Fri, 06 Jul 2018 18:50:13 +0000
8146115: Improve docker container detection and resource configuration usage
Reviewed-by: bobv, dbuck
1 /*
2 * Copyright (c) 1999, 2016, 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::Linux::is_initial_thread()) {
1000 // If current thread is initial 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 // initial thread
1087 // Check if current thread is the initial thread, similar to Solaris thr_main.
1088 bool os::Linux::is_initial_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 (initial_thread_stack_bottom() == NULL) return false;
1093 assert(initial_thread_stack_bottom() != NULL &&
1094 initial_thread_stack_size() != 0,
1095 "os::init did not locate initial thread's stack region");
1096 if ((address)&dummy >= initial_thread_stack_bottom() &&
1097 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1098 return true;
1099 else return false;
1100 }
1102 // Find the virtual memory area that contains addr
1103 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1104 FILE *fp = fopen("/proc/self/maps", "r");
1105 if (fp) {
1106 address low, high;
1107 while (!feof(fp)) {
1108 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1109 if (low <= addr && addr < high) {
1110 if (vma_low) *vma_low = low;
1111 if (vma_high) *vma_high = high;
1112 fclose (fp);
1113 return true;
1114 }
1115 }
1116 for (;;) {
1117 int ch = fgetc(fp);
1118 if (ch == EOF || ch == (int)'\n') break;
1119 }
1120 }
1121 fclose(fp);
1122 }
1123 return false;
1124 }
1126 // Locate initial thread stack. This special handling of initial thread stack
1127 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1128 // bogus value for the primordial process thread. While the launcher has created
1129 // the VM in a new thread since JDK 6, we still have to allow for the use of the
1130 // JNI invocation API from a primordial thread.
1131 void os::Linux::capture_initial_stack(size_t max_size) {
1133 // max_size is either 0 (which means accept OS default for thread stacks) or
1134 // a user-specified value known to be at least the minimum needed. If we
1135 // are actually on the primordial thread we can make it appear that we have a
1136 // smaller max_size stack by inserting the guard pages at that location. But we
1137 // cannot do anything to emulate a larger stack than what has been provided by
1138 // the OS or threading library. In fact if we try to use a stack greater than
1139 // what is set by rlimit then we will crash the hosting process.
1141 // Maximum stack size is the easy part, get it from RLIMIT_STACK.
1142 // If this is "unlimited" then it will be a huge value.
1143 struct rlimit rlim;
1144 getrlimit(RLIMIT_STACK, &rlim);
1145 size_t stack_size = rlim.rlim_cur;
1147 // 6308388: a bug in ld.so will relocate its own .data section to the
1148 // lower end of primordial stack; reduce ulimit -s value a little bit
1149 // so we won't install guard page on ld.so's data section.
1150 stack_size -= 2 * page_size();
1152 // Try to figure out where the stack base (top) is. This is harder.
1153 //
1154 // When an application is started, glibc saves the initial stack pointer in
1155 // a global variable "__libc_stack_end", which is then used by system
1156 // libraries. __libc_stack_end should be pretty close to stack top. The
1157 // variable is available since the very early days. However, because it is
1158 // a private interface, it could disappear in the future.
1159 //
1160 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1161 // to __libc_stack_end, it is very close to stack top, but isn't the real
1162 // stack top. Note that /proc may not exist if VM is running as a chroot
1163 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1164 // /proc/<pid>/stat could change in the future (though unlikely).
1165 //
1166 // We try __libc_stack_end first. If that doesn't work, look for
1167 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1168 // as a hint, which should work well in most cases.
1170 uintptr_t stack_start;
1172 // try __libc_stack_end first
1173 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1174 if (p && *p) {
1175 stack_start = *p;
1176 } else {
1177 // see if we can get the start_stack field from /proc/self/stat
1178 FILE *fp;
1179 int pid;
1180 char state;
1181 int ppid;
1182 int pgrp;
1183 int session;
1184 int nr;
1185 int tpgrp;
1186 unsigned long flags;
1187 unsigned long minflt;
1188 unsigned long cminflt;
1189 unsigned long majflt;
1190 unsigned long cmajflt;
1191 unsigned long utime;
1192 unsigned long stime;
1193 long cutime;
1194 long cstime;
1195 long prio;
1196 long nice;
1197 long junk;
1198 long it_real;
1199 uintptr_t start;
1200 uintptr_t vsize;
1201 intptr_t rss;
1202 uintptr_t rsslim;
1203 uintptr_t scodes;
1204 uintptr_t ecode;
1205 int i;
1207 // Figure what the primordial thread stack base is. Code is inspired
1208 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1209 // followed by command name surrounded by parentheses, state, etc.
1210 char stