level start routine
840 thread->run();
842 return 0;
843 }
845 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
846 assert(thread->osthread() == NULL, "caller responsible");
848 // Allocate the OSThread object
849 OSThread* osthread = new OSThread(NULL, NULL);
850 if (osthread == NULL) {
851 return false;
852 }
854 // set the correct thread state
855 osthread->set_thread_type(thr_type);
857 // Initial state is ALLOCATED but not INITIALIZED
858 osthread->set_state(ALLOCATED);
860 thread->set_osthread(osthread);
862 // init thread attributes
863 pthread_attr_t attr;
864 pthread_attr_init(&attr);
865 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
867 // stack size
868 if (os::Linux::supports_variable_stack_size()) {
869 // calculate stack size if it's not specified by caller
870 if (stack_size == 0) {
871 stack_size = os::Linux::default_stack_size(thr_type);
873 switch (thr_type) {
874 case os::java_thread:
875 // Java threads use ThreadStackSize which default value can be
876 // changed with the flag -Xss
877 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
878 stack_size = JavaThread::stack_size_at_create();
879 break;
880 case os::compiler_thread:
881 if (CompilerThreadStackSize > 0) {
882 stack_size = (size_t)(CompilerThreadStackSize * K);
883 break;
884 } // else fall through:
885 // use VMThreadStackSize if CompilerThreadStackSize is not defined
886 case os::vm_thread:
887 case os::pgc_thread:
888 case os::cgc_thread:
889 case os::watcher_thread:
890 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
891 break;
892 }
893 }
895 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
896 pthread_attr_setstacksize(&attr, stack_size);
897 } else {
898 // let pthread_create() pick the default value.
899 }
901 // glibc guard page
902 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
904 ThreadState state;
906 {
907 // Serialize thread creation if we are running with fixed stack LinuxThreads
908 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
909 if (lock) {
910 os::Linux::createThread_lock()->lock_without_safepoint_check();
911 }
913 pthread_t tid;
914 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
916 pthread_attr_destroy(&attr);
918 if (ret != 0) {
919 if (PrintMiscellaneous && (Verbose || WizardMode)) {
920 perror("pthread_create()");
921 }
922 // Need to clean up stuff we've allocated so far
923 thread->set_osthread(NULL);
924 delete osthread;
925 if (lock) os::Linux::createThread_lock()->unlock();
926 return false;
927 }
929 // Store pthread info into the OSThread
930 osthread->set_pthread_id(tid);
932 // Wait until child thread is either initialized or aborted
933 {
934 Monitor* sync_with_child = osthread->startThread_lock();
935 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
936 while ((state = osthread->get_state()) == ALLOCATED) {
937 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
938 }
939 }
941 if (lock) {
942 os::Linux::createThread_lock()->unlock();
943 }
944 }
946 // Aborted due to thread limit being reached
947 if (state == ZOMBIE) {
948 thread->set_osthread(NULL);
949 delete osthread;
950 return false;
951 }
953 // The thread is returned suspended (in state INITIALIZED),
954 // and is started higher up in the call chain
955 assert(state == INITIALIZED, "race condition");
956 return true;
957 }
959 /////////////////////////////////////////////////////////////////////////////
960 // attach existing thread
962 // bootstrap the main thread
963 bool os::create_main_thread(JavaThread* thread) {
964 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
965 return create_attached_thread(thread);
966 }
968 bool os::create_attached_thread(JavaThread* thread) {
969 #ifdef ASSERT
970 thread->verify_not_published();
971 #endif
973 // Allocate the OSThread object
974 OSThread* osthread = new OSThread(NULL, NULL);
976 if (osthread == NULL) {
977 return false;
978 }
980 // Store pthread info into the OSThread
981 osthread->set_thread_id(os::Linux::gettid());
982 osthread->set_pthread_id(::pthread_self());
984 // initialize floating point control register
985 os::Linux::init_thread_fpu_state();
987 // Initial thread state is RUNNABLE
988 osthread->set_state(RUNNABLE);
990 thread->set_osthread(osthread);
992 if (UseNUMA) {
993 int lgrp_id = os::numa_get_group_id();
994 if (lgrp_id != -1) {
995 thread->set_lgrp_id(lgrp_id);
996 }
997 }
999 if (os::is_primordial_thread()) {
1000 // If current thread is primordial thread, its stack is mapped on demand,
1001 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1002 // the entire stack region to avoid SEGV in stack banging.
1003 // It is also useful to get around the heap-stack-gap problem on SuSE
1004 // kernel (see 4821821 for details). We first expand stack to the top
1005 // of yellow zone, then enable stack yellow zone (order is significant,
1006 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1007 // is no gap between the last two virtual memory regions.
1009 JavaThread *jt = (JavaThread *)thread;
1010 address addr = jt->stack_yellow_zone_base();
1011 assert(addr != NULL, "initialization problem?");
1012 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1014 osthread->set_expanding_stack();
1015 os::Linux::manually_expand_stack(jt, addr);
1016 osthread->clear_expanding_stack();
1017 }
1019 // initialize signal mask for this thread
1020 // and save the caller's signal mask
1021 os::Linux::hotspot_sigmask(thread);
1023 return true;
1024 }
1026 void os::pd_start_thread(Thread* thread) {
1027 OSThread * osthread = thread->osthread();
1028 assert(osthread->get_state() != INITIALIZED, "just checking");
1029 Monitor* sync_with_child = osthread->startThread_lock();
1030 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1031 sync_with_child->notify();
1032 }
1034 // Free Linux resources related to the OSThread
1035 void os::free_thread(OSThread* osthread) {
1036 assert(osthread != NULL, "osthread not set");
1038 if (Thread::current()->osthread() == osthread) {
1039 // Restore caller's signal mask
1040 sigset_t sigmask = osthread->caller_sigmask();
1041 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1042 }
1044 delete osthread;
1045 }
1047 //////////////////////////////////////////////////////////////////////////////
1048 // thread local storage
1050 // Restore the thread pointer if the destructor is called. This is in case
1051 // someone from JNI code sets up a destructor with pthread_key_create to run
1052 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1053 // will hang or crash. When detachCurrentThread is called the key will be set
1054 // to null and we will not be called again. If detachCurrentThread is never
1055 // called we could loop forever depending on the pthread implementation.
1056 static void restore_thread_pointer(void* p) {
1057 Thread* thread = (Thread*) p;
1058 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1059 }
1061 int os::allocate_thread_local_storage() {
1062 pthread_key_t key;
1063 int rslt = pthread_key_create(&key, restore_thread_pointer);
1064 assert(rslt == 0, "cannot allocate thread local storage");
1065 return (int)key;
1066 }
1068 // Note: This is currently not used by VM, as we don't destroy TLS key
1069 // on VM exit.
1070 void os::free_thread_local_storage(int index) {
1071 int rslt = pthread_key_delete((pthread_key_t)index);
1072 assert(rslt == 0, "invalid index");
1073 }
1075 void os::thread_local_storage_at_put(int index, void* value) {
1076 int rslt = pthread_setspecific((pthread_key_t)index, value);
1077 assert(rslt == 0, "pthread_setspecific failed");
1078 }
1080 extern "C" Thread* get_thread() {
1081 return ThreadLocalStorage::thread();
1082 }
1084 //////////////////////////////////////////////////////////////////////////////
1085 // primordial thread
1087 // Check if current thread is the primordial thread, similar to Solaris thr_main.
1088 bool os::is_primordial_thread(void) {
1089 char dummy;
1090 // If called before init complete, thread stack bottom will be null.
1091 // Can be called if fatal error occurs before initialization.
1092 if (os::Linux::initial_thread_stack_bottom() == NULL) return false;
1093 assert(os::Linux::initial_thread_stack_bottom() != NULL &&
1094 os::Linux::initial_thread_stack_size() != 0,
1095 "os::init did not locate primordial thread's stack region");
1096 if ((address)&dummy >= os::Linux::initial_thread_stack_bottom() &&
1097 (address)&dummy < os::Linux::initial_thread_stack_bottom() +
1098 os::Linux::initial_thread_stack_size()) {
1099 return true;
1100 } else {
1101 return false;
1102 }
1103 }
1105 // Find the virtual memory area that contains addr
1106 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1107 FILE *fp = fopen("/proc/self/maps", "r");
1108 if (fp) {
1109 address low, high;
1110 while (!feof(fp)) {
1111 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1112 if (low <= addr && addr < high) {
1113 if (vma_low) *vma_low = low;
1114 if (vma_high) *vma_high = high;
1115 fclose (fp);
1116 return true;
1117 }
1118 }
1119 for (;;) {
1120 int ch = fgetc(fp);
1121 if (ch == EOF || ch == (int)'\n') break;
1122 }
1123 }
1124 fclose(fp);
1125 }
1126 return false;
1127 }
1129 // Locate primordial thread stack. This special handling of primordial thread stack
1130 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1131 // bogus value for the primordial process thread. While the launcher has created
1132 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1133 // JNI invocation API from a primordial thread.
1134 void os::Linux::capture_initial_stack(size_t max_size) {
1136 // max_size is either 0 (which means accept OS default for thread stacks) or
1137 // a user-specified value known to be at least the minimum needed. If we
1138 // are actually on the primordial thread we can make it appear that we have a
1139 // smaller max_size stack by inserting the guard pages at that location. But we
1140 // cannot do anything to emulate a larger stack than what has been provided by
1141 // the OS or threading library. In fact if we try to use a stack greater than
1142 // what is set by rlimit then we will crash the hosting process.
1144 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1145 // If this is "unlimited" then it will be a huge value.
1146 struct rlimit rlim;
1147 getrlimit(RLIMIT_STACK, &rlim);
1148 size_t stack_size = rlim.rlim_cur;
1150 // 6308388: a bug in ld.so will relocate its own .data section to the
1151 // lower end of primordial stack; reduce ulimit -s value a little bit
1152 // so we won't install guard page on ld.so's data section.
1153 // But ensure we don't underflow the stack size - allow 1 page spare
1154 if (stack_size >= (size_t)(3 * page_size())) {
1155 stack_size -= 2 * page_size();
1156 }
1158 // Try to figure out where the stack base (top) is. This is harder.
1159 //
1160 // When an application is started, glibc saves the initial stack pointer in
1161 // a global variable "__libc_stack_end", which is then used by system
1162 // libraries. __libc_stack_end should be pretty close to stack top. The
1163 // variable is available since the very early days. However, because it is
1164 // a private interface, it could disappear in the future.
1165 //
1166 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1167 // to __libc_stack_end, it is very close to stack top, but isn't the real
1168 // stack top. Note that /proc may not exist if VM is running as a chroot
1169 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1170 // /proc/<pid>/stat could change in the future (though unlikely).
1171 //
1172 // We try __libc_stack_end first. If that doesn't work, look for
1173 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1174 // as a hint, which should work well in most cases.
1176 uintptr_t stack_start;
1178 // try __libc_stack_end first
1179 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1180 if (p && *p) {
1181 stack_start = *p;
1182 } else {
1183 // see if we can get the start_stack field from /proc/self/stat
1184 FILE *fp;
1185 int pid;
1186 char state;
1187 int ppid;
1188 int pgrp;
1189 int session;
1190 int nr;
1191 int tpgrp;
1192 unsigned long flags;
1193 unsigned long minflt;
1194 unsigned long cminflt;
1195 unsigned long majflt;
1196 unsigned long cmajflt;
1197 unsigned long utime;
1198 unsigned long stime;
1199 long cutime;
1200 long cstime;
1201 long prio;
1202 long nice;
1203 long junk;
1204 long it_real;
1205 uintptr_t start;
1206 uintptr_t vsize;
1207 intptr_t rss;
1208 uintptr_t rsslim;
1209 uintptr_t scodes;
1210 uintptr_t ecode;
1211 int i;
1213 // Figure what the primordial thread stack base is. Code is inspired
1214 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1215 // followed by command name surrounded by parentheses, state, etc.
1216 char stat[2048];
1217 int statlen;
1219 fp = fopen("/proc/self/stat", "r");
1220 if (fp) {
1221 statlen = fread(stat, 1, 2047, fp);
1222 stat[statlen] = '\0';
1223 fclose(fp);
1225 // Skip pid and the command string. Note that we could be dealing with
1226 // weird command names, e.g. user could decide to rename java launcher
1227 // to "java 1.4.2 :)", then the stat file would look like
1228 // 1234 (java 1.4.2 :)) R ... ...
1229 // We don't really need to know the command string, just find the last
1230 // occurrence of ")" and then start parsing from there. See bug 4726580.
1231 char * s = strrchr(stat, ')');
1233 i = 0;
1234 if (s) {
1235 // Skip blank chars
1236 do s++; while (isspace(*s));
1238 #define _UFM UINTX_FORMAT
1239 #define _DFM INTX_FORMAT
1241 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1242 /* 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 */
1243 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,
1244 &state, /* 3 %c */
1245 &ppid, /* 4 %d */
1246 &pgrp, /* 5 %d */
1247 &session, /* 6 %d */
1248 &nr, /* 7 %d */
1249 &tpgrp, /* 8 %d */
1250 &flags, /* 9 %lu */
1251 &minflt, /* 10 %lu */
1252 &cminflt, /* 11 %lu */
1253 &majflt, /* 12 %lu */
1254 &cmajflt, /* 13 %lu */
1255 &utime, /* 14 %lu */
1256 &stime, /* 15 %lu */
1257 &cutime, /* 16 %ld */
1258 &cstime, /* 17 %ld */
1259 &prio, /* 18 %ld */
1260 &nice, /* 19 %ld */
1261 &junk, /* 20 %ld */
1262 &it_real, /* 21 %ld */
1263 &start, /* 22 UINTX_FORMAT */
1264 &vsize, /* 23 UINTX_FORMAT */
1265 &rss, /* 24 INTX_FORMAT */
1266 &rsslim, /* 25 UINTX_FORMAT */
1267 &scodes, /* 26 UINTX_FORMAT */
1268 &ecode, /* 27 UINTX_FORMAT */
1269 &stack_start); /* 28 UINTX_FORMAT */
1270 }
1272 #undef _UFM
1273 #undef _DFM
1275 if (i != 28 - 2) {
1276 assert(false, "Bad conversion from /proc/self/stat");
1277 // product mode - assume we are the primordial thread, good luck in the
1278 // embedded case.
1279 warning("Can't detect primordial thread stack location - bad conversion");
1280 stack_start = (uintptr_t) &rlim;
1281 }
1282 } else {
1283 // For some reason we can't open /proc/self/stat (for example, running on
1284 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1285 // most cases, so don't abort:
1286 warning("Can't detect primordial thread stack location - no /proc/self/stat");
1287 stack_start = (uintptr_t) &rlim;
1288 }
1289 }
1291 // Now we have a pointer (stack_start) very close to the stack top, the
1292 // next thing to do is to figure out the exact location of stack top. We
1293 // can find out the virtual memory area that contains stack_start by
1294 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1295 // and its upper limit is the real stack top. (again, this would fail if
1296 // running inside chroot, because /proc may not exist.)
1298 uintptr_t stack_top;
1299 address low, high;
1300 if (find_vma((address)stack_start, &low, &high)) {
1301 // success, "high" is the true stack top. (ignore "low", because initial
1302 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1303 stack_top = (uintptr_t)high;
1304 } else {
1305 // failed, likely because /proc/self/maps does not exist
1306 warning("Can't detect primordial thread stack location - find_vma failed");
1307 // best effort: stack_start is normally within a few pages below the real
1308 // stack top, use it as stack top, and reduce stack size so we won't put
1309 // guard page outside stack.
1310 stack_top = stack_start;
1311 stack_size -= 16 * page_size();
1312 }
1314 // stack_top could be partially down the page so align it
1315 stack_top = align_size_up(stack_top, page_size());
1317 // Allowed stack value is minimum of max_size and what we derived from rlimit
1318 if (max_size > 0) {
1319 _initial_thread_stack_size = MIN2(max_size, stack_size);
1320 } else {
1321 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1322 // clamp it at 8MB as we do on Solaris
1323 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1324 }
1326 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1327 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1328 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1329 }
1331 ////////////////////////////////////////////////////////////////////////////////
1332 // time support
1334 // Time since start-up in seconds to a fine granularity.
1335 // Used by VMSelfDestructTimer and the MemProfiler.
1336 double os::elapsedTime() {
1338 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1339 }
1341 jlong os::elapsed_counter() {
1342 return javaTimeNanos() - initial_time_count;
1343 }
1345 jlong os::elapsed_frequency() {
1346 return NANOSECS_PER_SEC; // nanosecond resolution
1347 }
1349 bool os::supports_vtime() { return true; }
1350 bool os::enable_vtime() { return false; }
1351 bool os::vtime_enabled() { return false; }
1353 double os::elapsedVTime() {
1354 struct rusage usage;
1355 int retval = getrusage(RUSAGE_THREAD, &usage);
1356 if (retval == 0) {
1357 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);
1358 } else {
1359 // better than nothing, but not much
1360 return elapsedTime();
1361 }
1362 }
1364 jlong os::javaTimeMillis() {
1365 timeval time;
1366 int status = gettimeofday(&time, NULL);
1367 assert(status != -1, "linux error");
1368 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1369 }
1371 #ifndef CLOCK_MONOTONIC
1372 #define CLOCK_MONOTONIC (1)
1373 #endif
1375 void os::Linux::clock_init() {
1376 // we do dlopen's in this particular order due to bug in linux
1377 // dynamical loader (see 6348968) leading to crash on exit
1378 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1379 if (handle == NULL) {
1380 handle = dlopen("librt.so", RTLD_LAZY);
1381 }
1383 if (handle) {
1384 int (*clock_getres_func)(clockid_t, struct timespec*) =
1385 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1386 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1387 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1388 if (clock_getres_func && clock_gettime_func) {
1389 // See if monotonic clock is supported by the kernel. Note that some
1390 // early implementations simply return kernel jiffies (updated every
1391 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1392 // for nano time (though the monotonic property is still nice to have).
1393 // It's fixed in newer kernels, however clock_getres() still returns
1394 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1395 // resolution for now. Hopefully as people move to new kernels, this
1396 // won't be a problem.
1397 struct timespec res;
1398 struct timespec tp;
1399 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1400 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1401 // yes, monotonic clock is supported
1402 _clock_gettime = clock_gettime_func;
1403 return;
1404 } else {
1405 // close librt if there is no monotonic clock
1406 dlclose(handle);
1407 }
1408 }
1409 }
1410 warning("No monotonic clock was available - timed services may " \
1411 "be adversely affected if the time-of-day clock changes");
1412 }
1414 #ifndef SYS_clock_getres
1416 #if defined(IA32) || defined(AMD64)
1417 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1418 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1419 #else
1420 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1421 #define sys_clock_getres(x,y) -1
1422 #endif
1424 #else
1425 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1426 #endif
1428 void os::Linux::fast_thread_clock_init() {
1429 if (!UseLinuxPosixThreadCPUClocks) {
1430 return;
1431 }
1432 clockid_t clockid;
1433 struct timespec tp;
1434 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1435 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1437 // Switch to using fast clocks for thread cpu time if
1438 // the sys_clock_getres() returns 0 error code.
1439 // Note, that some kernels may support the current thread
1440 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1441 // returned by the pthread_getcpuclockid().
1442 // If the fast Posix clocks are supported then the sys_clock_getres()
1443 // must return at least tp.tv_sec == 0 which means a resolution
1444 // better than 1 sec. This is extra check for reliability.
1446 if(pthread_getcpuclockid_func &&
1447 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1448 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1450 _supports_fast_thread_cpu_time = true;
1451 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1452 }
1453 }
1455 jlong os::javaTimeNanos() {
1456 if (Linux::supports_monotonic_clock()) {
1457 struct timespec tp;
1458 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1459 assert(status == 0, "gettime error");
1460 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1461 return result;
1462 } else {
1463 timeval time;
1464 int status = gettimeofday(&time, NULL);
1465 assert(status != -1, "linux error");
1466 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1467 return 1000 * usecs;
1468 }
1469 }
1471 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1472 if (Linux::supports_monotonic_clock()) {
1473 info_ptr->max_value = ALL_64_BITS;
1475 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1476 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1477 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1478 } else {
1479 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1480 info_ptr->max_value = ALL_64_BITS;
1482 // gettimeofday is a real time clock so it skips
1483 info_ptr->may_skip_backward = true;
1484 info_ptr->may_skip_forward = true;
1485 }
1487 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1488 }
1490 // Return the real, user, and system times in seconds from an
1491 // arbitrary fixed point in the past.
1492 bool os::getTimesSecs(double* process_real_time,
1493 double* process_user_time,
1494 double* process_system_time) {
1495 struct tms ticks;
1496 clock_t real_ticks = times(&ticks);
1498 if (real_ticks == (clock_t) (-1)) {
1499 return false;
1500 } else {
1501 double ticks_per_second = (double) clock_tics_per_sec;
1502 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1503 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1504 *process_real_time = ((double) real_ticks) / ticks_per_second;
1506 return true;
1507 }
1508 }
1511 char * os::local_time_string(char *buf, size_t buflen) {
1512 struct tm t;
1513 time_t long_time;
1514 time(&long_time);
1515 localtime_r(&long_time, &t);
1516 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1517 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1518 t.tm_hour, t.tm_min, t.tm_sec);
1519 return buf;
1520 }
1522 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1523 return localtime_r(clock, res);
1524 }
1526 ////////////////////////////////////////////////////////////////////////////////
1527 // runtime exit support
1529 // Note: os::shutdown() might be called very early during initialization, or
1530 // called from signal handler. Before adding something to os::shutdown(), make
1531 // sure it is async-safe and can handle partially initialized VM.
1532 void os::shutdown() {
1534 // allow PerfMemory to attempt cleanup of any persistent resources
1535 perfMemory_exit();
1537 // needs to remove object in file system
1538 AttachListener::abort();
1540 // flush buffered output, finish log files
1541 ostream_abort();
1543 // Check for abort hook
1544 abort_hook_t abort_hook = Arguments::abort_hook();
1545 if (abort_hook != NULL) {
1546 abort_hook();
1547 }
1549 }
1551 // Note: os::abort() might be called very early during initialization, or
1552 // called from signal handler. Before adding something to os::abort(), make
1553 // sure it is async-safe and can handle partially initialized VM.
1554 void os::abort(bool dump_core) {
1555 os::shutdown();
1556 if (dump_core) {
1557 #ifndef PRODUCT
1558 fdStream out(defaultStream::output_fd());
1559 out.print_raw("Current thread is ");
1560 char buf[16];
1561 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1562 out.print_raw_cr(buf);
1563 out.print_raw_cr("Dumping core ...");
1564 #endif
1565 ::abort(); // dump core
1566 }
1568 ::exit(1);
1569 }
1571 // Die immediately, no exit hook, no abort hook, no cleanup.
1572 void os::die() {
1573 // _exit() on LinuxThreads only kills current thread
1574 ::abort();
1575 }
1578 // This method is a copy of JDK's sysGetLastErrorString
1579 // from src/solaris/hpi/src/system_md.c
1581 size_t os::lasterror(char *buf, size_t len) {
1583 if (errno == 0) return 0;
1585 const char *s = ::strerror(errno);
1586 size_t n = ::strlen(s);
1587 if (n >= len) {
1588 n = len - 1;
1589 }
1590 ::strncpy(buf, s, n);
1591 buf[n] = '\0';
1592 return n;
1593 }
1595 intx os::current_thread_id() { return (intx)pthread_self(); }
1596 int os::current_process_id() {
1598 // Under the old linux thread library, linux gives each thread
1599 // its own process id. Because of this each thread will return
1600 // a different pid if this method were to return the result
1601 // of getpid(2). Linux provides no api that returns the pid
1602 // of the launcher thread for the vm. This implementation
1603 // returns a unique pid, the pid of the launcher thread
1604 // that starts the vm 'process'.
1606 // Under the NPTL, getpid() returns the same pid as the
1607 // launcher thread rather than a unique pid per thread.
1608 // Use gettid() if you want the old pre NPTL behaviour.
1610 // if you are looking for the result of a call to getpid() that
1611 // returns a unique pid for the calling thread, then look at the
1612 // OSThread::thread_id() method in osThread_linux.hpp file
1614 return (int)(_initial_pid ? _initial_pid : getpid());
1615 }
1617 // DLL functions
1619 const char* os::dll_file_extension() { return ".so"; }
1621 // This must be hard coded because it's the system's temporary
1622 // directory not the java application's temp directory, ala java.io.tmpdir.
1623 const char* os::get_temp_directory() { return "/tmp"; }
1625 static bool file_exists(const char* filename) {
1626 struct stat statbuf;
1627 if (filename == NULL || strlen(filename) == 0) {
1628 return false;
1629 }
1630 return os::stat(filename, &statbuf) == 0;
1631 }
1633 bool os::dll_build_name(char* buffer, size_t buflen,
1634 const char* pname, const char* fname) {
1635 bool retval = false;
1636 // Copied from libhpi
1637 const size_t pnamelen = pname ? strlen(pname) : 0;
1639 // Return error on buffer overflow.
1640 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1641 return retval;
1642 }
1644 if (pnamelen == 0) {
1645 snprintf(buffer, buflen, "lib%s.so", fname);
1646 retval = true;
1647 } else if (strchr(pname, *os::path_separator()) != NULL) {
1648 int n;
1649 char** pelements = split_path(pname, &n);
1650 if (pelements == NULL) {
1651 return false;
1652 }
1653 for (int i = 0 ; i < n ; i++) {
1654 // Really shouldn't be NULL, but check can't hurt
1655 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1656 continue; // skip the empty path values
1657 }
1658 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1659 if (file_exists(buffer)) {
1660 retval = true;
1661 break;
1662 }
1663 }
1664 // release the storage
1665 for (int i = 0 ; i < n ; i++) {
1666 if (pelements[i] != NULL) {
1667 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1668 }
1669 }
1670 if (pelements != NULL) {
1671 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1672 }
1673 } else {
1674 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1675 retval = true;
1676 }
1677 return retval;
1678 }
1680 // check if addr is inside libjvm.so
1681 bool os::address_is_in_vm(address addr) {
1682 static address libjvm_base_addr;
1683 Dl_info dlinfo;
1685 if (libjvm_base_addr == NULL) {
1686 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1687 libjvm_base_addr = (address)dlinfo.dli_fbase;
1688 }
1689 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1690 }
1692 if (dladdr((void *)addr, &dlinfo) != 0) {
1693 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1694 }
1696 return false;
1697 }
1699 bool os::dll_address_to_function_name(address addr, char *buf,
1700 int buflen, int *offset) {
1701 // buf is not optional, but offset is optional
1702 assert(buf != NULL, "sanity check");
1704 Dl_info dlinfo;
1706 if (dladdr((void*)addr, &dlinfo) != 0) {
1707 // see if we have a matching symbol
1708 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1709 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1710 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1711 }
1712 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1713 return true;
1714 }
1715 // no matching symbol so try for just file info
1716 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1717 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1718 buf, buflen, offset, dlinfo.dli_fname)) {
1719 return true;
1720 }
1721 }
1722 }
1724 buf[0] = '\0';
1725 if (offset != NULL) *offset = -1;
1726 return false;
1727 }
1729 struct _address_to_library_name {
1730 address addr; // input : memory address
1731 size_t buflen; // size of fname
1732 char* fname; // output: library name
1733 address base; // library base addr
1734 };
1736 static int address_to_library_name_callback(struct dl_phdr_info *info,
1737 size_t size, void *data) {
1738 int i;
1739 bool found = false;
1740 address libbase = NULL;
1741 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1743 // iterate through all loadable segments
1744 for (i = 0; i < info->dlpi_phnum; i++) {
1745 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1746 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1747 // base address of a library is the lowest address of its loaded
1748 // segments.
1749 if (libbase == NULL || libbase > segbase) {
1750 libbase = segbase;
1751 }
1752 // see if 'addr' is within current segment
1753 if (segbase <= d->addr &&
1754 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1755 found = true;
1756 }
1757 }
1758 }
1760 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1761 // so dll_address_to_library_name() can fall through to use dladdr() which
1762 // can figure out executable name from argv[0].
1763 if (found && info->dlpi_name && info->dlpi_name[0]) {
1764 d->base = libbase;
1765 if (d->fname) {
1766 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1767 }
1768 return 1;
1769 }
1770 return 0;
1771 }
1773 bool os::dll_address_to_library_name(address addr, char* buf,
1774 int buflen, int* offset) {
1775 // buf is not optional, but offset is optional
1776 assert(buf != NULL, "sanity check");
1778 Dl_info dlinfo;
1779 struct _address_to_library_name data;
1781 // There is a bug in old glibc dladdr() implementation that it could resolve
1782 // to wrong library name if the .so file has a base address != NULL. Here
1783 // we iterate through the program headers of all loaded libraries to find
1784 // out which library 'addr' really belongs to. This workaround can be
1785 // removed once the minimum requirement for glibc is moved to 2.3.x.
1786 data.addr = addr;
1787 data.fname = buf;
1788 data.buflen = buflen;
1789 data.base = NULL;
1790 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1792 if (rslt) {
1793 // buf already contains library name
1794 if (offset) *offset = addr - data.base;
1795 return true;
1796 }
1797 if (dladdr((void*)addr, &dlinfo) != 0) {
1798 if (dlinfo.dli_fname != NULL) {
1799 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1800 }
1801 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1802 *offset = addr - (address)dlinfo.dli_fbase;
1803 }
1804 return true;
1805 }
1807 buf[0] = '\0';
1808 if (offset) *offset = -1;
1809 return false;
1810 }
1812 // Loads .dll/.so and
1813 // in case of error it checks if .dll/.so was built for the
1814 // same architecture as Hotspot is running on
1817 // Remember the stack's state. The Linux dynamic linker will change
1818 // the stack to 'executable' at most once, so we must safepoint only once.
1819 bool os::Linux::_stack_is_executable = false;
1821 // VM operation that loads a library. This is necessary if stack protection
1822 // of the Java stacks can be lost during loading the library. If we
1823 // do not stop the Java threads, they can stack overflow before the stacks
1824 // are protected again.
1825 class VM_LinuxDllLoad: public VM_Operation {
1826 private:
1827 const char *_filename;
1828 char *_ebuf;
1829 int _ebuflen;
1830 void *_lib;
1831 public:
1832 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1833 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1834 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1835 void doit() {
1836 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1837 os::Linux::_stack_is_executable = true;
1838 }
1839 void* loaded_library() { return _lib; }
1840 };
1842 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1843 {
1844 void * result = NULL;
1845 bool load_attempted = false;
1847 // Check whether the library to load might change execution rights
1848 // of the stack. If they are changed, the protection of the stack
1849 // guard pages will be lost. We need a safepoint to fix this.
1850 //
1851 // See Linux man page execstack(8) for more info.
1852 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1853 ElfFile ef(filename);
1854 if (!ef.specifies_noexecstack()) {
1855 if (!is_init_completed()) {
1856 os::Linux::_stack_is_executable = true;
1857 // This is OK - No Java threads have been created yet, and hence no
1858 // stack guard pages to fix.
1859 //
1860 // This should happen only when you are building JDK7 using a very
1861 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1862 //
1863 // Dynamic loader will make all stacks executable after
1864 // this function returns, and will not do that again.
1865 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1866 } else {
1867 warning("You have loaded library %s which might have disabled stack guard. "
1868 "The VM will try to fix the stack guard now.\n"
1869 "It's highly recommended that you fix the library with "
1870 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1871 filename);
1873 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1874 JavaThread *jt = JavaThread::current();
1875 if (jt->thread_state() != _thread_in_native) {
1876 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1877 // that requires ExecStack. Cannot enter safe point. Let's give up.
1878 warning("Unable to fix stack guard. Giving up.");
1879 } else {
1880 if (!LoadExecStackDllInVMThread) {
1881 // This is for the case where the DLL has an static
1882 // constructor function that executes JNI code. We cannot
1883 // load such DLLs in the VMThread.
1884 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1885 }
1887 ThreadInVMfromNative tiv(jt);
1888 debug_only(VMNativeEntryWrapper vew;)
1890 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1891 VMThread::execute(&op);
1892 if (LoadExecStackDllInVMThread) {
1893 result = op.loaded_library();
1894 }
1895 load_attempted = true;
1896 }
1897 }
1898 }
1899 }
1901 if (!load_attempted) {
1902 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1903 }
1905 if (result != NULL) {
1906 // Successful loading
1907 return result;
1908 }
1910 Elf32_Ehdr elf_head;
1911 int diag_msg_max_length=ebuflen-strlen(ebuf);
1912 char* diag_msg_buf=ebuf+strlen(ebuf);
1914 if (diag_msg_max_length==0) {
1915 // No more space in ebuf for additional diagnostics message
1916 return NULL;
1917 }
1920 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1922 if (file_descriptor < 0) {
1923 // Can't open library, report dlerror() message
1924 return NULL;
1925 }
1927 bool failed_to_read_elf_head=
1928 (sizeof(elf_head)!=
1929 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1931 ::close(file_descriptor);
1932 if (failed_to_read_elf_head) {
1933 // file i/o error - report dlerror() msg
1934 return NULL;
1935 }
1937 typedef struct {
1938 Elf32_Half code; // Actual value as defined in elf.h
1939 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1940 char elf_class; // 32 or 64 bit
1941 char endianess; // MSB or LSB
1942 char* name; // String representation
1943 } arch_t;
1945 #ifndef EM_486
1946 #define EM_486 6 /* Intel 80486 */
1947 #endif
1949 static const arch_t arch_array[]={
1950 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1951 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1952 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1953 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1954 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1955 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1956 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1957 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1958 #if defined(VM_LITTLE_ENDIAN)
1959 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64 LE"},
1960 #else
1961 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1962 #endif
1963 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1964 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1965 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1966 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1967 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1968 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1969 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1970 };
1972 #if (defined IA32)
1973 static Elf32_Half running_arch_code=EM_386;
1974 #elif (defined AMD64)
1975 static Elf32_Half running_arch_code=EM_X86_64;
1976 #elif (defined IA64)
1977 static Elf32_Half running_arch_code=EM_IA_64;
1978 #elif (defined __sparc) && (defined _LP64)
1979 static Elf32_Half running_arch_code=EM_SPARCV9;
1980 #elif (defined __sparc) && (!defined _LP64)
1981 static Elf32_Half running_arch_code=EM_SPARC;
1982 #elif (defined __powerpc64__)
1983 static Elf32_Half running_arch_code=EM_PPC64;
1984 #elif (defined __powerpc__)
1985 static Elf32_Half running_arch_code=EM_PPC;
1986 #elif (defined ARM)
1987 static Elf32_Half running_arch_code=EM_ARM;
1988 #elif (defined S390)
1989 static Elf32_Half running_arch_code=EM_S390;
1990 #elif (defined ALPHA)
1991 static Elf32_Half running_arch_code=EM_ALPHA;
1992 #elif (defined MIPSEL)
1993 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1994 #elif (defined PARISC)
1995 static Elf32_Half running_arch_code=EM_PARISC;
1996 #elif (defined MIPS)
1997 static Elf32_Half running_arch_code=EM_MIPS;
1998 #elif (defined M68K)
1999 static Elf32_Half running_arch_code=EM_68K;
2000 #else
2001 #error Method os::dll_load requires that one of following is defined:\
2002 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2003 #endif
2005 // Identify compatability class for VM's architecture and library's architecture
2006 // Obtain string descriptions for architectures
2008 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2009 int running_arch_index=-1;
2011 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2012 if (running_arch_code == arch_array[i].code) {
2013 running_arch_index = i;
2014 }
2015 if (lib_arch.code == arch_array[i].code) {
2016 lib_arch.compat_class = arch_array[i].compat_class;
2017 lib_arch.name = arch_array[i].name;
2018 }
2019 }
2021 assert(running_arch_index != -1,
2022 "Didn't find running architecture code (running_arch_code) in arch_array");
2023 if (running_arch_index == -1) {
2024 // Even though running architecture detection failed
2025 // we may still continue with reporting dlerror() message
2026 return NULL;
2027 }
2029 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2030 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2031 return NULL;
2032 }
2034 #ifndef S390
2035 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2036 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2037 return NULL;
2038 }
2039 #endif // !S390
2041 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2042 if ( lib_arch.name!=NULL ) {
2043 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2044 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2045 lib_arch.name, arch_array[running_arch_index].name);
2046 } else {
2047 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2048 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2049 lib_arch.code,
2050 arch_array[running_arch_index].name);
2051 }
2052 }
2054 return NULL;
2055 }
2057 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2058 void * result = ::dlopen(filename, RTLD_LAZY);
2059 if (result == NULL) {
2060 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2061 ebuf[ebuflen-1] = '\0';
2062 }
2063 return result;
2064 }
2066 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2067 void * result = NULL;
2068 if (LoadExecStackDllInVMThread) {
2069 result = dlopen_helper(filename, ebuf, ebuflen);
2070 }
2072 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2073 // library that requires an executable stack, or which does not have this
2074 // stack attribute set, dlopen changes the stack attribute to executable. The
2075 // read protection of the guard pages gets lost.
2076 //
2077 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2078 // may have been queued at the same time.
2080 if (!_stack_is_executable) {
2081 JavaThread *jt = Threads::first();
2083 while (jt) {
2084 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2085 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2086 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2087 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2088 warning("Attempt to reguard stack yellow zone failed.");
2089 }
2090 }
2091 jt = jt->next();
2092 }
2093 }
2095 return result;
2096 }
2098 /*
2099 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2100 * chances are you might want to run the generated bits against glibc-2.0
2101 * libdl.so, so always use locking for any version of glibc.
2102 */
2103 void* os::dll_lookup(void* handle, const char* name) {
2104 pthread_mutex_lock(&dl_mutex);
2105 void* res = dlsym(handle, name);
2106 pthread_mutex_unlock(&dl_mutex);
2107 return res;
2108 }
2110 void* os::get_default_process_handle() {
2111 return (void*)::dlopen(NULL, RTLD_LAZY);
2112 }
2114 static bool _print_ascii_file(const char* filename, outputStream* st) {
2115 int fd = ::open(filename, O_RDONLY);
2116 if (fd == -1) {
2117 return false;
2118 }
2120 char buf[32];
2121 int bytes;
2122 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2123 st->print_raw(buf, bytes);
2124 }
2126 ::close(fd);
2128 return true;
2129 }
2131 void os::print_dll_info(outputStream *st) {
2132 st->print_cr("Dynamic libraries:");
2134 char fname[32];
2135 pid_t pid = os::Linux::gettid();
2137 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2139 if (!_print_ascii_file(fname, st)) {
2140 st->print("Can not get library information for pid = %d\n", pid);
2141 }
2142 }
2144 void os::print_os_info_brief(outputStream* st) {
2145 os::Linux::print_distro_info(st);
2147 os::Posix::print_uname_info(st);
2149 os::Linux::print_libversion_info(st);
2151 }
2153 void os::print_os_info(outputStream* st) {
2154 st->print("OS:");
2156 os::Linux::print_distro_info(st);
2158 os::Posix::print_uname_info(st);
2160 // Print warning if unsafe chroot environment detected
2161 if (unsafe_chroot_detected) {
2162 st->print("WARNING!! ");
2163 st->print_cr("%s", unstable_chroot_error);
2164 }
2166 os::Linux::print_libversion_info(st);
2168 os::Posix::print_rlimit_info(st);
2170 os::Posix::print_load_average(st);
2172 os::Linux::print_full_memory_info(st);
2174 os::Linux::print_container_info(st);
2175 }
2177 // Try to identify popular distros.
2178 // Most Linux distributions have a /etc/XXX-release file, which contains
2179 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2180 // file that also contains the OS version string. Some have more than one
2181 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2182 // /etc/redhat-release.), so the order is important.
2183 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2184 // their own specific XXX-release file as well as a redhat-release file.
2185 // Because of this the XXX-release file needs to be searched for before the
2186 // redhat-release file.
2187 // Since Red Hat has a lsb-release file that is not very descriptive the
2188 // search for redhat-release needs to be before lsb-release.
2189 // Since the lsb-release file is the new standard it needs to be searched
2190 // before the older style release files.
2191 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2192 // next to last resort. The os-release file is a new standard that contains
2193 // distribution information and the system-release file seems to be an old
2194 // standard that has been replaced by the lsb-release and os-release files.
2195 // Searching for the debian_version file is the last resort. It contains
2196 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2197 // "Debian " is printed before the contents of the debian_version file.
2198 void os::Linux::print_distro_info(outputStream* st) {
2199 if (!_print_ascii_file("/etc/oracle-release", st) &&
2200 !_print_ascii_file("/etc/mandriva-release", st) &&
2201 !_print_ascii_file("/etc/mandrake-release", st) &&
2202 !_print_ascii_file("/etc/sun-release", st) &&
2203 !_print_ascii_file("/etc/redhat-release", st) &&
2204 !_print_ascii_file("/etc/lsb-release", st) &&
2205 !_print_ascii_file("/etc/SuSE-release", st) &&
2206 !_print_ascii_file("/etc/turbolinux-release", st) &&
2207 !_print_ascii_file("/etc/gentoo-release", st) &&
2208 !_print_ascii_file("/etc/ltib-release", st) &&
2209 !_print_ascii_file("/etc/angstrom-version", st) &&
2210 !_print_ascii_file("/etc/system-release", st) &&
2211 !_print_ascii_file("/etc/os-release", st)) {
2213 if (file_exists("/etc/debian_version")) {
2214 st->print("Debian ");
2215 _print_ascii_file("/etc/debian_version", st);
2216 } else {
2217 st->print("Linux");
2218 }
2219 }
2220 st->cr();
2221 }
2223 void os::Linux::print_libversion_info(outputStream* st) {
2224 // libc, pthread
2225 st->print("libc:");
2226 st->print("%s ", os::Linux::glibc_version());
2227 st->print("%s ", os::Linux::libpthread_version());
2228 if (os::Linux::is_LinuxThreads()) {
2229 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2230 }
2231 st->cr();
2232 }
2234 void os::Linux::print_full_memory_info(outputStream* st) {
2235 st->print("\n/proc/meminfo:\n");
2236 _print_ascii_file("/proc/meminfo", st);
2237 st->cr();
2238 }
2240 void os::Linux::print_container_info(outputStream* st) {
2241 if (!OSContainer::is_containerized()) {
2242 return;
2243 }
2245 st->print("container (cgroup) information:\n");
2247 const char *p_ct = OSContainer::container_type();
2248 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2250 char *p = OSContainer::cpu_cpuset_cpus();
2251 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2252 free(p);
2254 p = OSContainer::cpu_cpuset_memory_nodes();
2255 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2256 free(p);
2258 int i = OSContainer::active_processor_count();
2259 if (i > 0) {
2260 st->print("active_processor_count: %d\n", i);
2261 } else {
2262 st->print("active_processor_count: failed\n");
2263 }
2265 i = OSContainer::cpu_quota();
2266 st->print("cpu_quota: %d\n", i);
2268 i = OSContainer::cpu_period();
2269 st->print("cpu_period: %d\n", i);
2271 i = OSContainer::cpu_shares();
2272 st->print("cpu_shares: %d\n", i);
2274 jlong j = OSContainer::memory_limit_in_bytes();
2275 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2277 j = OSContainer::memory_and_swap_limit_in_bytes();
2278 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2280 j = OSContainer::memory_soft_limit_in_bytes();
2281 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2283 j = OSContainer::OSContainer::memory_usage_in_bytes();
2284 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2286 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2287 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2288 st->cr();
2289 }
2291 void os::print_memory_info(outputStream* st) {
2293 st->print("Memory:");
2294 st->print(" %dk page", os::vm_page_size()>>10);
2296 // values in struct sysinfo are "unsigned long"
2297 struct sysinfo si;
2298 sysinfo(&si);
2300 st->print(", physical " UINT64_FORMAT "k",
2301 os::physical_memory() >> 10);
2302 st->print("(" UINT64_FORMAT "k free)",
2303 os::available_memory() >> 10);
2304 st->print(", swap " UINT64_FORMAT "k",
2305 ((jlong)si.totalswap * si.mem_unit) >> 10);
2306 st->print("(" UINT64_FORMAT "k free)",
2307 ((jlong)si.freeswap * si.mem_unit) >> 10);
2308 st->cr();
2309 }
2311 void os::pd_print_cpu_info(outputStream* st) {
2312 st->print("\n/proc/cpuinfo:\n");
2313 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2314 st->print(" <Not Available>");
2315 }
2316 st->cr();
2317 }
2319 void os::print_siginfo(outputStream* st, void* siginfo) {
2320 const siginfo_t* si = (const siginfo_t*)siginfo;
2322 os::Posix::print_siginfo_brief(st, si);
2323 #if INCLUDE_CDS
2324 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2325 UseSharedSpaces) {
2326 FileMapInfo* mapinfo = FileMapInfo::current_info();
2327 if (mapinfo->is_in_shared_space(si->si_addr)) {
2328 st->print("\n\nError accessing class data sharing archive." \
2329 " Mapped file inaccessible during execution, " \
2330 " possible disk/network problem.");
2331 }
2332 }
2333 #endif
2334 st->cr();
2335 }
2338 static void print_signal_handler(outputStream* st, int sig,
2339 char* buf, size_t buflen);
2341 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2342 st->print_cr("Signal Handlers:");
2343 print_signal_handler(st, SIGSEGV, buf, buflen);
2344 print_signal_handler(st, SIGBUS , buf, buflen);
2345 print_signal_handler(st, SIGFPE , buf, buflen);
2346 print_signal_handler(st, SIGPIPE, buf, buflen);
2347 print_signal_handler(st, SIGXFSZ, buf, buflen);
2348 print_signal_handler(st, SIGILL , buf, buflen);
2349 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2350 print_signal_handler(st, SR_signum, buf, buflen);
2351 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2352 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2353 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2354 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2355 #if defined(PPC64)
2356 print_signal_handler(st, SIGTRAP, buf, buflen);
2357 #endif
2358 }
2360 static char saved_jvm_path[MAXPATHLEN] = {0};
2362 // Find the full path to the current module, libjvm.so
2363 void os::jvm_path(char *buf, jint buflen) {
2364 // Error checking.
2365 if (buflen < MAXPATHLEN) {
2366 assert(false, "must use a large-enough buffer");
2367 buf[0] = '\0';
2368 return;
2369 }
2370 // Lazy resolve the path to current module.
2371 if (saved_jvm_path[0] != 0) {
2372 strcpy(buf, saved_jvm_path);
2373 return;
2374 }
2376 char dli_fname[MAXPATHLEN];
2377 bool ret = dll_address_to_library_name(
2378 CAST_FROM_FN_PTR(address, os::jvm_path),
2379 dli_fname, sizeof(dli_fname), NULL);
2380 assert(ret, "cannot locate libjvm");
2381 char *rp = NULL;
2382 if (ret && dli_fname[0] != '\0') {
2383 rp = realpath(dli_fname, buf);
2384 }
2385 if (rp == NULL)
2386 return;
2388 if (Arguments::created_by_gamma_launcher()) {
2389 // Support for the gamma launcher. Typical value for buf is
2390 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2391 // the right place in the string, then assume we are installed in a JDK and
2392 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2393 // up the path so it looks like libjvm.so is installed there (append a
2394 // fake suffix hotspot/libjvm.so).
2395 const char *p = buf + strlen(buf) - 1;
2396 for (int count = 0; p > buf && count < 5; ++count) {
2397 for (--p; p > buf && *p != '/'; --p)
2398 /* empty */ ;
2399 }
2401 if (strncmp(p, "/jre/lib/", 9) != 0) {
2402 // Look for JAVA_HOME in the environment.
2403 char* java_home_var = ::getenv("JAVA_HOME");
2404 if (java_home_var != NULL && java_home_var[0] != 0) {
2405 char* jrelib_p;
2406 int len;
2408 // Check the current module name "libjvm.so".
2409 p = strrchr(buf, '/');
2410 assert(strstr(p, "/libjvm") == p, "invalid library name");
2412 rp = realpath(java_home_var, buf);
2413 if (rp == NULL)
2414 return;
2416 // determine if this is a legacy image or modules image
2417 // modules image doesn't have "jre" subdirectory
2418 len = strlen(buf);
2419 assert(len < buflen, "Ran out of buffer room");
2420 jrelib_p = buf + len;
2421 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2422 if (0 != access(buf, F_OK)) {
2423 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2424 }
2426 if (0 == access(buf, F_OK)) {
2427 // Use current module name "libjvm.so"
2428 len = strlen(buf);
2429 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2430 } else {
2431 // Go back to path of .so
2432 rp = realpath(dli_fname, buf);
2433 if (rp == NULL)
2434 return;
2435 }
2436 }
2437 }
2438 }
2440 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2441 }
2443 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2444 // no prefix required, not even "_"
2445 }
2447 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2448 // no suffix required
2449 }
2451 ////////////////////////////////////////////////////////////////////////////////
2452 // sun.misc.Signal support
2454 static volatile jint sigint_count = 0;
2456 static void
2457 UserHandler(int sig, void *siginfo, void *context) {
2458 // 4511530 - sem_post is serialized and handled by the manager thread. When
2459 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2460 // don't want to flood the manager thread with sem_post requests.
2461 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2462 return;
2464 // Ctrl-C is pressed during error reporting, likely because the error
2465 // handler fails to abort. Let VM die immediately.
2466 if (sig == SIGINT && is_error_reported()) {
2467 os::die();
2468 }
2470 os::signal_notify(sig);
2471 }
2473 void* os::user_handler() {
2474 return CAST_FROM_FN_PTR(void*, UserHandler);
2475 }
2477 class Semaphore : public StackObj {
2478 public:
2479 Semaphore();
2480 ~Semaphore();
2481 void signal();
2482 void wait();
2483 bool trywait();
2484 bool timedwait(unsigned int sec, int nsec);
2485 private:
2486 sem_t _semaphore;
2487 };
2489 Semaphore::Semaphore() {
2490 sem_init(&_semaphore, 0, 0);
2491 }
2493 Semaphore::~Semaphore() {
2494 sem_destroy(&_semaphore);
2495 }
2497 void Semaphore::signal() {
2498 sem_post(&_semaphore);
2499 }
2501 void Semaphore::wait() {
2502 sem_wait(&_semaphore);
2503 }
2505 bool Semaphore::trywait() {
2506 return sem_trywait(&_semaphore) == 0;
2507 }
2509 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2511 struct timespec ts;
2512 // Semaphore's are always associated with CLOCK_REALTIME
2513 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2514 // see unpackTime for discussion on overflow checking
2515 if (sec >= MAX_SECS) {
2516 ts.tv_sec += MAX_SECS;
2517 ts.tv_nsec = 0;
2518 } else {
2519 ts.tv_sec += sec;
2520 ts.tv_nsec += nsec;
2521 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2522 ts.tv_nsec -= NANOSECS_PER_SEC;
2523 ++ts.tv_sec; // note: this must be <= max_secs
2524 }
2525 }
2527 while (1) {
2528 int result = sem_timedwait(&_semaphore, &ts);
2529 if (result == 0) {
2530 return true;
2531 } else if (errno == EINTR) {
2532 continue;
2533 } else if (errno == ETIMEDOUT) {
2534 return false;
2535 } else {
2536 return false;
2537 }
2538 }
2539 }
2541 extern "C" {
2542 typedef void (*sa_handler_t)(int);
2543 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2544 }
2546 void* os::signal(int signal_number, void* handler) {
2547 struct sigaction sigAct, oldSigAct;
2549 sigfillset(&(sigAct.sa_mask));
2550 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2551 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2553 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2554 // -1 means registration failed
2555 return (void *)-1;
2556 }
2558 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2559 }
2561 void os::signal_raise(int signal_number) {
2562 ::raise(signal_number);
2563 }
2565 /*
2566 * The following code is moved from os.cpp for making this
2567 * code platform specific, which it is by its very nature.
2568 */
2570 // Will be modified when max signal is changed to be dynamic
2571 int os::sigexitnum_pd() {
2572 return NSIG;
2573 }
2575 // a counter for each possible signal value
2576 static volatile jint pending_signals[NSIG+1] = { 0 };
2578 // Linux(POSIX) specific hand shaking semaphore.
2579 static sem_t sig_sem;
2580 static Semaphore sr_semaphore;
2582 void os::signal_init_pd() {
2583 // Initialize signal structures
2584 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2586 // Initialize signal semaphore
2587 ::sem_init(&sig_sem, 0, 0);
2588 }
2590 void os::signal_notify(int sig) {
2591 Atomic::inc(&pending_signals[sig]);
2592 ::sem_post(&sig_sem);
2593 }
2595 static int check_pending_signals(bool wait) {
2596 Atomic::store(0, &sigint_count);
2597 for (;;) {
2598 for (int i = 0; i < NSIG + 1; i++) {
2599 jint n = pending_signals[i];
2600 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2601 return i;
2602 }
2603 }
2604 if (!wait) {
2605 return -1;
2606 }
2607 JavaThread *thread = JavaThread::current();
2608 ThreadBlockInVM tbivm(thread);
2610 bool threadIsSuspended;
2611 do {
2612 thread->set_suspend_equivalent();
2613 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2614 ::sem_wait(&sig_sem);
2616 // were we externally suspended while we were waiting?
2617 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2618 if (threadIsSuspended) {
2619 //
2620 // The semaphore has been incremented, but while we were waiting
2621 // another thread suspended us. We don't want to continue running
2622 // while suspended because that would surprise the thread that
2623 // suspended us.
2624 //
2625 ::sem_post(&sig_sem);
2627 thread->java_suspend_self();
2628 }
2629 } while (threadIsSuspended);
2630 }
2631 }
2633 int os::signal_lookup() {
2634 return check_pending_signals(false);
2635 }
2637 int os::signal_wait() {
2638 return check_pending_signals(true);
2639 }
2641 ////////////////////////////////////////////////////////////////////////////////
2642 // Virtual Memory
2644 int os::vm_page_size() {
2645 // Seems redundant as all get out
2646 assert(os::Linux::page_size() != -1, "must call os::init");
2647 return os::Linux::page_size();
2648 }
2650 // Solaris allocates memory by pages.
2651 int os::vm_allocation_granularity() {
2652 assert(os::Linux::page_size() != -1, "must call os::init");
2653 return os::Linux::page_size();
2654 }
2656 // Rationale behind this function:
2657 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2658 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2659 // samples for JITted code. Here we create private executable mapping over the code cache
2660 // and then we can use standard (well, almost, as mapping can change) way to provide
2661 // info for the reporting script by storing timestamp and location of symbol
2662 void linux_wrap_code(char* base, size_t size) {
2663 static volatile jint cnt = 0;
2665 if (!UseOprofile) {
2666 return;
2667 }
2669 char buf[PATH_MAX+1];
2670 int num = Atomic::add(1, &cnt);
2672 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2673 os::get_temp_directory(), os::current_process_id(), num);
2674 unlink(buf);
2676 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2678 if (fd != -1) {
2679 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2680 if (rv != (off_t)-1) {
2681 if (::write(fd, "", 1) == 1) {
2682 mmap(base, size,
2683 PROT_READ|PROT_WRITE|PROT_EXEC,
2684 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2685 }
2686 }
2687 ::close(fd);
2688 unlink(buf);
2689 }
2690 }
2692 static bool recoverable_mmap_error(int err) {
2693 // See if the error is one we can let the caller handle. This
2694 // list of errno values comes from JBS-6843484. I can't find a
2695 // Linux man page that documents this specific set of errno
2696 // values so while this list currently matches Solaris, it may
2697 // change as we gain experience with this failure mode.
2698 switch (err) {
2699 case EBADF:
2700 case EINVAL:
2701 case ENOTSUP:
2702 // let the caller deal with these errors
2703 return true;
2705 default:
2706 // Any remaining errors on this OS can cause our reserved mapping
2707 // to be lost. That can cause confusion where different data
2708 // structures think they have the same memory mapped. The worst
2709 // scenario is if both the VM and a library think they have the
2710 // same memory mapped.
2711 return false;
2712 }
2713 }
2715 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2716 int err) {
2717 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2718 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2719 strerror(err), err);
2720 }
2722 static void warn_fail_commit_memory(char* addr, size_t size,
2723 size_t alignment_hint, bool exec,
2724 int err) {
2725 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2726 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2727 alignment_hint, exec, strerror(err), err);
2728 }
2730 // NOTE: Linux kernel does not really reserve the pages for us.
2731 // All it does is to check if there are enough free pages
2732 // left at the time of mmap(). This could be a potential
2733 // problem.
2734 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2735 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2736 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2737 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2738 if (res != (uintptr_t) MAP_FAILED) {
2739 if (UseNUMAInterleaving) {
2740 numa_make_global(addr, size);
2741 }
2742 return 0;
2743 }
2745 int err = errno; // save errno from mmap() call above
2747 if (!recoverable_mmap_error(err)) {
2748 warn_fail_commit_memory(addr, size, exec, err);
2749 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2750 }
2752 return err;
2753 }
2755 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2756 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2757 }
2759 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2760 const char* mesg) {
2761 assert(mesg != NULL, "mesg must be specified");
2762 int err = os::Linux::commit_memory_impl(addr, size, exec);
2763 if (err != 0) {
2764 // the caller wants all commit errors to exit with the specified mesg:
2765 warn_fail_commit_memory(addr, size, exec, err);
2766 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2767 }
2768 }
2770 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2771 #ifndef MAP_HUGETLB
2772 #define MAP_HUGETLB 0x40000
2773 #endif
2775 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2776 #ifndef MADV_HUGEPAGE
2777 #define MADV_HUGEPAGE 14
2778 #endif
2780 int os::Linux::commit_memory_impl(char* addr, size_t size,
2781 size_t alignment_hint, bool exec) {
2782 int err = os::Linux::commit_memory_impl(addr, size, exec);
2783 if (err == 0) {
2784 realign_memory(addr, size, alignment_hint);
2785 }
2786 return err;
2787 }
2789 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2790 bool exec) {
2791 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2792 }
2794 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2795 size_t alignment_hint, bool exec,
2796 const char* mesg) {
2797 assert(mesg != NULL, "mesg must be specified");
2798 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2799 if (err != 0) {
2800 // the caller wants all commit errors to exit with the specified mesg:
2801 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2802 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2803 }
2804 }
2806 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2807 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2808 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2809 // be supported or the memory may already be backed by huge pages.
2810 ::madvise(addr, bytes, MADV_HUGEPAGE);
2811 }
2812 }
2814 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2815 // This method works by doing an mmap over an existing mmaping and effectively discarding
2816 // the existing pages. However it won't work for SHM-based large pages that cannot be
2817 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2818 // small pages on top of the SHM segment. This method always works for small pages, so we
2819 // allow that in any case.
2820 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2821 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2822 }
2823 }
2825 void os::numa_make_global(char *addr, size_t bytes) {
2826 Linux::numa_interleave_memory(addr, bytes);
2827 }
2829 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2830 // bind policy to MPOL_PREFERRED for the current thread.
2831 #define USE_MPOL_PREFERRED 0
2833 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2834 // To make NUMA and large pages more robust when both enabled, we need to ease
2835 // the requirements on where the memory should be allocated. MPOL_BIND is the
2836 // default policy and it will force memory to be allocated on the specified
2837 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2838 // the specified node, but will not force it. Using this policy will prevent
2839 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2840 // free large pages.
2841 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2842 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2843 }
2845 bool os::numa_topology_changed() { return false; }
2847 size_t os::numa_get_groups_num() {
2848 // Return just the number of nodes in which it's possible to allocate memory
2849 // (in numa terminology, configured nodes).
2850 return Linux::numa_num_configured_nodes();
2851 }
2853 int os::numa_get_group_id() {
2854 int cpu_id = Linux::sched_getcpu();
2855 if (cpu_id != -1) {
2856 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2857 if (lgrp_id != -1) {
2858 return lgrp_id;
2859 }
2860 }
2861 return 0;
2862 }
2864 int os::Linux::get_existing_num_nodes() {
2865 size_t node;
2866 size_t highest_node_number = Linux::numa_max_node();
2867 int num_nodes = 0;
2869 // Get the total number of nodes in the system including nodes without memory.
2870 for (node = 0; node <= highest_node_number; node++) {
2871 if (isnode_in_existing_nodes(node)) {
2872 num_nodes++;
2873 }
2874 }
2875 return num_nodes;
2876 }
2878 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2879 size_t highest_node_number = Linux::numa_max_node();
2880 size_t i = 0;
2882 // Map all node ids in which is possible to allocate memory. Also nodes are
2883 // not always consecutively available, i.e. available from 0 to the highest
2884 // node number.
2885 for (size_t node = 0; node <= highest_node_number; node++) {
2886 if (Linux::isnode_in_configured_nodes(node)) {
2887 ids[i++] = node;
2888 }
2889 }
2890 return i;
2891 }
2893 bool os::get_page_info(char *start, page_info* info) {
2894 return false;
2895 }
2897 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2898 return end;
2899 }
2902 int os::Linux::sched_getcpu_syscall(void) {
2903 unsigned int cpu = 0;
2904 int retval = -1;
2906 #if defined(IA32)
2907 # ifndef SYS_getcpu
2908 # define SYS_getcpu 318
2909 # endif
2910 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2911 #elif defined(AMD64)
2912 // Unfortunately we have to bring all these macros here from vsyscall.h
2913 // to be able to compile on old linuxes.
2914 # define __NR_vgetcpu 2
2915 # define VSYSCALL_START (-10UL << 20)
2916 # define VSYSCALL_SIZE 1024
2917 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2918 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2919 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2920 retval = vgetcpu(&cpu, NULL, NULL);
2921 #endif
2923 return (retval == -1) ? retval : cpu;
2924 }
2926 // Something to do with the numa-aware allocator needs these symbols
2927 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2928 extern "C" JNIEXPORT void numa_error(char *where) { }
2929 extern "C" JNIEXPORT int fork1() { return fork(); }
2931 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2932 // load symbol from base version instead.
2933 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2934 void *f = dlvsym(handle, name, "libnuma_1.1");
2935 if (f == NULL) {
2936 f = dlsym(handle, name);
2937 }
2938 return f;
2939 }
2941 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2942 // Return NULL if the symbol is not defined in this particular version.
2943 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2944 return dlvsym(handle, name, "libnuma_1.2");
2945 }
2947 bool os::Linux::libnuma_init() {
2948 // sched_getcpu() should be in libc.
2949 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2950 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2952 // If it's not, try a direct syscall.
2953 if (sched_getcpu() == -1)
2954 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2956 if (sched_getcpu() != -1) { // Does it work?
2957 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2958 if (handle != NULL) {
2959 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2960 libnuma_dlsym(handle, "numa_node_to_cpus")));
2961 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2962 libnuma_dlsym(handle, "numa_max_node")));
2963 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2964 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2965 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2966 libnuma_dlsym(handle, "numa_available")));
2967 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2968 libnuma_dlsym(handle, "numa_tonode_memory")));
2969 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2970 libnuma_dlsym(handle, "numa_interleave_memory")));
2971 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2972 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2973 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2974 libnuma_dlsym(handle, "numa_set_bind_policy")));
2975 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2976 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2977 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2978 libnuma_dlsym(handle, "numa_distance")));
2980 if (numa_available() != -1) {
2981 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2982 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2983 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2984 // Create an index -> node mapping, since nodes are not always consecutive
2985 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2986 rebuild_nindex_to_node_map();
2987 // Create a cpu -> node mapping
2988 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2989 rebuild_cpu_to_node_map();
2990 return true;
2991 }
2992 }
2993 }
2994 return false;
2995 }
2997 void os::Linux::rebuild_nindex_to_node_map() {
2998 int highest_node_number = Linux::numa_max_node();
3000 nindex_to_node()->clear();
3001 for (int node = 0; node <= highest_node_number; node++) {
3002 if (Linux::isnode_in_existing_nodes(node)) {
3003 nindex_to_node()->append(node);
3004 }
3005 }
3006 }
3008 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3009 // The table is later used in get_node_by_cpu().
3010 void os::Linux::rebuild_cpu_to_node_map() {
3011 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3012 // in libnuma (possible values are starting from 16,
3013 // and continuing up with every other power of 2, but less
3014 // than the maximum number of CPUs supported by kernel), and
3015 // is a subject to change (in libnuma version 2 the requirements
3016 // are more reasonable) we'll just hardcode the number they use
3017 // in the library.
3018 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3020 size_t cpu_num = processor_count();
3021 size_t cpu_map_size = NCPUS / BitsPerCLong;
3022 size_t cpu_map_valid_size =
3023 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3025 cpu_to_node()->clear();
3026 cpu_to_node()->at_grow(cpu_num - 1);
3028 size_t node_num = get_existing_num_nodes();
3030 int distance = 0;
3031 int closest_distance = INT_MAX;
3032 int closest_node = 0;
3033 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3034 for (size_t i = 0; i < node_num; i++) {
3035 // Check if node is configured (not a memory-less node). If it is not, find
3036 // the closest configured node.
3037 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3038 closest_distance = INT_MAX;
3039 // Check distance from all remaining nodes in the system. Ignore distance
3040 // from itself and from another non-configured node.
3041 for (size_t m = 0; m < node_num; m++) {
3042 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3043 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3044 // If a closest node is found, update. There is always at least one
3045 // configured node in the system so there is always at least one node
3046 // close.
3047 if (distance != 0 && distance < closest_distance) {
3048 closest_distance = distance;
3049 closest_node = nindex_to_node()->at(m);
3050 }
3051 }
3052 }
3053 } else {
3054 // Current node is already a configured node.
3055 closest_node = nindex_to_node()->at(i);
3056 }
3058 // Get cpus from the original node and map them to the closest node. If node
3059 // is a configured node (not a memory-less node), then original node and
3060 // closest node are the same.
3061 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3062 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3063 if (cpu_map[j] != 0) {
3064 for (size_t k = 0; k < BitsPerCLong; k++) {
3065 if (cpu_map[j] & (1UL << k)) {
3066 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3067 }
3068 }
3069 }
3070 }
3071 }
3072 }
3073 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
3074 }
3076 int os::Linux::get_node_by_cpu(int cpu_id) {
3077 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3078 return cpu_to_node()->at(cpu_id);
3079 }
3080 return -1;
3081 }
3083 GrowableArray<int>* os::Linux::_cpu_to_node;
3084 GrowableArray<int>* os::Linux::_nindex_to_node;
3085 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3086 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3087 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3088 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3089 os::Linux::numa_available_func_t os::Linux::_numa_available;
3090 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3091 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3092 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3093 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3094 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3095 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3096 unsigned long* os::Linux::_numa_all_nodes;
3097 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3098 struct bitmask* os::Linux::_numa_nodes_ptr;
3100 bool os::pd_uncommit_memory(char* addr, size_t size) {
3101 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3102 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3103 return res != (uintptr_t) MAP_FAILED;
3104 }
3106 static
3107 address get_stack_commited_bottom(address bottom, size_t size) {
3108 address nbot = bottom;
3109 address ntop = bottom + size;
3111 size_t page_sz = os::vm_page_size();
3112 unsigned pages = size / page_sz;
3114 unsigned char vec[1];
3115 unsigned imin = 1, imax = pages + 1, imid;
3116 int mincore_return_value = 0;
3118 assert(imin <= imax, "Unexpected page size");
3120 while (imin < imax) {
3121 imid = (imax + imin) / 2;
3122 nbot = ntop - (imid * page_sz);
3124 // Use a trick with mincore to check whether the page is mapped or not.
3125 // mincore sets vec to 1 if page resides in memory and to 0 if page
3126 // is swapped output but if page we are asking for is unmapped
3127 // it returns -1,ENOMEM
3128 mincore_return_value = mincore(nbot, page_sz, vec);
3130 if (mincore_return_value == -1) {
3131 // Page is not mapped go up
3132 // to find first mapped page
3133 if (errno != EAGAIN) {
3134 assert(errno == ENOMEM, "Unexpected mincore errno");
3135 imax = imid;
3136 }
3137 } else {
3138 // Page is mapped go down
3139 // to find first not mapped page
3140 imin = imid + 1;
3141 }
3142 }
3144 nbot = nbot + page_sz;
3146 // Adjust stack bottom one page up if last checked page is not mapped
3147 if (mincore_return_value == -1) {
3148 nbot = nbot + page_sz;
3149 }
3151 return nbot;
3152 }
3155 // Linux uses a growable mapping for the stack, and if the mapping for
3156 // the stack guard pages is not removed when we detach a thread the
3157 // stack cannot grow beyond the pages where the stack guard was
3158 // mapped. If at some point later in the process the stack expands to
3159 // that point, the Linux kernel cannot expand the stack any further
3160 // because the guard pages are in the way, and a segfault occurs.
3161 //
3162 // However, it's essential not to split the stack region by unmapping
3163 // a region (leaving a hole) that's already part of the stack mapping,
3164 // so if the stack mapping has already grown beyond the guard pages at
3165 // the time we create them, we have to truncate the stack mapping.
3166 // So, we need to know the extent of the stack mapping when
3167 // create_stack_guard_pages() is called.
3169 // We only need this for stacks that are growable: at the time of
3170 // writing thread stacks don't use growable mappings (i.e. those
3171 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3172 // only applies to the main thread.
3174 // If the (growable) stack mapping already extends beyond the point
3175 // where we're going to put our guard pages, truncate the mapping at
3176 // that point by munmap()ping it. This ensures that when we later
3177 // munmap() the guard pages we don't leave a hole in the stack
3178 // mapping. This only affects the main/primordial thread
3180 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3182 if (os::is_primordial_thread()) {
3183 // As we manually grow stack up to bottom inside create_attached_thread(),
3184 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3185 // we don't need to do anything special.
3186 // Check it first, before calling heavy function.
3187 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3188 unsigned char vec[1];
3190 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3191 // Fallback to slow path on all errors, including EAGAIN
3192 stack_extent = (uintptr_t) get_stack_commited_bottom(
3193 os::Linux::initial_thread_stack_bottom(),
3194 (size_t)addr - stack_extent);
3195 }
3197 if (stack_extent < (uintptr_t)addr) {
3198 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3199 }
3200 }
3202 return os::commit_memory(addr, size, !ExecMem);
3203 }
3205 // If this is a growable mapping, remove the guard pages entirely by
3206 // munmap()ping them. If not, just call uncommit_memory(). This only
3207 // affects the main/primordial thread, but guard against future OS changes.
3208 // It's safe to always unmap guard pages for primordial thread because we
3209 // always place it right after end of the mapped region.
3211 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3212 uintptr_t stack_extent, stack_base;
3214 if (os::is_primordial_thread()) {
3215 return ::munmap(addr, size) == 0;
3216 }
3218 return os::uncommit_memory(addr, size);
3219 }
3221 static address _highest_vm_reserved_address = NULL;
3223 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3224 // at 'requested_addr'. If there are existing memory mappings at the same
3225 // location, however, they will be overwritten. If 'fixed' is false,
3226 // 'requested_addr' is only treated as a hint, the return value may or
3227 // may not start from the requested address. Unlike Linux mmap(), this
3228 // function returns NULL to indicate failure.
3229 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3230 char * addr;
3231 int flags;
3233 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3234 if (fixed) {
3235 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3236 flags |= MAP_FIXED;
3237 }
3239 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3240 // touch an uncommitted page. Otherwise, the read/write might
3241 // succeed if we have enough swap space to back the physical page.
3242 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3243 flags, -1, 0);
3245 if (addr != MAP_FAILED) {
3246 // anon_mmap() should only get called during VM initialization,
3247 // don't need lock (actually we can skip locking even it can be called
3248 // from multiple threads, because _highest_vm_reserved_address is just a
3249 // hint about the upper limit of non-stack memory regions.)
3250 if ((address)addr + bytes > _highest_vm_reserved_address) {
3251 _highest_vm_reserved_address = (address)addr + bytes;
3252 }
3253 }
3255 return addr == MAP_FAILED ? NULL : addr;
3256 }
3258 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3259 // (req_addr != NULL) or with a given alignment.
3260 // - bytes shall be a multiple of alignment.
3261 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3262 // - alignment sets the alignment at which memory shall be allocated.
3263 // It must be a multiple of allocation granularity.
3264 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3265 // req_addr or NULL.
3266 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3268 size_t extra_size = bytes;
3269 if (req_addr == NULL && alignment > 0) {
3270 extra_size += alignment;
3271 }
3273 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3274 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3275 -1, 0);
3276 if (start == MAP_FAILED) {
3277 start = NULL;
3278 } else {
3279 if (req_addr != NULL) {
3280 if (start != req_addr) {
3281 ::munmap(start, extra_size);
3282 start = NULL;
3283 }
3284 } else {
3285 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3286 char* const end_aligned = start_aligned + bytes;
3287 char* const end = start + extra_size;
3288 if (start_aligned > start) {
3289 ::munmap(start, start_aligned - start);
3290 }
3291 if (end_aligned < end) {
3292 ::munmap(end_aligned, end - end_aligned);
3293 }
3294 start = start_aligned;
3295 }
3296 }
3297 return start;
3298 }
3300 // Don't update _highest_vm_reserved_address, because there might be memory
3301 // regions above addr + size. If so, releasing a memory region only creates
3302 // a hole in the address space, it doesn't help prevent heap-stack collision.
3303 //
3304 static int anon_munmap(char * addr, size_t size) {
3305 return ::munmap(addr, size) == 0;
3306 }
3308 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3309 size_t alignment_hint) {
3310 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3311 }
3313 bool os::pd_release_memory(char* addr, size_t size) {
3314 return anon_munmap(addr, size);
3315 }
3317 static address highest_vm_reserved_address() {
3318 return _highest_vm_reserved_address;
3319 }
3321 static bool linux_mprotect(char* addr, size_t size, int prot) {
3322 // Linux wants the mprotect address argument to be page aligned.
3323 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3325 // According to SUSv3, mprotect() should only be used with mappings
3326 // established by mmap(), and mmap() always maps whole pages. Unaligned
3327 // 'addr' likely indicates problem in the VM (e.g. trying to change
3328 // protection of malloc'ed or statically allocated memory). Check the
3329 // caller if you hit this assert.
3330 assert(addr == bottom, "sanity check");
3332 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3333 return ::mprotect(bottom, size, prot) == 0;
3334 }
3336 // Set protections specified
3337 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3338 bool is_committed) {
3339 unsigned int p = 0;
3340 switch (prot) {
3341 case MEM_PROT_NONE: p = PROT_NONE; break;
3342 case MEM_PROT_READ: p = PROT_READ; break;
3343 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3344 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3345 default:
3346 ShouldNotReachHere();
3347 }
3348 // is_committed is unused.
3349 return linux_mprotect(addr, bytes, p);
3350 }
3352 bool os::guard_memory(char* addr, size_t size) {
3353 return linux_mprotect(addr, size, PROT_NONE);
3354 }
3356 bool os::unguard_memory(char* addr, size_t size) {
3357 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3358 }
3360 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3361 bool result = false;
3362 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3363 MAP_ANONYMOUS|MAP_PRIVATE,
3364 -1, 0);
3365 if (p != MAP_FAILED) {
3366 void *aligned_p = align_ptr_up(p, page_size);
3368 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3370 munmap(p, page_size * 2);
3371 }
3373 if (warn && !result) {
3374 warning("TransparentHugePages is not supported by the operating system.");
3375 }
3377 return result;
3378 }
3380 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3381 bool result = false;
3382 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3383 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3384 -1, 0);
3386 if (p != MAP_FAILED) {
3387 // We don't know if this really is a huge page or not.
3388 FILE *fp = fopen("/proc/self/maps", "r");
3389 if (fp) {
3390 while (!feof(fp)) {
3391 char chars[257];
3392 long x = 0;
3393 if (fgets(chars, sizeof(chars), fp)) {
3394 if (sscanf(chars, "%lx-%*x", &x) == 1
3395 && x == (long)p) {
3396 if (strstr (chars, "hugepage")) {
3397 result = true;
3398 break;
3399 }
3400 }
3401 }
3402 }
3403 fclose(fp);
3404 }
3405 munmap(p, page_size);
3406 }
3408 if (warn && !result) {
3409 warning("HugeTLBFS is not supported by the operating system.");
3410 }
3412 return result;
3413 }
3415 /*
3416 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3417 *
3418 * From the coredump_filter documentation:
3419 *
3420 * - (bit 0) anonymous private memory
3421 * - (bit 1) anonymous shared memory
3422 * - (bit 2) file-backed private memory
3423 * - (bit 3) file-backed shared memory
3424 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3425 * effective only if the bit 2 is cleared)
3426 * - (bit 5) hugetlb private memory
3427 * - (bit 6) hugetlb shared memory
3428 */
3429 static void set_coredump_filter(void) {
3430 FILE *f;
3431 long cdm;
3433 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3434 return;
3435 }
3437 if (fscanf(f, "%lx", &cdm) != 1) {
3438 fclose(f);
3439 return;
3440 }
3442 rewind(f);
3444 if ((cdm & LARGEPAGES_BIT) == 0) {
3445 cdm |= LARGEPAGES_BIT;
3446 fprintf(f, "%#lx", cdm);
3447 }
3449 fclose(f);
3450 }
3452 // Large page support
3454 static size_t _large_page_size = 0;
3456 size_t os::Linux::find_large_page_size() {
3457 size_t large_page_size = 0;
3459 // large_page_size on Linux is used to round up heap size. x86 uses either
3460 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3461 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3462 // page as large as 256M.
3463 //
3464 // Here we try to figure out page size by parsing /proc/meminfo and looking
3465 // for a line with the following format:
3466 // Hugepagesize: 2048 kB
3467 //
3468 // If we can't determine the value (e.g. /proc is not mounted, or the text
3469 // format has been changed), we'll use the largest page size supported by
3470 // the processor.
3472 #ifndef ZERO
3473 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3474 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3475 #endif // ZERO
3477 FILE *fp = fopen("/proc/meminfo", "r");
3478 if (fp) {
3479 while (!feof(fp)) {
3480 int x = 0;
3481 char buf[16];
3482 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3483 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3484 large_page_size = x * K;
3485 break;
3486 }
3487 } else {
3488 // skip to next line
3489 for (;;) {
3490 int ch = fgetc(fp);
3491 if (ch == EOF || ch == (int)'\n') break;
3492 }
3493 }
3494 }
3495 fclose(fp);
3496 }
3498 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3499 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3500 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3501 proper_unit_for_byte_size(large_page_size));
3502 }
3504 return large_page_size;
3505 }
3507 size_t os::Linux::setup_large_page_size() {
3508 _large_page_size = Linux::find_large_page_size();
3509 const size_t default_page_size = (size_t)Linux::page_size();
3510 if (_large_page_size > default_page_size) {
3511 _page_sizes[0] = _large_page_size;
3512 _page_sizes[1] = default_page_size;
3513 _page_sizes[2] = 0;
3514 }
3516 return _large_page_size;
3517 }
3519 bool os::Linux::setup_large_page_type(size_t page_size) {
3520 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3521 FLAG_IS_DEFAULT(UseSHM) &&
3522 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3524 // The type of large pages has not been specified by the user.
3526 // Try UseHugeTLBFS and then UseSHM.
3527 UseHugeTLBFS = UseSHM = true;
3529 // Don't try UseTransparentHugePages since there are known
3530 // performance issues with it turned on. This might change in the future.
3531 UseTransparentHugePages = false;
3532 }
3534 if (UseTransparentHugePages) {
3535 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3536 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3537 UseHugeTLBFS = false;
3538 UseSHM = false;
3539 return true;
3540 }
3541 UseTransparentHugePages = false;
3542 }
3544 if (UseHugeTLBFS) {
3545 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3546 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3547 UseSHM = false;
3548 return true;
3549 }
3550 UseHugeTLBFS = false;
3551 }
3553 return UseSHM;
3554 }
3556 void os::large_page_init() {
3557 if (!UseLargePages &&
3558 !UseTransparentHugePages &&
3559 !UseHugeTLBFS &&
3560 !UseSHM) {
3561 // Not using large pages.
3562 return;
3563 }
3565 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3566 // The user explicitly turned off large pages.
3567 // Ignore the rest of the large pages flags.
3568 UseTransparentHugePages = false;
3569 UseHugeTLBFS = false;
3570 UseSHM = false;
3571 return;
3572 }
3574 size_t large_page_size = Linux::setup_large_page_size();
3575 UseLargePages = Linux::setup_large_page_type(large_page_size);
3577 set_coredump_filter();
3578 }
3580 #ifndef SHM_HUGETLB
3581 #define SHM_HUGETLB 04000
3582 #endif
3584 #define shm_warning_format(format, ...) \
3585 do { \
3586 if (UseLargePages && \
3587 (!FLAG_IS_DEFAULT(UseLargePages) || \
3588 !FLAG_IS_DEFAULT(UseSHM) || \
3589 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3590 warning(format, __VA_ARGS__); \
3591 } \
3592 } while (0)
3594 #define shm_warning(str) shm_warning_format("%s", str)
3596 #define shm_warning_with_errno(str) \
3597 do { \
3598 int err = errno; \
3599 shm_warning_format(str " (error = %d)", err); \
3600 } while (0)
3602 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3603 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3605 if (!is_size_aligned(alignment, SHMLBA)) {
3606 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3607 return NULL;
3608 }
3610 // To ensure that we get 'alignment' aligned memory from shmat,
3611 // we pre-reserve aligned virtual memory and then attach to that.
3613 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3614 if (pre_reserved_addr == NULL) {
3615 // Couldn't pre-reserve aligned memory.
3616 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3617 return NULL;
3618 }
3620 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3621 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3623 if ((intptr_t)addr == -1) {
3624 int err = errno;
3625 shm_warning_with_errno("Failed to attach shared memory.");
3627 assert(err != EACCES, "Unexpected error");
3628 assert(err != EIDRM, "Unexpected error");
3629 assert(err != EINVAL, "Unexpected error");
3631 // Since we don't know if the kernel unmapped the pre-reserved memory area
3632 // we can't unmap it, since that would potentially unmap memory that was
3633 // mapped from other threads.
3634 return NULL;
3635 }
3637 return addr;
3638 }
3640 static char* shmat_at_address(int shmid, char* req_addr) {
3641 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3642 assert(false, "Requested address needs to be SHMLBA aligned");
3643 return NULL;
3644 }
3646 char* addr = (char*)shmat(shmid, req_addr, 0);
3648 if ((intptr_t)addr == -1) {
3649 shm_warning_with_errno("Failed to attach shared memory.");
3650 return NULL;
3651 }
3653 return addr;
3654 }
3656 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3657 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3658 if (req_addr != NULL) {
3659 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3660 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3661 return shmat_at_address(shmid, req_addr);
3662 }
3664 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3665 // return large page size aligned memory addresses when req_addr == NULL.
3666 // However, if the alignment is larger than the large page size, we have
3667 // to manually ensure that the memory returned is 'alignment' aligned.
3668 if (alignment > os::large_page_size()) {
3669 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3670 return shmat_with_alignment(shmid, bytes, alignment);
3671 } else {
3672 return shmat_at_address(shmid, NULL);
3673 }
3674 }
3676 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3677 // "exec" is passed in but not used. Creating the shared image for
3678 // the code cache doesn't have an SHM_X executable permission to check.
3679 assert(UseLargePages && UseSHM, "only for SHM large pages");
3680 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3681 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3683 if (!is_size_aligned(bytes, os::large_page_size())) {
3684 return NULL; // Fallback to small pages.
3685 }
3687 // Create a large shared memory region to attach to based on size.
3688 // Currently, size is the total size of the heap.
3689 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3690 if (shmid == -1) {
3691 // Possible reasons for shmget failure:
3692 // 1. shmmax is too small for Java heap.
3693 // > check shmmax value: cat /proc/sys/kernel/shmmax
3694 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3695 // 2. not enough large page memory.
3696 // > check available large pages: cat /proc/meminfo
3697 // > increase amount of large pages:
3698 // echo new_value > /proc/sys/vm/nr_hugepages
3699 // Note 1: different Linux may use different name for this property,
3700 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3701 // Note 2: it's possible there's enough physical memory available but
3702 // they are so fragmented after a long run that they can't
3703 // coalesce into large pages. Try to reserve large pages when
3704 // the system is still "fresh".
3705 shm_warning_with_errno("Failed to reserve shared memory.");
3706 return NULL;
3707 }
3709 // Attach to the region.
3710 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3712 // Remove shmid. If shmat() is successful, the actual shared memory segment
3713 // will be deleted when it's detached by shmdt() or when the process
3714 // terminates. If shmat() is not successful this will remove the shared
3715 // segment immediately.
3716 shmctl(shmid, IPC_RMID, NULL);
3718 return addr;
3719 }
3721 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3722 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3724 bool warn_on_failure = UseLargePages &&
3725 (!FLAG_IS_DEFAULT(UseLargePages) ||
3726 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3727 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3729 if (warn_on_failure) {
3730 char msg[128];
3731 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3732 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3733 warning("%s", msg);
3734 }
3735 }
3737 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3738 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3739 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3740 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3742 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3743 char* addr = (char*)::mmap(req_addr, bytes, prot,
3744 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3745 -1, 0);
3747 if (addr == MAP_FAILED) {
3748 warn_on_large_pages_failure(req_addr, bytes, errno);
3749 return NULL;
3750 }
3752 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3754 return addr;
3755 }
3757 // Reserve memory using mmap(MAP_HUGETLB).
3758 // - bytes shall be a multiple of alignment.
3759 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3760 // - alignment sets the alignment at which memory shall be allocated.
3761 // It must be a multiple of allocation granularity.
3762 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3763 // req_addr or NULL.
3764 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3765 size_t large_page_size = os::large_page_size();
3766 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3768 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3769 assert(is_size_aligned(bytes, alignment), "Must be");
3771 // First reserve - but not commit - the address range in small pages.
3772 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3774 if (start == NULL) {
3775 return NULL;
3776 }
3778 assert(is_ptr_aligned(start, alignment), "Must be");
3780 char* end = start + bytes;
3782 // Find the regions of the allocated chunk that can be promoted to large pages.
3783 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3784 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3786 size_t lp_bytes = lp_end - lp_start;
3788 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3790 if (lp_bytes == 0) {
3791 // The mapped region doesn't even span the start and the end of a large page.
3792 // Fall back to allocate a non-special area.
3793 ::munmap(start, end - start);
3794 return NULL;
3795 }
3797 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3799 void* result;
3801 // Commit small-paged leading area.
3802 if (start != lp_start) {
3803 result = ::mmap(start, lp_start - start, prot,
3804 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3805 -1, 0);
3806 if (result == MAP_FAILED) {
3807 ::munmap(lp_start, end - lp_start);
3808 return NULL;
3809 }
3810 }
3812 // Commit large-paged area.
3813 result = ::mmap(lp_start, lp_bytes, prot,
3814 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3815 -1, 0);
3816 if (result == MAP_FAILED) {
3817 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3818 // If the mmap above fails, the large pages region will be unmapped and we
3819 // have regions before and after with small pages. Release these regions.
3820 //
3821 // | mapped | unmapped | mapped |
3822 // ^ ^ ^ ^
3823 // start lp_start lp_end end
3824 //
3825 ::munmap(start, lp_start - start);
3826 ::munmap(lp_end, end - lp_end);
3827 return NULL;
3828 }
3830 // Commit small-paged trailing area.
3831 if (lp_end != end) {
3832 result = ::mmap(lp_end, end - lp_end, prot,
3833 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3834 -1, 0);
3835 if (result == MAP_FAILED) {
3836 ::munmap(start, lp_end - start);
3837 return NULL;
3838 }
3839 }
3841 return start;
3842 }
3844 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3845 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3846 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3847 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3848 assert(is_power_of_2(os::large_page_size()), "Must be");
3849 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3851 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3852 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3853 } else {
3854 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3855 }
3856 }
3858 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3859 assert(UseLargePages, "only for large pages");
3861 char* addr;
3862 if (UseSHM) {
3863 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3864 } else {
3865 assert(UseHugeTLBFS, "must be");
3866 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3867 }
3869 if (addr != NULL) {
3870 if (UseNUMAInterleaving) {
3871 numa_make_global(addr, bytes);
3872 }
3874 // The memory is committed
3875 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3876 }
3878 return addr;
3879 }
3881 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3882 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3883 return shmdt(base) == 0;
3884 }
3886 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3887 return pd_release_memory(base, bytes);
3888 }
3890 bool os::release_memory_special(char* base, size_t bytes) {
3891 bool res;
3892 if (MemTracker::tracking_level() > NMT_minimal) {
3893 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3894 res = os::Linux::release_memory_special_impl(base, bytes);
3895 if (res) {
3896 tkr.record((address)base, bytes);
3897 }
3899 } else {
3900 res = os::Linux::release_memory_special_impl(base, bytes);
3901 }
3902 return res;
3903 }
3905 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3906 assert(UseLargePages, "only for large pages");
3907 bool res;
3909 if (UseSHM) {
3910 res = os::Linux::release_memory_special_shm(base, bytes);
3911 } else {
3912 assert(UseHugeTLBFS, "must be");
3913 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3914 }
3915 return res;
3916 }
3918 size_t os::large_page_size() {
3919 return _large_page_size;
3920 }
3922 // With SysV SHM the entire memory region must be allocated as shared
3923 // memory.
3924 // HugeTLBFS allows application to commit large page memory on demand.
3925 // However, when committing memory with HugeTLBFS fails, the region
3926 // that was supposed to be committed will lose the old reservation
3927 // and allow other threads to steal that memory region. Because of this
3928 // behavior we can't commit HugeTLBFS memory.
3929 bool os::can_commit_large_page_memory() {
3930 return UseTransparentHugePages;
3931 }
3933 bool os::can_execute_large_page_memory() {
3934 return UseTransparentHugePages || UseHugeTLBFS;
3935 }
3937 // Reserve memory at an arbitrary address, only if that area is
3938 // available (and not reserved for something else).
3940 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3941 const int max_tries = 10;
3942 char* base[max_tries];
3943 size_t size[max_tries];
3944 const size_t gap = 0x000000;
3946 // Assert only that the size is a multiple of the page size, since
3947 // that's all that mmap requires, and since that's all we really know
3948 // about at this low abstraction level. If we need higher alignment,
3949 // we can either pass an alignment to this method or verify alignment
3950 // in one of the methods further up the call chain. See bug 5044738.
3951 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3953 // Repeatedly allocate blocks until the block is allocated at the
3954 // right spot. Give up after max_tries. Note that reserve_memory() will
3955 // automatically update _highest_vm_reserved_address if the call is
3956 // successful. The variable tracks the highest memory address every reserved
3957 // by JVM. It is used to detect heap-stack collision if running with
3958 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3959 // space than needed, it could confuse the collision detecting code. To
3960 // solve the problem, save current _highest_vm_reserved_address and
3961 // calculate the correct value before return.
3962 address old_highest = _highest_vm_reserved_address;
3964 // Linux mmap allows caller to pass an address as hint; give it a try first,
3965 // if kernel honors the hint then we can return immediately.
3966 char * addr = anon_mmap(requested_addr, bytes, false);
3967 if (addr == requested_addr) {
3968 return requested_addr;
3969 }
3971 if (addr != NULL) {
3972 // mmap() is successful but it fails to reserve at the requested address
3973 anon_munmap(addr, bytes);
3974 }
3976 int i;
3977 for (i = 0; i < max_tries; ++i) {
3978 base[i] = reserve_memory(bytes);
3980 if (base[i] != NULL) {
3981 // Is this the block we wanted?
3982 if (base[i] == requested_addr) {
3983 size[i] = bytes;
3984 break;
3985 }
3987 // Does this overlap the block we wanted? Give back the overlapped
3988 // parts and try again.
3990 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3991 if (top_overlap >= 0 && top_overlap < bytes) {
3992 unmap_memory(base[i], top_overlap);
3993 base[i] += top_overlap;
3994 size[i] = bytes - top_overlap;
3995 } else {
3996 size_t bottom_overlap = base[i] + bytes - requested_addr;
3997 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3998 unmap_memory(requested_addr, bottom_overlap);
3999 size[i] = bytes - bottom_overlap;
4000 } else {
4001 size[i] = bytes;
4002 }
4003 }
4004 }
4005 }
4007 // Give back the unused reserved pieces.
4009 for (int j = 0; j < i; ++j) {
4010 if (base[j] != NULL) {
4011 unmap_memory(base[j], size[j]);
4012 }
4013 }
4015 if (i < max_tries) {
4016 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
4017 return requested_addr;
4018 } else {
4019 _highest_vm_reserved_address = old_highest;
4020 return NULL;
4021 }
4022 }
4024 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4025 return ::read(fd, buf, nBytes);
4026 }
4028 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
4029 // Solaris uses poll(), linux uses park().
4030 // Poll() is likely a better choice, assuming that Thread.interrupt()
4031 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
4032 // SIGSEGV, see 4355769.
4034 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
4035 assert(thread == Thread::current(), "thread consistency check");
4037 ParkEvent * const slp = thread->_SleepEvent ;
4038 slp->reset() ;
4039 OrderAccess::fence() ;
4041 if (interruptible) {
4042 jlong prevtime = javaTimeNanos();
4044 for (;;) {
4045 if (os::is_interrupted(thread, true)) {
4046 return OS_INTRPT;
4047 }
4049 jlong newtime = javaTimeNanos();
4051 if (newtime - prevtime < 0) {
4052 // time moving backwards, should only happen if no monotonic clock
4053 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4054 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4055 } else {
4056 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4057 }
4059 if(millis <= 0) {
4060 return OS_OK;
4061 }
4063 prevtime = newtime;
4065 {
4066 assert(thread->is_Java_thread(), "sanity check");
4067 JavaThread *jt = (JavaThread *) thread;
4068 ThreadBlockInVM tbivm(jt);
4069 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
4071 jt->set_suspend_equivalent();
4072 // cleared by handle_special_suspend_equivalent_condition() or
4073 // java_suspend_self() via check_and_wait_while_suspended()
4075 slp->park(millis);
4077 // were we externally suspended while we were waiting?
4078 jt->check_and_wait_while_suspended();
4079 }
4080 }
4081 } else {
4082 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4083 jlong prevtime = javaTimeNanos();
4085 for (;;) {
4086 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
4087 // the 1st iteration ...
4088 jlong newtime = javaTimeNanos();
4090 if (newtime - prevtime < 0) {
4091 // time moving backwards, should only happen if no monotonic clock
4092 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4093 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4094 } else {
4095 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4096 }
4098 if(millis <= 0) break ;
4100 prevtime = newtime;
4101 slp->park(millis);
4102 }
4103 return OS_OK ;
4104 }
4105 }
4107 //
4108 // Short sleep, direct OS call.
4109 //
4110 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4111 // sched_yield(2) will actually give up the CPU:
4112 //
4113 // * Alone on this pariticular CPU, keeps running.
4114 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4115 // (pre 2.6.39).
4116 //
4117 // So calling this with 0 is an alternative.
4118 //
4119 void os::naked_short_sleep(jlong ms) {
4120 struct timespec req;
4122 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4123 req.tv_sec = 0;
4124 if (ms > 0) {
4125 req.tv_nsec = (ms % 1000) * 1000000;
4126 }
4127 else {
4128 req.tv_nsec = 1;
4129 }
4131 nanosleep(&req, NULL);
4133 return;
4134 }
4136 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4137 void os::infinite_sleep() {
4138 while (true) { // sleep forever ...
4139 ::sleep(100); // ... 100 seconds at a time
4140 }
4141 }
4143 // Used to convert frequent JVM_Yield() to nops
4144 bool os::dont_yield() {
4145 return DontYieldALot;
4146 }
4148 void os::yield() {
4149 sched_yield();
4150 }
4152 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
4154 void os::yield_all(int attempts) {
4155 // Yields to all threads, including threads with lower priorities
4156 // Threads on Linux are all with same priority. The Solaris style
4157 // os::yield_all() with nanosleep(1ms) is not necessary.
4158 sched_yield();
4159 }
4161 // Called from the tight loops to possibly influence time-sharing heuristics
4162 void os::loop_breaker(int attempts) {
4163 os::yield_all(attempts);
4164 }
4166 ////////////////////////////////////////////////////////////////////////////////
4167 // thread priority support
4169 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4170 // only supports dynamic priority, static priority must be zero. For real-time
4171 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4172 // However, for large multi-threaded applications, SCHED_RR is not only slower
4173 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4174 // of 5 runs - Sep 2005).
4175 //
4176 // The following code actually changes the niceness of kernel-thread/LWP. It
4177 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4178 // not the entire user process, and user level threads are 1:1 mapped to kernel
4179 // threads. It has always been the case, but could change in the future. For
4180 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4181 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4183 int os::java_to_os_priority[CriticalPriority + 1] = {
4184 19, // 0 Entry should never be used
4186 4, // 1 MinPriority
4187 3, // 2
4188 2, // 3
4190 1, // 4
4191 0, // 5 NormPriority
4192 -1, // 6
4194 -2, // 7
4195 -3, // 8
4196 -4, // 9 NearMaxPriority
4198 -5, // 10 MaxPriority
4200 -5 // 11 CriticalPriority
4201 };
4203 static int prio_init() {
4204 if (ThreadPriorityPolicy == 1) {
4205 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4206 // if effective uid is not root. Perhaps, a more elegant way of doing
4207 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4208 if (geteuid() != 0) {
4209 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4210 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4211 }
4212 ThreadPriorityPolicy = 0;
4213 }
4214 }
4215 if (UseCriticalJavaThreadPriority) {
4216 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4217 }
4218 return 0;
4219 }
4221 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4222 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4224 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4225 return (ret == 0) ? OS_OK : OS_ERR;
4226 }
4228 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4229 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4230 *priority_ptr = java_to_os_priority[NormPriority];
4231 return OS_OK;
4232 }
4234 errno = 0;
4235 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4236 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4237 }
4239 // Hint to the underlying OS that a task switch would not be good.
4240 // Void return because it's a hint and can fail.
4241 void os::hint_no_preempt() {}
4243 ////////////////////////////////////////////////////////////////////////////////
4244 // suspend/resume support
4246 // the low-level signal-based suspend/resume support is a remnant from the
4247 // old VM-suspension that used to be for java-suspension, safepoints etc,
4248 // within hotspot. Now there is a single use-case for this:
4249 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4250 // that runs in the watcher thread.
4251 // The remaining code is greatly simplified from the more general suspension
4252 // code that used to be used.
4253 //
4254 // The protocol is quite simple:
4255 // - suspend:
4256 // - sends a signal to the target thread
4257 // - polls the suspend state of the osthread using a yield loop
4258 // - target thread signal handler (SR_handler) sets suspend state
4259 // and blocks in sigsuspend until continued
4260 // - resume:
4261 // - sets target osthread state to continue
4262 // - sends signal to end the sigsuspend loop in the SR_handler
4263 //
4264 // Note that the SR_lock plays no role in this suspend/resume protocol.
4265 //
4267 static void resume_clear_context(OSThread *osthread) {
4268 osthread->set_ucontext(NULL);
4269 osthread->set_siginfo(NULL);
4270 }
4272 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4273 osthread->set_ucontext(context);
4274 osthread->set_siginfo(siginfo);
4275 }
4277 //
4278 // Handler function invoked when a thread's execution is suspended or
4279 // resumed. We have to be careful that only async-safe functions are
4280 // called here (Note: most pthread functions are not async safe and
4281 // should be avoided.)
4282 //
4283 // Note: sigwait() is a more natural fit than sigsuspend() from an
4284 // interface point of view, but sigwait() prevents the signal hander
4285 // from being run. libpthread would get very confused by not having
4286 // its signal handlers run and prevents sigwait()'s use with the
4287 // mutex granting granting signal.
4288 //
4289 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4290 //
4291 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4292 // Save and restore errno to avoid confusing native code with EINTR
4293 // after sigsuspend.
4294 int old_errno = errno;
4296 Thread* thread = Thread::current();
4297 OSThread* osthread = thread->osthread();
4298 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4300 os::SuspendResume::State current = osthread->sr.state();
4301 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4302 suspend_save_context(osthread, siginfo, context);
4304 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4305 os::SuspendResume::State state = osthread->sr.suspended();
4306 if (state == os::SuspendResume::SR_SUSPENDED) {
4307 sigset_t suspend_set; // signals for sigsuspend()
4309 // get current set of blocked signals and unblock resume signal
4310 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4311 sigdelset(&suspend_set, SR_signum);
4313 sr_semaphore.signal();
4314 // wait here until we are resumed
4315 while (1) {
4316 sigsuspend(&suspend_set);
4318 os::SuspendResume::State result = osthread->sr.running();
4319 if (result == os::SuspendResume::SR_RUNNING) {
4320 sr_semaphore.signal();
4321 break;
4322 }
4323 }
4325 } else if (state == os::SuspendResume::SR_RUNNING) {
4326 // request was cancelled, continue
4327 } else {
4328 ShouldNotReachHere();
4329 }
4331 resume_clear_context(osthread);
4332 } else if (current == os::SuspendResume::SR_RUNNING) {
4333 // request was cancelled, continue
4334 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4335 // ignore
4336 } else {
4337 // ignore
4338 }
4340 errno = old_errno;
4341 }
4344 static int SR_initialize() {
4345 struct sigaction act;
4346 char *s;
4347 /* Get signal number to use for suspend/resume */
4348 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4349 int sig = ::strtol(s, 0, 10);
4350 if (sig > 0 || sig < _NSIG) {
4351 SR_signum = sig;
4352 }
4353 }
4355 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4356 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4358 sigemptyset(&SR_sigset);
4359 sigaddset(&SR_sigset, SR_signum);
4361 /* Set up signal handler for suspend/resume */
4362 act.sa_flags = SA_RESTART|SA_SIGINFO;
4363 act.sa_handler = (void (*)(int)) SR_handler;
4365 // SR_signum is blocked by default.
4366 // 4528190 - We also need to block pthread restart signal (32 on all
4367 // supported Linux platforms). Note that LinuxThreads need to block
4368 // this signal for all threads to work properly. So we don't have
4369 // to use hard-coded signal number when setting up the mask.
4370 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4372 if (sigaction(SR_signum, &act, 0) == -1) {
4373 return -1;
4374 }
4376 // Save signal flag
4377 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4378 return 0;
4379 }
4381 static int sr_notify(OSThread* osthread) {
4382 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4383 assert_status(status == 0, status, "pthread_kill");
4384 return status;
4385 }
4387 // "Randomly" selected value for how long we want to spin
4388 // before bailing out on suspending a thread, also how often
4389 // we send a signal to a thread we want to resume
4390 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4391 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4393 // returns true on success and false on error - really an error is fatal
4394 // but this seems the normal response to library errors
4395 static bool do_suspend(OSThread* osthread) {
4396 assert(osthread->sr.is_running(), "thread should be running");
4397 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4399 // mark as suspended and send signal
4400 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4401 // failed to switch, state wasn't running?
4402 ShouldNotReachHere();
4403 return false;
4404 }
4406 if (sr_notify(osthread) != 0) {
4407 ShouldNotReachHere();
4408 }
4410 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4411 while (true) {
4412 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4413 break;
4414 } else {
4415 // timeout
4416 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4417 if (cancelled == os::SuspendResume::SR_RUNNING) {
4418 return false;
4419 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4420 // make sure that we consume the signal on the semaphore as well
4421 sr_semaphore.wait();
4422 break;
4423 } else {
4424 ShouldNotReachHere();
4425 return false;
4426 }
4427 }
4428 }
4430 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4431 return true;
4432 }
4434 static void do_resume(OSThread* osthread) {
4435 assert(osthread->sr.is_suspended(), "thread should be suspended");
4436 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4438 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4439 // failed to switch to WAKEUP_REQUEST
4440 ShouldNotReachHere();
4441 return;
4442 }
4444 while (true) {
4445 if (sr_notify(osthread) == 0) {
4446 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4447 if (osthread->sr.is_running()) {
4448 return;
4449 }
4450 }
4451 } else {
4452 ShouldNotReachHere();
4453 }
4454 }
4456 guarantee(osthread->sr.is_running(), "Must be running!");
4457 }
4459 ////////////////////////////////////////////////////////////////////////////////
4460 // interrupt support
4462 void os::interrupt(Thread* thread) {
4463 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4464 "possibility of dangling Thread pointer");
4466 OSThread* osthread = thread->osthread();
4468 if (!osthread->interrupted()) {
4469 osthread->set_interrupted(true);
4470 // More than one thread can get here with the same value of osthread,
4471 // resulting in multiple notifications. We do, however, want the store
4472 // to interrupted() to be visible to other threads before we execute unpark().
4473 OrderAccess::fence();
4474 ParkEvent * const slp = thread->_SleepEvent ;
4475 if (slp != NULL) slp->unpark() ;
4476 }
4478 // For JSR166. Unpark even if interrupt status already was set
4479 if (thread->is_Java_thread())
4480 ((JavaThread*)thread)->parker()->unpark();
4482 ParkEvent * ev = thread->_ParkEvent ;
4483 if (ev != NULL) ev->unpark() ;
4485 }
4487 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4488 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4489 "possibility of dangling Thread pointer");
4491 OSThread* osthread = thread->osthread();
4493 bool interrupted = osthread->interrupted();
4495 if (interrupted && clear_interrupted) {
4496 osthread->set_interrupted(false);
4497 // consider thread->_SleepEvent->reset() ... optional optimization
4498 }
4500 return interrupted;
4501 }
4503 ///////////////////////////////////////////////////////////////////////////////////
4504 // signal handling (except suspend/resume)
4506 // This routine may be used by user applications as a "hook" to catch signals.
4507 // The user-defined signal handler must pass unrecognized signals to this
4508 // routine, and if it returns true (non-zero), then the signal handler must
4509 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4510 // routine will never retun false (zero), but instead will execute a VM panic
4511 // routine kill the process.
4512 //
4513 // If this routine returns false, it is OK to call it again. This allows
4514 // the user-defined signal handler to perform checks either before or after
4515 // the VM performs its own checks. Naturally, the user code would be making
4516 // a serious error if it tried to handle an exception (such as a null check
4517 // or breakpoint) that the VM was generating for its own correct operation.
4518 //
4519 // This routine may recognize any of the following kinds of signals:
4520 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4521 // It should be consulted by handlers for any of those signals.
4522 //
4523 // The caller of this routine must pass in the three arguments supplied
4524 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4525 // field of the structure passed to sigaction(). This routine assumes that
4526 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4527 //
4528 // Note that the VM will print warnings if it detects conflicting signal
4529 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4530 //
4531 extern "C" JNIEXPORT int
4532 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4533 void* ucontext, int abort_if_unrecognized);
4535 void signalHandler(int sig, siginfo_t* info, void* uc) {
4536 assert(info != NULL && uc != NULL, "it must be old kernel");
4537 int orig_errno = errno; // Preserve errno value over signal handler.
4538 JVM_handle_linux_signal(sig, info, uc, true);
4539 errno = orig_errno;
4540 }
4543 // This boolean allows users to forward their own non-matching signals
4544 // to JVM_handle_linux_signal, harmlessly.
4545 bool os::Linux::signal_handlers_are_installed = false;
4547 // For signal-chaining
4548 struct sigaction os::Linux::sigact[MAXSIGNUM];
4549 unsigned int os::Linux::sigs = 0;
4550 bool os::Linux::libjsig_is_loaded = false;
4551 typedef struct sigaction *(*get_signal_t)(int);
4552 get_signal_t os::Linux::get_signal_action = NULL;
4554 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4555 struct sigaction *actp = NULL;
4557 if (libjsig_is_loaded) {
4558 // Retrieve the old signal handler from libjsig
4559 actp = (*get_signal_action)(sig);
4560 }
4561 if (actp == NULL) {
4562 // Retrieve the preinstalled signal handler from jvm
4563 actp = get_preinstalled_handler(sig);
4564 }
4566 return actp;
4567 }
4569 static bool call_chained_handler(struct sigaction *actp, int sig,
4570 siginfo_t *siginfo, void *context) {
4571 // Call the old signal handler
4572 if (actp->sa_handler == SIG_DFL) {
4573 // It's more reasonable to let jvm treat it as an unexpected exception
4574 // instead of taking the default action.
4575 return false;
4576 } else if (actp->sa_handler != SIG_IGN) {
4577 if ((actp->sa_flags & SA_NODEFER) == 0) {
4578 // automaticlly block the signal
4579 sigaddset(&(actp->sa_mask), sig);
4580 }
4582 sa_handler_t hand = NULL;
4583 sa_sigaction_t sa = NULL;
4584 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4585 // retrieve the chained handler
4586 if (siginfo_flag_set) {
4587 sa = actp->sa_sigaction;
4588 } else {
4589 hand = actp->sa_handler;
4590 }
4592 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4593 actp->sa_handler = SIG_DFL;
4594 }
4596 // try to honor the signal mask
4597 sigset_t oset;
4598 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4600 // call into the chained handler
4601 if (siginfo_flag_set) {
4602 (*sa)(sig, siginfo, context);
4603 } else {
4604 (*hand)(sig);
4605 }
4607 // restore the signal mask
4608 pthread_sigmask(SIG_SETMASK, &oset, 0);
4609 }
4610 // Tell jvm's signal handler the signal is taken care of.
4611 return true;
4612 }
4614 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4615 bool chained = false;
4616 // signal-chaining
4617 if (UseSignalChaining) {
4618 struct sigaction *actp = get_chained_signal_action(sig);
4619 if (actp != NULL) {
4620 chained = call_chained_handler(actp, sig, siginfo, context);
4621 }
4622 }
4623 return chained;
4624 }
4626 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4627 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4628 return &sigact[sig];
4629 }
4630 return NULL;
4631 }
4633 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4634 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4635 sigact[sig] = oldAct;
4636 sigs |= (unsigned int)1 << sig;
4637 }
4639 // for diagnostic
4640 int os::Linux::sigflags[MAXSIGNUM];
4642 int os::Linux::get_our_sigflags(int sig) {
4643 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4644 return sigflags[sig];
4645 }
4647 void os::Linux::set_our_sigflags(int sig, int flags) {
4648 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4649 sigflags[sig] = flags;
4650 }
4652 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4653 // Check for overwrite.
4654 struct sigaction oldAct;
4655 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4657 void* oldhand = oldAct.sa_sigaction
4658 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4659 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4660 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4661 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4662 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4663 if (AllowUserSignalHandlers || !set_installed) {
4664 // Do not overwrite; user takes responsibility to forward to us.
4665 return;
4666 } else if (UseSignalChaining) {
4667 // save the old handler in jvm
4668 save_preinstalled_handler(sig, oldAct);
4669 // libjsig also interposes the sigaction() call below and saves the
4670 // old sigaction on it own.
4671 } else {
4672 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4673 "%#lx for signal %d.", (long)oldhand, sig));
4674 }
4675 }
4677 struct sigaction sigAct;
4678 sigfillset(&(sigAct.sa_mask));
4679 sigAct.sa_handler = SIG_DFL;
4680 if (!set_installed) {
4681 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4682 } else {
4683 sigAct.sa_sigaction = signalHandler;
4684 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4685 }
4686 // Save flags, which are set by ours
4687 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4688 sigflags[sig] = sigAct.sa_flags;
4690 int ret = sigaction(sig, &sigAct, &oldAct);
4691 assert(ret == 0, "check");
4693 void* oldhand2 = oldAct.sa_sigaction
4694 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4695 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4696 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4697 }
4699 // install signal handlers for signals that HotSpot needs to
4700 // handle in order to support Java-level exception handling.
4702 void os::Linux::install_signal_handlers() {
4703 if (!signal_handlers_are_installed) {
4704 signal_handlers_are_installed = true;
4706 // signal-chaining
4707 typedef void (*signal_setting_t)();
4708 signal_setting_t begin_signal_setting = NULL;
4709 signal_setting_t end_signal_setting = NULL;
4710 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4711 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4712 if (begin_signal_setting != NULL) {
4713 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4714 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4715 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4716 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4717 libjsig_is_loaded = true;
4718 assert(UseSignalChaining, "should enable signal-chaining");
4719 }
4720 if (libjsig_is_loaded) {
4721 // Tell libjsig jvm is setting signal handlers
4722 (*begin_signal_setting)();
4723 }
4725 set_signal_handler(SIGSEGV, true);
4726 set_signal_handler(SIGPIPE, true);
4727 set_signal_handler(SIGBUS, true);
4728 set_signal_handler(SIGILL, true);
4729 set_signal_handler(SIGFPE, true);
4730 #if defined(PPC64)
4731 set_signal_handler(SIGTRAP, true);
4732 #endif
4733 set_signal_handler(SIGXFSZ, true);
4735 if (libjsig_is_loaded) {
4736 // Tell libjsig jvm finishes setting signal handlers
4737 (*end_signal_setting)();
4738 }
4740 // We don't activate signal checker if libjsig is in place, we trust ourselves
4741 // and if UserSignalHandler is installed all bets are off.
4742 // Log that signal checking is off only if -verbose:jni is specified.
4743 if (CheckJNICalls) {
4744 if (libjsig_is_loaded) {
4745 if (PrintJNIResolving) {
4746 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4747 }
4748 check_signals = false;
4749 }
4750 if (AllowUserSignalHandlers) {
4751 if (PrintJNIResolving) {
4752 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4753 }
4754 check_signals = false;
4755 }
4756 }
4757 }
4758 }
4760 // This is the fastest way to get thread cpu time on Linux.
4761 // Returns cpu time (user+sys) for any thread, not only for current.
4762 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4763 // It might work on 2.6.10+ with a special kernel/glibc patch.
4764 // For reference, please, see IEEE Std 1003.1-2004:
4765 // http://www.unix.org/single_unix_specification
4767 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4768 struct timespec tp;
4769 int rc = os::Linux::clock_gettime(clockid, &tp);
4770 assert(rc == 0, "clock_gettime is expected to return 0 code");
4772 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4773 }
4775 /////
4776 // glibc on Linux platform uses non-documented flag
4777 // to indicate, that some special sort of signal
4778 // trampoline is used.
4779 // We will never set this flag, and we should
4780 // ignore this flag in our diagnostic
4781 #ifdef SIGNIFICANT_SIGNAL_MASK
4782 #undef SIGNIFICANT_SIGNAL_MASK
4783 #endif
4784 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4786 static const char* get_signal_handler_name(address handler,
4787 char* buf, int buflen) {
4788 int offset = 0;
4789 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4790 if (found) {
4791 // skip directory names
4792 const char *p1, *p2;
4793 p1 = buf;
4794 size_t len = strlen(os::file_separator());
4795 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4796 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4797 } else {
4798 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4799 }
4800 return buf;
4801 }
4803 static void print_signal_handler(outputStream* st, int sig,
4804 char* buf, size_t buflen) {
4805 struct sigaction sa;
4807 sigaction(sig, NULL, &sa);
4809 // See comment for SIGNIFICANT_SIGNAL_MASK define
4810 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4812 st->print("%s: ", os::exception_name(sig, buf, buflen));
4814 address handler = (sa.sa_flags & SA_SIGINFO)
4815 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4816 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4818 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4819 st->print("SIG_DFL");
4820 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4821 st->print("SIG_IGN");
4822 } else {
4823 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4824 }
4826 st->print(", sa_mask[0]=");
4827 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4829 address rh = VMError::get_resetted_sighandler(sig);
4830 // May be, handler was resetted by VMError?
4831 if(rh != NULL) {
4832 handler = rh;
4833 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4834 }
4836 st->print(", sa_flags=");
4837 os::Posix::print_sa_flags(st, sa.sa_flags);
4839 // Check: is it our handler?
4840 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4841 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4842 // It is our signal handler
4843 // check for flags, reset system-used one!
4844 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4845 st->print(
4846 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4847 os::Linux::get_our_sigflags(sig));
4848 }
4849 }
4850 st->cr();
4851 }
4854 #define DO_SIGNAL_CHECK(sig) \
4855 if (!sigismember(&check_signal_done, sig)) \
4856 os::Linux::check_signal_handler(sig)
4858 // This method is a periodic task to check for misbehaving JNI applications
4859 // under CheckJNI, we can add any periodic checks here
4861 void os::run_periodic_checks() {
4863 if (check_signals == false) return;
4865 // SEGV and BUS if overridden could potentially prevent
4866 // generation of hs*.log in the event of a crash, debugging
4867 // such a case can be very challenging, so we absolutely
4868 // check the following for a good measure:
4869 DO_SIGNAL_CHECK(SIGSEGV);
4870 DO_SIGNAL_CHECK(SIGILL);
4871 DO_SIGNAL_CHECK(SIGFPE);
4872 DO_SIGNAL_CHECK(SIGBUS);
4873 DO_SIGNAL_CHECK(SIGPIPE);
4874 DO_SIGNAL_CHECK(SIGXFSZ);
4875 #if defined(PPC64)
4876 DO_SIGNAL_CHECK(SIGTRAP);
4877 #endif
4879 // ReduceSignalUsage allows the user to override these handlers
4880 // see comments at the very top and jvm_solaris.h
4881 if (!ReduceSignalUsage) {
4882 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4883 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4884 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4885 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4886 }
4888 DO_SIGNAL_CHECK(SR_signum);
4889 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4890 }
4892 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4894 static os_sigaction_t os_sigaction = NULL;
4896 void os::Linux::check_signal_handler(int sig) {
4897 char buf[O_BUFLEN];
4898 address jvmHandler = NULL;
4901 struct sigaction act;
4902 if (os_sigaction == NULL) {
4903 // only trust the default sigaction, in case it has been interposed
4904 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4905 if (os_sigaction == NULL) return;
4906 }
4908 os_sigaction(sig, (struct sigaction*)NULL, &act);
4911 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4913 address thisHandler = (act.sa_flags & SA_SIGINFO)
4914 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4915 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4918 switch(sig) {
4919 case SIGSEGV:
4920 case SIGBUS:
4921 case SIGFPE:
4922 case SIGPIPE:
4923 case SIGILL:
4924 case SIGXFSZ:
4925 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4926 break;
4928 case SHUTDOWN1_SIGNAL:
4929 case SHUTDOWN2_SIGNAL:
4930 case SHUTDOWN3_SIGNAL:
4931 case BREAK_SIGNAL:
4932 jvmHandler = (address)user_handler();
4933 break;
4935 case INTERRUPT_SIGNAL:
4936 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4937 break;
4939 default:
4940 if (sig == SR_signum) {
4941 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4942 } else {
4943 return;
4944 }
4945 break;
4946 }
4948 if (thisHandler != jvmHandler) {
4949 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4950 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4951 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4952 // No need to check this sig any longer
4953 sigaddset(&check_signal_done, sig);
4954 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4955 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4956 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4957 exception_name(sig, buf, O_BUFLEN));
4958 }
4959 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4960 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4961 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4962 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4963 // No need to check this sig any longer
4964 sigaddset(&check_signal_done, sig);
4965 }
4967 // Dump all the signal
4968 if (sigismember(&check_signal_done, sig)) {
4969 print_signal_handlers(tty, buf, O_BUFLEN);
4970 }
4971 }
4973 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4975 extern bool signal_name(int signo, char* buf, size_t len);
4977 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4978 if (0 < exception_code && exception_code <= SIGRTMAX) {
4979 // signal
4980 if (!signal_name(exception_code, buf, size)) {
4981 jio_snprintf(buf, size, "SIG%d", exception_code);
4982 }
4983 return buf;
4984 } else {
4985 return NULL;
4986 }
4987 }
4989 // this is called _before_ most of the global arguments have been parsed
4990 void os::init(void) {
4991 char dummy; /* used to get a guess on initial stack address */
4993 // With LinuxThreads the JavaMain thread pid (primordial thread)
4994 // is different than the pid of the java launcher thread.
4995 // So, on Linux, the launcher thread pid is passed to the VM
4996 // via the sun.java.launcher.pid property.
4997 // Use this property instead of getpid() if it was correctly passed.
4998 // See bug 6351349.
4999 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
5001 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
5003 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5005 init_random(1234567);
5007 ThreadCritical::initialize();
5009 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5010 if (Linux::page_size() == -1) {
5011 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
5012 strerror(errno)));
5013 }
5014 init_page_sizes((size_t) Linux::page_size());
5016 Linux::initialize_system_info();
5018 // _main_thread points to the thread that created/loaded the JVM.
5019 Linux::_main_thread = pthread_self();
5021 Linux::clock_init();
5022 initial_time_count = javaTimeNanos();
5024 // pthread_condattr initialization for monotonic clock
5025 int status;
5026 pthread_condattr_t* _condattr = os::Linux::condAttr();
5027 if ((status = pthread_condattr_init(_condattr)) != 0) {
5028 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
5029 }
5030 // Only set the clock if CLOCK_MONOTONIC is available
5031 if (Linux::supports_monotonic_clock()) {
5032 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
5033 if (status == EINVAL) {
5034 warning("Unable to use monotonic clock with relative timed-waits" \
5035 " - changes to the time-of-day clock may have adverse affects");
5036 } else {
5037 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
5038 }
5039 }
5040 }
5041 // else it defaults to CLOCK_REALTIME
5043 pthread_mutex_init(&dl_mutex, NULL);
5045 // If the pagesize of the VM is greater than 8K determine the appropriate
5046 // number of initial guard pages. The user can change this with the
5047 // command line arguments, if needed.
5048 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
5049 StackYellowPages = 1;
5050 StackRedPages = 1;
5051 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
5052 }
5053 }
5055 // To install functions for atexit system call
5056 extern "C" {
5057 static void perfMemory_exit_helper() {
5058 perfMemory_exit();
5059 }
5060 }
5062 void os::pd_init_container_support() {
5063 OSContainer::init();
5064 }
5066 // this is called _after_ the global arguments have been parsed
5067 jint os::init_2(void)
5068 {
5069 Linux::fast_thread_clock_init();
5071 // Allocate a single page and mark it as readable for safepoint polling
5072 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5073 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
5075 os::set_polling_page( polling_page );
5077 #ifndef PRODUCT
5078 if(Verbose && PrintMiscellaneous)
5079 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5080 #endif
5082 if (!UseMembar) {
5083 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5084 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
5085 os::set_memory_serialize_page( mem_serialize_page );
5087 #ifndef PRODUCT
5088 if(Verbose && PrintMiscellaneous)
5089 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5090 #endif
5091 }
5093 // initialize suspend/resume support - must do this before signal_sets_init()
5094 if (SR_initialize() != 0) {
5095 perror("SR_initialize failed");
5096 return JNI_ERR;
5097 }
5099 Linux::signal_sets_init();
5100 Linux::install_signal_handlers();
5102 // Check minimum allowable stack size for thread creation and to initialize
5103 // the java system classes, including StackOverflowError - depends on page
5104 // size. Add a page for compiler2 recursion in main thread.
5105 // Add in 2*BytesPerWord times page size to account for VM stack during
5106 // class initialization depending on 32 or 64 bit VM.
5107 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
5108 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
5109 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
5111 size_t threadStackSizeInBytes = ThreadStackSize * K;
5112 if (threadStackSizeInBytes != 0 &&
5113 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
5114 tty->print_cr("\nThe stack size specified is too small, "
5115 "Specify at least %dk",
5116 os::Linux::min_stack_allowed/ K);
5117 return JNI_ERR;
5118 }
5120 // Make the stack size a multiple of the page size so that
5121 // the yellow/red zones can be guarded.
5122 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5123 vm_page_size()));
5125 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5127 #if defined(IA32)
5128 workaround_expand_exec_shield_cs_limit();
5129 #endif
5131 Linux::libpthread_init();
5132 if (PrintMiscellaneous && (Verbose || WizardMode)) {
5133 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
5134 Linux::glibc_version(), Linux::libpthread_version(),
5135 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
5136 }
5138 if (UseNUMA) {
5139 if (!Linux::libnuma_init()) {
5140 UseNUMA = false;
5141 } else {
5142 if ((Linux::numa_max_node() < 1)) {
5143 // There's only one node(they start from 0), disable NUMA.
5144 UseNUMA = false;
5145 }
5146 }
5147 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5148 // we can make the adaptive lgrp chunk resizing work. If the user specified
5149 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
5150 // disable adaptive resizing.
5151 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5152 if (FLAG_IS_DEFAULT(UseNUMA)) {
5153 UseNUMA = false;
5154 } else {
5155 if (FLAG_IS_DEFAULT(UseLargePages) &&
5156 FLAG_IS_DEFAULT(UseSHM) &&
5157 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
5158 UseLargePages = false;
5159 } else {
5160 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
5161 UseAdaptiveSizePolicy = false;
5162 UseAdaptiveNUMAChunkSizing = false;
5163 }
5164 }
5165 }
5166 if (!UseNUMA && ForceNUMA) {
5167 UseNUMA = true;
5168 }
5169 }
5171 if (MaxFDLimit) {
5172 // set the number of file descriptors to max. print out error
5173 // if getrlimit/setrlimit fails but continue regardless.
5174 struct rlimit nbr_files;
5175 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5176 if (status != 0) {
5177 if (PrintMiscellaneous && (Verbose || WizardMode))
5178 perror("os::init_2 getrlimit failed");
5179 } else {
5180 nbr_files.rlim_cur = nbr_files.rlim_max;
5181 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5182 if (status != 0) {
5183 if (PrintMiscellaneous && (Verbose || WizardMode))
5184 perror("os::init_2 setrlimit failed");
5185 }
5186 }
5187 }
5189 // Initialize lock used to serialize thread creation (see os::create_thread)
5190 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5192 // at-exit methods are called in the reverse order of their registration.
5193 // atexit functions are called on return from main or as a result of a
5194 // call to exit(3C). There can be only 32 of these functions registered
5195 // and atexit() does not set errno.
5197 if (PerfAllowAtExitRegistration) {
5198 // only register atexit functions if PerfAllowAtExitRegistration is set.
5199 // atexit functions can be delayed until process exit time, which
5200 // can be problematic for embedded VM situations. Embedded VMs should
5201 // call DestroyJavaVM() to assure that VM resources are released.
5203 // note: perfMemory_exit_helper atexit function may be removed in
5204 // the future if the appropriate cleanup code can be added to the
5205 // VM_Exit VMOperation's doit method.
5206 if (atexit(perfMemory_exit_helper) != 0) {
5207 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5208 }
5209 }
5211 // initialize thread priority policy
5212 prio_init();
5214 return JNI_OK;
5215 }
5217 // Mark the polling page as unreadable
5218 void os::make_polling_page_unreadable(void) {
5219 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5220 fatal("Could not disable polling page");
5221 };
5223 // Mark the polling page as readable
5224 void os::make_polling_page_readable(void) {
5225 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5226 fatal("Could not enable polling page");
5227 }
5228 };
5230 static int os_cpu_count(const cpu_set_t* cpus) {
5231 int count = 0;
5232 // only look up to the number of configured processors
5233 for (int i = 0; i < os::processor_count(); i++) {
5234 if (CPU_ISSET(i, cpus)) {
5235 count++;
5236 }
5237 }
5238 return count;
5239 }
5241 // Get the current number of available processors for this process.
5242 // This value can change at any time during a process's lifetime.
5243 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5244 // If anything goes wrong we fallback to returning the number of online
5245 // processors - which can be greater than the number available to the process.
5246 int os::Linux::active_processor_count() {
5247 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5248 int cpus_size = sizeof(cpu_set_t);
5249 int cpu_count = 0;
5251 // pid 0 means the current thread - which we have to assume represents the process
5252 if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
5253 cpu_count = os_cpu_count(&cpus);
5254 if (PrintActiveCpus) {
5255 tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5256 }
5257 }
5258 else {
5259 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5260 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5261 "which may exceed available processors", strerror(errno), cpu_count);
5262 }
5264 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5265 return cpu_count;
5266 }
5268 // Determine the active processor count from one of
5269 // three different sources:
5270 //
5271 // 1. User option -XX:ActiveProcessorCount
5272 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5273 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5274 //
5275 // Option 1, if specified, will always override.
5276 // If the cgroup subsystem is active and configured, we
5277 // will return the min of the cgroup and option 2 results.
5278 // This is required since tools, such as numactl, that
5279 // alter cpu affinity do not update cgroup subsystem
5280 // cpuset configuration files.
5281 int os::active_processor_count() {
5282 // User has overridden the number of active processors
5283 if (ActiveProcessorCount > 0) {
5284 if (PrintActiveCpus) {
5285 tty->print_cr("active_processor_count: "
5286 "active processor count set by user : %d",
5287 ActiveProcessorCount);
5288 }
5289 return ActiveProcessorCount;
5290 }
5292 int active_cpus;
5293 if (OSContainer::is_containerized()) {
5294 active_cpus = OSContainer::active_processor_count();
5295 if (PrintActiveCpus) {
5296 tty->print_cr("active_processor_count: determined by OSContainer: %d",
5297 active_cpus);
5298 }
5299 } else {
5300 active_cpus = os::Linux::active_processor_count();
5301 }
5303 return active_cpus;
5304 }
5306 void os::set_native_thread_name(const char *name) {
5307 // Not yet implemented.
5308 return;
5309 }
5311 bool os::distribute_processes(uint length, uint* distribution) {
5312 // Not yet implemented.
5313 return false;
5314 }
5316 bool os::bind_to_processor(uint processor_id) {
5317 // Not yet implemented.
5318 return false;
5319 }
5321 ///
5323 void os::SuspendedThreadTask::internal_do_task() {
5324 if (do_suspend(_thread->osthread())) {
5325 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5326 do_task(context);
5327 do_resume(_thread->osthread());
5328 }
5329 }
5331 class PcFetcher : public os::SuspendedThreadTask {
5332 public:
5333 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5334 ExtendedPC result();
5335 protected:
5336 void do_task(const os::SuspendedThreadTaskContext& context);
5337 private:
5338 ExtendedPC _epc;
5339 };
5341 ExtendedPC PcFetcher::result() {
5342 guarantee(is_done(), "task is not done yet.");
5343 return _epc;
5344 }
5346 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5347 Thread* thread = context.thread();
5348 OSThread* osthread = thread->osthread();
5349 if (osthread->ucontext() != NULL) {
5350 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5351 } else {
5352 // NULL context is unexpected, double-check this is the VMThread
5353 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5354 }
5355 }
5357 // Suspends the target using the signal mechanism and then grabs the PC before
5358 // resuming the target. Used by the flat-profiler only
5359 ExtendedPC os::get_thread_pc(Thread* thread) {
5360 // Make sure that it is called by the watcher for the VMThread
5361 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5362 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5364 PcFetcher fetcher(thread);
5365 fetcher.run();
5366 return fetcher.result();
5367 }
5369 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5370 {
5371 if (is_NPTL()) {
5372 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5373 } else {
5374 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5375 // word back to default 64bit precision if condvar is signaled. Java
5376 // wants 53bit precision. Save and restore current value.
5377 int fpu = get_fpu_control_word();
5378 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5379 set_fpu_control_word(fpu);
5380 return status;
5381 }
5382 }
5384 ////////////////////////////////////////////////////////////////////////////////
5385 // debug support
5387 bool os::find(address addr, outputStream* st) {
5388 Dl_info dlinfo;
5389 memset(&dlinfo, 0, sizeof(dlinfo));
5390 if (dladdr(addr, &dlinfo) != 0) {
5391 st->print(PTR_FORMAT ": ", addr);
5392 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5393 st->print("%s+%#x", dlinfo.dli_sname,
5394 addr - (intptr_t)dlinfo.dli_saddr);
5395 } else if (dlinfo.dli_fbase != NULL) {
5396 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5397 } else {
5398 st->print("<absolute address>");
5399 }
5400 if (dlinfo.dli_fname != NULL) {
5401 st->print(" in %s", dlinfo.dli_fname);
5402 }
5403 if (dlinfo.dli_fbase != NULL) {
5404 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5405 }
5406 st->cr();
5408 if (Verbose) {
5409 // decode some bytes around the PC
5410 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5411 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5412 address lowest = (address) dlinfo.dli_sname;
5413 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5414 if (begin < lowest) begin = lowest;
5415 Dl_info dlinfo2;
5416 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5417 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5418 end = (address) dlinfo2.dli_saddr;
5419 Disassembler::decode(begin, end, st);
5420 }
5421 return true;
5422 }
5423 return false;
5424 }
5426 ////////////////////////////////////////////////////////////////////////////////
5427 // misc
5429 // This does not do anything on Linux. This is basically a hook for being
5430 // able to use structured exception handling (thread-local exception filters)
5431 // on, e.g., Win32.
5432 void
5433 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5434 JavaCallArguments* args, Thread* thread) {
5435 f(value, method, args, thread);
5436 }
5438 void os::print_statistics() {
5439 }
5441 int os::message_box(const char* title, const char* message) {
5442 int i;
5443 fdStream err(defaultStream::error_fd());
5444 for (i = 0; i < 78; i++) err.print_raw("=");
5445 err.cr();
5446 err.print_raw_cr(title);
5447 for (i = 0; i < 78; i++) err.print_raw("-");
5448 err.cr();
5449 err.print_raw_cr(message);
5450 for (i = 0; i < 78; i++) err.print_raw("=");
5451 err.cr();
5453 char buf[16];
5454 // Prevent process from exiting upon "read error" without consuming all CPU
5455 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5457 return buf[0] == 'y' || buf[0] == 'Y';
5458 }
5460 int os::stat(const char *path, struct stat *sbuf) {
5461 char pathbuf[MAX_PATH];
5462 if (strlen(path) > MAX_PATH - 1) {
5463 errno = ENAMETOOLONG;
5464 return -1;
5465 }
5466 os::native_path(strcpy(pathbuf, path));
5467 return ::stat(pathbuf, sbuf);
5468 }
5470 bool os::check_heap(bool force) {
5471 return true;
5472 }
5474 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5475 return ::vsnprintf(buf, count, format, args);
5476 }
5478 // Is a (classpath) directory empty?
5479 bool os::dir_is_empty(const char* path) {
5480 DIR *dir = NULL;
5481 struct dirent *ptr;
5483 dir = opendir(path);
5484 if (dir == NULL) return true;
5486 /* Scan the directory */
5487 bool result = true;
5488 char buf[sizeof(struct dirent) + MAX_PATH];
5489 while (result && (ptr = ::readdir(dir)) != NULL) {
5490 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5491 result = false;
5492 }
5493 }
5494 closedir(dir);
5495 return result;
5496 }
5498 // This code originates from JDK's sysOpen and open64_w
5499 // from src/solaris/hpi/src/system_md.c
5501 #ifndef O_DELETE
5502 #define O_DELETE 0x10000
5503 #endif
5505 // Open a file. Unlink the file immediately after open returns
5506 // if the specified oflag has the O_DELETE flag set.
5507 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5509 int os::open(const char *path, int oflag, int mode) {
5511 if (strlen(path) > MAX_PATH - 1) {
5512 errno = ENAMETOOLONG;
5513 return -1;
5514 }
5515 int fd;
5516 int o_delete = (oflag & O_DELETE);
5517 oflag = oflag & ~O_DELETE;
5519 fd = ::open64(path, oflag, mode);
5520 if (fd == -1) return -1;
5522 //If the open succeeded, the file might still be a directory
5523 {
5524 struct stat64 buf64;
5525 int ret = ::fstat64(fd, &buf64);
5526 int st_mode = buf64.st_mode;
5528 if (ret != -1) {
5529 if ((st_mode & S_IFMT) == S_IFDIR) {
5530 errno = EISDIR;
5531 ::close(fd);
5532 return -1;
5533 }
5534 } else {
5535 ::close(fd);
5536 return -1;
5537 }
5538 }
5540 /*
5541 * All file descriptors that are opened in the JVM and not
5542 * specifically destined for a subprocess should have the
5543 * close-on-exec flag set. If we don't set it, then careless 3rd
5544 * party native code might fork and exec without closing all
5545 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5546 * UNIXProcess.c), and this in turn might:
5547 *
5548 * - cause end-of-file to fail to be detected on some file
5549 * descriptors, resulting in mysterious hangs, or
5550 *
5551 * - might cause an fopen in the subprocess to fail on a system
5552 * suffering from bug 1085341.
5553 *
5554 * (Yes, the default setting of the close-on-exec flag is a Unix
5555 * design flaw)
5556 *
5557 * See:
5558 * 1085341: 32-bit stdio routines should support file descriptors >255
5559 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5560 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5561 */
5562 #ifdef FD_CLOEXEC
5563 {
5564 int flags = ::fcntl(fd, F_GETFD);
5565 if (flags != -1)
5566 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5567 }
5568 #endif
5570 if (o_delete != 0) {
5571 ::unlink(path);
5572 }
5573 return fd;
5574 }
5577 // create binary file, rewriting existing file if required
5578 int os::create_binary_file(const char* path, bool rewrite_existing) {
5579 int oflags = O_WRONLY | O_CREAT;
5580 if (!rewrite_existing) {
5581 oflags |= O_EXCL;
5582 }
5583 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5584 }
5586 // return current position of file pointer
5587 jlong os::current_file_offset(int fd) {
5588 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5589 }
5591 // move file pointer to the specified offset
5592 jlong os::seek_to_file_offset(int fd, jlong offset) {
5593 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5594 }
5596 // This code originates from JDK's sysAvailable
5597 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5599 int os::available(int fd, jlong *bytes) {
5600 jlong cur, end;
5601 int mode;
5602 struct stat64 buf64;
5604 if (::fstat64(fd, &buf64) >= 0) {
5605 mode = buf64.st_mode;
5606 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5607 /*
5608 * XXX: is the following call interruptible? If so, this might
5609 * need to go through the INTERRUPT_IO() wrapper as for other
5610 * blocking, interruptible calls in this file.
5611 */
5612 int n;
5613 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5614 *bytes = n;
5615 return 1;
5616 }
5617 }
5618 }
5619 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5620 return 0;
5621 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5622 return 0;
5623 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5624 return 0;
5625 }
5626 *bytes = end - cur;
5627 return 1;
5628 }
5630 int os::socket_available(int fd, jint *pbytes) {
5631 // Linux doc says EINTR not returned, unlike Solaris
5632 int ret = ::ioctl(fd, FIONREAD, pbytes);
5634 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5635 // is expected to return 0 on failure and 1 on success to the jdk.
5636 return (ret < 0) ? 0 : 1;
5637 }
5639 // Map a block of memory.
5640 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5641 char *addr, size_t bytes, bool read_only,
5642 bool allow_exec) {
5643 int prot;
5644 int flags = MAP_PRIVATE;
5646 if (read_only) {
5647 prot = PROT_READ;
5648 } else {
5649 prot = PROT_READ | PROT_WRITE;
5650 }
5652 if (allow_exec) {
5653 prot |= PROT_EXEC;
5654 }
5656 if (addr != NULL) {
5657 flags |= MAP_FIXED;
5658 }
5660 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5661 fd, file_offset);
5662 if (mapped_address == MAP_FAILED) {
5663 return NULL;
5664 }
5665 return mapped_address;
5666 }
5669 // Remap a block of memory.
5670 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5671 char *addr, size_t bytes, bool read_only,
5672 bool allow_exec) {
5673 // same as map_memory() on this OS
5674 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5675 allow_exec);
5676 }
5679 // Unmap a block of memory.
5680 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5681 return munmap(addr, bytes) == 0;
5682 }
5684 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5686 static clockid_t thread_cpu_clockid(Thread* thread) {
5687 pthread_t tid = thread->osthread()->pthread_id();
5688 clockid_t clockid;
5690 // Get thread clockid
5691 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5692 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5693 return clockid;
5694 }
5696 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5697 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5698 // of a thread.
5699 //
5700 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5701 // the fast estimate available on the platform.
5703 jlong os::current_thread_cpu_time() {
5704 if (os::Linux::supports_fast_thread_cpu_time()) {
5705 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5706 } else {
5707 // return user + sys since the cost is the same
5708 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5709 }
5710 }
5712 jlong os::thread_cpu_time(Thread* thread) {
5713 // consistent with what current_thread_cpu_time() returns
5714 if (os::Linux::supports_fast_thread_cpu_time()) {
5715 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5716 } else {
5717 return slow_thread_cpu_time(thread, true /* user + sys */);
5718 }
5719 }
5721 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5722 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5723 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5724 } else {
5725 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5726 }
5727 }
5729 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5730 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5731 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5732 } else {
5733 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5734 }
5735 }
5737 //
5738 // -1 on error.
5739 //
5741 PRAGMA_DIAG_PUSH
5742 PRAGMA_FORMAT_NONLITERAL_IGNORED
5743 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5744 static bool proc_task_unchecked = true;
5745 static const char *proc_stat_path = "/proc/%d/stat";
5746 pid_t tid = thread->osthread()->thread_id();
5747 char *s;
5748 char stat[2048];
5749 int statlen;
5750 char proc_name[64];
5751 int count;
5752 long sys_time, user_time;
5753 char cdummy;
5754 int idummy;
5755 long ldummy;
5756 FILE *fp;
5758 // The /proc/<tid>/stat aggregates per-process usage on
5759 // new Linux kernels 2.6+ where NPTL is supported.
5760 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5761 // See bug 6328462.
5762 // There possibly can be cases where there is no directory
5763 // /proc/self/task, so we check its availability.
5764 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5765 // This is executed only once
5766 proc_task_unchecked = false;
5767 fp = fopen("/proc/self/task", "r");
5768 if (fp != NULL) {
5769 proc_stat_path = "/proc/self/task/%d/stat";
5770 fclose(fp);
5771 }
5772 }
5774 sprintf(proc_name, proc_stat_path, tid);
5775 fp = fopen(proc_name, "r");
5776 if ( fp == NULL ) return -1;
5777 statlen = fread(stat, 1, 2047, fp);
5778 stat[statlen] = '\0';
5779 fclose(fp);
5781 // Skip pid and the command string. Note that we could be dealing with
5782 // weird command names, e.g. user could decide to rename java launcher
5783 // to "java 1.4.2 :)", then the stat file would look like
5784 // 1234 (java 1.4.2 :)) R ... ...
5785 // We don't really need to know the command string, just find the last
5786 // occurrence of ")" and then start parsing from there. See bug 4726580.
5787 s = strrchr(stat, ')');
5788 if (s == NULL ) return -1;
5790 // Skip blank chars
5791 do s++; while (isspace(*s));
5793 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5794 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5795 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5796 &user_time, &sys_time);
5797 if ( count != 13 ) return -1;
5798 if (user_sys_cpu_time) {
5799 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5800 } else {
5801 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5802 }
5803 }
5804 PRAGMA_DIAG_POP
5806 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5807 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5808 info_ptr->may_skip_backward = false; // elapsed time not wall time
5809 info_ptr->may_skip_forward = false; // elapsed time not wall time
5810 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5811 }
5813 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5814 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5815 info_ptr->may_skip_backward = false; // elapsed time not wall time
5816 info_ptr->may_skip_forward = false; // elapsed time not wall time
5817 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5818 }
5820 bool os::is_thread_cpu_time_supported() {
5821 return true;
5822 }
5824 // System loadavg support. Returns -1 if load average cannot be obtained.
5825 // Linux doesn't yet have a (official) notion of processor sets,
5826 // so just return the system wide load average.
5827 int os::loadavg(double loadavg[], int nelem) {
5828 return ::getloadavg(loadavg, nelem);
5829 }
5831 void os::pause() {
5832 char filename[MAX_PATH];
5833 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5834 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5835 } else {
5836 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5837 }
5839 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5840 if (fd != -1) {
5841 struct stat buf;
5842 ::close(fd);
5843 while (::stat(filename, &buf) == 0) {
5844 (void)::poll(NULL, 0, 100);
5845 }
5846 } else {
5847 jio_fprintf(stderr,
5848 "Could not open pause file '%s', continuing immediately.\n", filename);
5849 }
5850 }
5853 // Refer to the comments in os_solaris.cpp park-unpark.
5854 //
5855 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5856 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5857 // For specifics regarding the bug see GLIBC BUGID 261237 :
5858 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5859 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5860 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5861 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5862 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5863 // and monitorenter when we're using 1-0 locking. All those operations may result in
5864 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5865 // of libpthread avoids the problem, but isn't practical.
5866 //
5867 // Possible remedies:
5868 //
5869 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5870 // This is palliative and probabilistic, however. If the thread is preempted
5871 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5872 // than the minimum period may have passed, and the abstime may be stale (in the
5873 // past) resultin in a hang. Using this technique reduces the odds of a hang
5874 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5875 //
5876 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5877 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5878 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5879 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5880 // thread.
5881 //
5882 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5883 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5884 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5885 // This also works well. In fact it avoids kernel-level scalability impediments
5886 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5887 // timers in a graceful fashion.
5888 //
5889 // 4. When the abstime value is in the past it appears that control returns
5890 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5891 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5892 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5893 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5894 // It may be possible to avoid reinitialization by checking the return
5895 // value from pthread_cond_timedwait(). In addition to reinitializing the
5896 // condvar we must establish the invariant that cond_signal() is only called
5897 // within critical sections protected by the adjunct mutex. This prevents
5898 // cond_signal() from "seeing" a condvar that's in the midst of being
5899 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5900 // desirable signal-after-unlock optimization that avoids futile context switching.
5901 //
5902 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5903 // structure when a condvar is used or initialized. cond_destroy() would
5904 // release the helper structure. Our reinitialize-after-timedwait fix
5905 // put excessive stress on malloc/free and locks protecting the c-heap.
5906 //
5907 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5908 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5909 // and only enabling the work-around for vulnerable environments.
5911 // utility to compute the abstime argument to timedwait:
5912 // millis is the relative timeout time
5913 // abstime will be the absolute timeout time
5914 // TODO: replace compute_abstime() with unpackTime()
5916 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5917 if (millis < 0) millis = 0;
5919 jlong seconds = millis / 1000;
5920 millis %= 1000;
5921 if (seconds > 50000000) { // see man cond_timedwait(3T)
5922 seconds = 50000000;
5923 }
5925 if (os::Linux::supports_monotonic_clock()) {
5926 struct timespec now;
5927 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5928 assert_status(status == 0, status, "clock_gettime");
5929 abstime->tv_sec = now.tv_sec + seconds;
5930 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5931 if (nanos >= NANOSECS_PER_SEC) {
5932 abstime->tv_sec += 1;
5933 nanos -= NANOSECS_PER_SEC;
5934 }
5935 abstime->tv_nsec = nanos;
5936 } else {
5937 struct timeval now;
5938 int status = gettimeofday(&now, NULL);
5939 assert(status == 0, "gettimeofday");
5940 abstime->tv_sec = now.tv_sec + seconds;
5941 long usec = now.tv_usec + millis * 1000;
5942 if (usec >= 1000000) {
5943 abstime->tv_sec += 1;
5944 usec -= 1000000;
5945 }
5946 abstime->tv_nsec = usec * 1000;
5947 }
5948 return abstime;
5949 }
5952 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5953 // Conceptually TryPark() should be equivalent to park(0).
5955 int os::PlatformEvent::TryPark() {
5956 for (;;) {
5957 const int v = _Event ;
5958 guarantee ((v == 0) || (v == 1), "invariant") ;
5959 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5960 }
5961 }
5963 void os::PlatformEvent::park() { // AKA "down()"
5964 // Invariant: Only the thread associated with the Event/PlatformEvent
5965 // may call park().
5966 // TODO: assert that _Assoc != NULL or _Assoc == Self
5967 int v ;
5968 for (;;) {
5969 v = _Event ;
5970 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5971 }
5972 guarantee (v >= 0, "invariant") ;
5973 if (v == 0) {
5974 // Do this the hard way by blocking ...
5975 int status = pthread_mutex_lock(_mutex);
5976 assert_status(status == 0, status, "mutex_lock");
5977 guarantee (_nParked == 0, "invariant") ;
5978 ++ _nParked ;
5979 while (_Event < 0) {
5980 status = pthread_cond_wait(_cond, _mutex);
5981 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5982 // Treat this the same as if the wait was interrupted
5983 if (status == ETIME) { status = EINTR; }
5984 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5985 }
5986 -- _nParked ;
5988 _Event = 0 ;
5989 status = pthread_mutex_unlock(_mutex);
5990 assert_status(status == 0, status, "mutex_unlock");
5991 // Paranoia to ensure our locked and lock-free paths interact
5992 // correctly with each other.
5993 OrderAccess::fence();
5994 }
5995 guarantee (_Event >= 0, "invariant") ;
5996 }
5998 int os::PlatformEvent::park(jlong millis) {
5999 guarantee (_nParked == 0, "invariant") ;
6001 int v ;
6002 for (;;) {
6003 v = _Event ;
6004 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6005 }
6006 guarantee (v >= 0, "invariant") ;
6007 if (v != 0) return OS_OK ;
6009 // We do this the hard way, by blocking the thread.
6010 // Consider enforcing a minimum timeout value.
6011 struct timespec abst;
6012 compute_abstime(&abst, millis);
6014 int ret = OS_TIMEOUT;
6015 int status = pthread_mutex_lock(_mutex);
6016 assert_status(status == 0, status, "mutex_lock");
6017 guarantee (_nParked == 0, "invariant") ;
6018 ++_nParked ;
6020 // Object.wait(timo) will return because of
6021 // (a) notification
6022 // (b) timeout
6023 // (c) thread.interrupt
6024 //
6025 // Thread.interrupt and object.notify{All} both call Event::set.
6026 // That is, we treat thread.interrupt as a special case of notification.
6027 // The underlying Solaris implementation, cond_timedwait, admits
6028 // spurious/premature wakeups, but the JLS/JVM spec prevents the
6029 // JVM from making those visible to Java code. As such, we must
6030 // filter out spurious wakeups. We assume all ETIME returns are valid.
6031 //
6032 // TODO: properly differentiate simultaneous notify+interrupt.
6033 // In that case, we should propagate the notify to another waiter.
6035 while (_Event < 0) {
6036 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
6037 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6038 pthread_cond_destroy (_cond);
6039 pthread_cond_init (_cond, os::Linux::condAttr()) ;
6040 }
6041 assert_status(status == 0 || status == EINTR ||
6042 status == ETIME || status == ETIMEDOUT,
6043 status, "cond_timedwait");
6044 if (!FilterSpuriousWakeups) break ; // previous semantics
6045 if (status == ETIME || status == ETIMEDOUT) break ;
6046 // We consume and ignore EINTR and spurious wakeups.
6047 }
6048 --_nParked ;
6049 if (_Event >= 0) {
6050 ret = OS_OK;
6051 }
6052 _Event = 0 ;
6053 status = pthread_mutex_unlock(_mutex);
6054 assert_status(status == 0, status, "mutex_unlock");
6055 assert (_nParked == 0, "invariant") ;
6056 // Paranoia to ensure our locked and lock-free paths interact
6057 // correctly with each other.
6058 OrderAccess::fence();
6059 return ret;
6060 }
6062 void os::PlatformEvent::unpark() {
6063 // Transitions for _Event:
6064 // 0 :=> 1
6065 // 1 :=> 1
6066 // -1 :=> either 0 or 1; must signal target thread
6067 // That is, we can safely transition _Event from -1 to either
6068 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
6069 // unpark() calls.
6070 // See also: "Semaphores in Plan 9" by Mullender & Cox
6071 //
6072 // Note: Forcing a transition from "-1" to "1" on an unpark() means
6073 // that it will take two back-to-back park() calls for the owning
6074 // thread to block. This has the benefit of forcing a spurious return
6075 // from the first park() call after an unpark() call which will help
6076 // shake out uses of park() and unpark() without condition variables.
6078 if (Atomic::xchg(1, &_Event) >= 0) return;
6080 // Wait for the thread associated with the event to vacate
6081 int status = pthread_mutex_lock(_mutex);
6082 assert_status(status == 0, status, "mutex_lock");
6083 int AnyWaiters = _nParked;
6084 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6085 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
6086 AnyWaiters = 0;
6087 pthread_cond_signal(_cond);
6088 }
6089 status = pthread_mutex_unlock(_mutex);
6090 assert_status(status == 0, status, "mutex_unlock");
6091 if (AnyWaiters != 0) {
6092 status = pthread_cond_signal(_cond);
6093 assert_status(status == 0, status, "cond_signal");
6094 }
6096 // Note that we signal() _after dropping the lock for "immortal" Events.
6097 // This is safe and avoids a common class of futile wakeups. In rare
6098 // circumstances this can cause a thread to return prematurely from
6099 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
6100 // simply re-test the condition and re-park itself.
6101 }
6104 // JSR166
6105 // -------------------------------------------------------
6107 /*
6108 * The solaris and linux implementations of park/unpark are fairly
6109 * conservative for now, but can be improved. They currently use a
6110 * mutex/condvar pair, plus a a count.
6111 * Park decrements count if > 0, else does a condvar wait. Unpark
6112 * sets count to 1 and signals condvar. Only one thread ever waits
6113 * on the condvar. Contention seen when trying to park implies that someone
6114 * is unparking you, so don't wait. And spurious returns are fine, so there
6115 * is no need to track notifications.
6116 */
6118 /*
6119 * This code is common to linux and solaris and will be moved to a
6120 * common place in dolphin.
6121 *
6122 * The passed in time value is either a relative time in nanoseconds
6123 * or an absolute time in milliseconds. Either way it has to be unpacked
6124 * into suitable seconds and nanoseconds components and stored in the
6125 * given timespec structure.
6126 * Given time is a 64-bit value and the time_t used in the timespec is only
6127 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6128 * overflow if times way in the future are given. Further on Solaris versions
6129 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6130 * number of seconds, in abstime, is less than current_time + 100,000,000.
6131 * As it will be 28 years before "now + 100000000" will overflow we can
6132 * ignore overflow and just impose a hard-limit on seconds using the value
6133 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6134 * years from "now".
6135 */
6137 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6138 assert (time > 0, "convertTime");
6139 time_t max_secs = 0;
6141 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
6142 struct timeval now;
6143 int status = gettimeofday(&now, NULL);
6144 assert(status == 0, "gettimeofday");
6146 max_secs = now.tv_sec + MAX_SECS;
6148 if (isAbsolute) {
6149 jlong secs = time / 1000;
6150 if (secs > max_secs) {
6151 absTime->tv_sec = max_secs;
6152 } else {
6153 absTime->tv_sec = secs;
6154 }
6155 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6156 } else {
6157 jlong secs = time / NANOSECS_PER_SEC;
6158 if (secs >= MAX_SECS) {
6159 absTime->tv_sec = max_secs;
6160 absTime->tv_nsec = 0;
6161 } else {
6162 absTime->tv_sec = now.tv_sec + secs;
6163 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6164 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6165 absTime->tv_nsec -= NANOSECS_PER_SEC;
6166 ++absTime->tv_sec; // note: this must be <= max_secs
6167 }
6168 }
6169 }
6170 } else {
6171 // must be relative using monotonic clock
6172 struct timespec now;
6173 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6174 assert_status(status == 0, status, "clock_gettime");
6175 max_secs = now.tv_sec + MAX_SECS;
6176 jlong secs = time / NANOSECS_PER_SEC;
6177 if (secs >= MAX_SECS) {
6178 absTime->tv_sec = max_secs;
6179 absTime->tv_nsec = 0;
6180 } else {
6181 absTime->tv_sec = now.tv_sec + secs;
6182 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
6183 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6184 absTime->tv_nsec -= NANOSECS_PER_SEC;
6185 ++absTime->tv_sec; // note: this must be <= max_secs
6186 }
6187 }
6188 }
6189 assert(absTime->tv_sec >= 0, "tv_sec < 0");
6190 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6191 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6192 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6193 }
6195 void Parker::park(bool isAbsolute, jlong time) {
6196 // Ideally we'd do something useful while spinning, such
6197 // as calling unpackTime().
6199 // Optional fast-path check:
6200 // Return immediately if a permit is available.
6201 // We depend on Atomic::xchg() having full barrier semantics
6202 // since we are doing a lock-free update to _counter.
6203 if (Atomic::xchg(0, &_counter) > 0) return;
6205 Thread* thread = Thread::current();
6206 assert(thread->is_Java_thread(), "Must be JavaThread");
6207 JavaThread *jt = (JavaThread *)thread;
6209 // Optional optimization -- avoid state transitions if there's an interrupt pending.
6210 // Check interrupt before trying to wait
6211 if (Thread::is_interrupted(thread, false)) {
6212 return;
6213 }
6215 // Next, demultiplex/decode time arguments
6216 timespec absTime;
6217 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6218 return;
6219 }
6220 if (time > 0) {
6221 unpackTime(&absTime, isAbsolute, time);
6222 }
6225 // Enter safepoint region
6226 // Beware of deadlocks such as 6317397.
6227 // The per-thread Parker:: mutex is a classic leaf-lock.
6228 // In particular a thread must never block on the Threads_lock while
6229 // holding the Parker:: mutex. If safepoints are pending both the
6230 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6231 ThreadBlockInVM tbivm(jt);
6233 // Don't wait if cannot get lock since interference arises from
6234 // unblocking. Also. check interrupt before trying wait
6235 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6236 return;
6237 }
6239 int status ;
6240 if (_counter > 0) { // no wait needed
6241 _counter = 0;
6242 status = pthread_mutex_unlock(_mutex);
6243 assert (status == 0, "invariant") ;
6244 // Paranoia to ensure our locked and lock-free paths interact
6245 // correctly with each other and Java-level accesses.
6246 OrderAccess::fence();
6247 return;
6248 }
6250 #ifdef ASSERT
6251 // Don't catch signals while blocked; let the running threads have the signals.
6252 // (This allows a debugger to break into the running thread.)
6253 sigset_t oldsigs;
6254 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6255 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6256 #endif
6258 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6259 jt->set_suspend_equivalent();
6260 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6262 assert(_cur_index == -1, "invariant");
6263 if (time == 0) {
6264 _cur_index = REL_INDEX; // arbitrary choice when not timed
6265 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6266 } else {
6267 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6268 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6269 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6270 pthread_cond_destroy (&_cond[_cur_index]) ;
6271 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6272 }
6273 }
6274 _cur_index = -1;
6275 assert_status(status == 0 || status == EINTR ||
6276 status == ETIME || status == ETIMEDOUT,
6277 status, "cond_timedwait");
6279 #ifdef ASSERT
6280 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6281 #endif
6283 _counter = 0 ;
6284 status = pthread_mutex_unlock(_mutex) ;
6285 assert_status(status == 0, status, "invariant") ;
6286 // Paranoia to ensure our locked and lock-free paths interact
6287 // correctly with each other and Java-level accesses.
6288 OrderAccess::fence();
6290 // If externally suspended while waiting, re-suspend
6291 if (jt->handle_special_suspend_equivalent_condition()) {
6292 jt->java_suspend_self();
6293 }
6294 }
6296 void Parker::unpark() {
6297 int s, status ;
6298 status = pthread_mutex_lock(_mutex);
6299 assert (status == 0, "invariant") ;
6300 s = _counter;
6301 _counter = 1;
6302 if (s < 1) {
6303 // thread might be parked
6304 if (_cur_index != -1) {
6305 // thread is definitely parked
6306 if (WorkAroundNPTLTimedWaitHang) {
6307 status = pthread_cond_signal (&_cond[_cur_index]);
6308 assert (status == 0, "invariant");
6309 status = pthread_mutex_unlock(_mutex);
6310 assert (status == 0, "invariant");
6311 } else {
6312 // must capture correct index before unlocking
6313 int index = _cur_index;
6314 status = pthread_mutex_unlock(_mutex);
6315 assert (status == 0, "invariant");
6316 status = pthread_cond_signal (&_cond[index]);
6317 assert (status == 0, "invariant");
6318 }
6319 } else {
6320 pthread_mutex_unlock(_mutex);
6321 assert (status == 0, "invariant") ;
6322 }
6323 } else {
6324 pthread_mutex_unlock(_mutex);
6325 assert (status == 0, "invariant") ;
6326 }
6327 }
6330 extern char** environ;
6332 // Run the specified command in a separate process. Return its exit value,
6333 // or -1 on failure (e.g. can't fork a new process).
6334 // Unlike system(), this function can be called from signal handler. It
6335 // doesn't block SIGINT et al.
6336 int os::fork_and_exec(char* cmd) {
6337 const char * argv[4] = {"sh", "-c", cmd, NULL};
6339 pid_t pid = fork();
6341 if (pid < 0) {
6342 // fork failed
6343 return -1;
6345 } else if (pid == 0) {
6346 // child process
6348 execve("/bin/sh", (char* const*)argv, environ);
6350 // execve failed
6351 _exit(-1);
6353 } else {
6354 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6355 // care about the actual exit code, for now.
6357 int status;
6359 // Wait for the child process to exit. This returns immediately if
6360 // the child has already exited. */
6361 while (waitpid(pid, &status, 0) < 0) {
6362 switch (errno) {
6363 case ECHILD: return 0;
6364 case EINTR: break;
6365 default: return -1;
6366 }
6367 }
6369 if (WIFEXITED(status)) {
6370 // The child exited normally; get its exit code.
6371 return WEXITSTATUS(status);
6372 } else if (WIFSIGNALED(status)) {
6373 // The child exited because of a signal
6374 // The best value to return is 0x80 + signal number,
6375 // because that is what all Unix shells do, and because
6376 // it allows callers to distinguish between process exit and
6377 // process death by signal.
6378 return 0x80 + WTERMSIG(status);
6379 } else {
6380 // Unknown exit code; pass it through
6381 return status;
6382 }
6383 }
6384 }
6386 // is_headless_jre()
6387 //
6388 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6389 // in order to report if we are running in a headless jre
6390 //
6391 // Since JDK8 xawt/libmawt.so was moved into the same directory
6392 // as libawt.so, and renamed libawt_xawt.so
6393 //
6394 bool os::is_headless_jre() {
6395 struct stat statbuf;
6396 char buf[MAXPATHLEN];
6397 char libmawtpath[MAXPATHLEN];
6398 const char *xawtstr = "/xawt/libmawt.so";
6399 const char *new_xawtstr = "/libawt_xawt.so";
6400 char *p;
6402 // Get path to libjvm.so
6403 os::jvm_path(buf, sizeof(buf));
6405 // Get rid of libjvm.so
6406 p = strrchr(buf, '/');
6407 if (p == NULL) return false;
6408 else *p = '\0';
6410 // Get rid of client or server
6411 p = strrchr(buf, '/');
6412 if (p == NULL) return false;
6413 else *p = '\0';
6415 // check xawt/libmawt.so
6416 strcpy(libmawtpath, buf);
6417 strcat(libmawtpath, xawtstr);
6418 if (::stat(libmawtpath, &statbuf) == 0) return false;
6420 // check libawt_xawt.so
6421 strcpy(libmawtpath, buf);
6422 strcat(libmawtpath, new_xawtstr);
6423 if (::stat(libmawtpath, &statbuf) == 0) return false;
6425 return true;
6426 }
6428 // Get the default path to the core file
6429 // Returns the length of the string
6430 int os::get_core_path(char* buffer, size_t bufferSize) {
6431 const char* p = get_current_directory(buffer, bufferSize);
6433 if (p == NULL) {
6434 assert(p != NULL, "failed to get current directory");
6435 return 0;
6436 }
6438 return strlen(buffer);
6439 }
6441 /////////////// Unit tests ///////////////
6443 #ifndef PRODUCT
6445 #define test_log(...) \
6446 do {\
6447 if (VerboseInternalVMTests) { \
6448 tty->print_cr(__VA_ARGS__); \
6449 tty->flush(); \
6450 }\
6451 } while (false)
6453 class TestReserveMemorySpecial : AllStatic {
6454 public:
6455 static void small_page_write(void* addr, size_t size) {
6456 size_t page_size = os::vm_page_size();
6458 char* end = (char*)addr + size;
6459 for (char* p = (char*)addr; p < end; p += page_size) {
6460 *p = 1;
6461 }
6462 }
6464 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6465 if (!UseHugeTLBFS) {
6466 return;
6467 }
6469 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6471 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6473 if (addr != NULL) {
6474 small_page_write(addr, size);
6476 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6477 }
6478 }
6480 static void test_reserve_memory_special_huge_tlbfs_only() {
6481 if (!UseHugeTLBFS) {
6482 return;
6483 }
6485 size_t lp = os::large_page_size();
6487 for (size_t size = lp; size <= lp * 10; size += lp) {
6488 test_reserve_memory_special_huge_tlbfs_only(size);
6489 }
6490 }
6492 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6493 size_t lp = os::large_page_size();
6494 size_t ag = os::vm_allocation_granularity();
6496 // sizes to test
6497 const size_t sizes[] = {
6498 lp, lp + ag, lp + lp / 2, lp * 2,
6499 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6500 lp * 10, lp * 10 + lp / 2
6501 };
6502 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6504 // For each size/alignment combination, we test three scenarios:
6505 // 1) with req_addr == NULL
6506 // 2) with a non-null req_addr at which we expect to successfully allocate
6507 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6508 // expect the allocation to either fail or to ignore req_addr
6510 // Pre-allocate two areas; they shall be as large as the largest allocation
6511 // and aligned to the largest alignment we will be testing.
6512 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6513 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6514 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6515 -1, 0);
6516 assert(mapping1 != MAP_FAILED, "should work");
6518 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6519 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6520 -1, 0);
6521 assert(mapping2 != MAP_FAILED, "should work");
6523 // Unmap the first mapping, but leave the second mapping intact: the first
6524 // mapping will serve as a value for a "good" req_addr (case 2). The second
6525 // mapping, still intact, as "bad" req_addr (case 3).
6526 ::munmap(mapping1, mapping_size);
6528 // Case 1
6529 test_log("%s, req_addr NULL:", __FUNCTION__);
6530 test_log("size align result");
6532 for (int i = 0; i < num_sizes; i++) {
6533 const size_t size = sizes[i];
6534 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6535 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6536 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6537 size, alignment, p, (p != NULL ? "" : "(failed)"));
6538 if (p != NULL) {
6539 assert(is_ptr_aligned(p, alignment), "must be");
6540 small_page_write(p, size);
6541 os::Linux::release_memory_special_huge_tlbfs(p, size);
6542 }
6543 }
6544 }
6546 // Case 2
6547 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6548 test_log("size align req_addr result");
6550 for (int i = 0; i < num_sizes; i++) {
6551 const size_t size = sizes[i];
6552 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6553 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6554 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6555 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6556 size, alignment, req_addr, p,
6557 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6558 if (p != NULL) {
6559 assert(p == req_addr, "must be");
6560 small_page_write(p, size);
6561 os::Linux::release_memory_special_huge_tlbfs(p, size);
6562 }
6563 }
6564 }
6566 // Case 3
6567 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6568 test_log("size align req_addr result");
6570 for (int i = 0; i < num_sizes; i++) {
6571 const size_t size = sizes[i];
6572 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6573 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6574 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6575 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6576 size, alignment, req_addr, p,
6577 ((p != NULL ? "" : "(failed)")));
6578 // as the area around req_addr contains already existing mappings, the API should always
6579 // return NULL (as per contract, it cannot return another address)
6580 assert(p == NULL, "must be");
6581 }
6582 }
6584 ::munmap(mapping2, mapping_size);
6586 }
6588 static void test_reserve_memory_special_huge_tlbfs() {
6589 if (!UseHugeTLBFS) {
6590 return;
6591 }
6593 test_reserve_memory_special_huge_tlbfs_only();
6594 test_reserve_memory_special_huge_tlbfs_mixed();
6595 }
6597 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6598 if (!UseSHM) {
6599 return;
6600 }
6602 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6604 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6606 if (addr != NULL) {
6607 assert(is_ptr_aligned(addr, alignment), "Check");
6608 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6610 small_page_write(addr, size);
6612 os::Linux::release_memory_special_shm(addr, size);
6613 }
6614 }
6616 static void test_reserve_memory_special_shm() {
6617 size_t lp = os::large_page_size();
6618 size_t ag = os::vm_allocation_granularity();
6620 for (size_t size = ag; size < lp * 3; size += ag) {
6621 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6622 test_reserve_memory_special_shm(size, alignment);
6623 }
6624 }
6625 }
6627 static void test() {
6628 test_reserve_memory_special_huge_tlbfs();
6629 test_reserve_memory_special_shm();
6630 }
6631 };
6633 void TestReserveMemorySpecial_test() {
6634 TestReserveMemorySpecial::test();
6635 }
6637 #endif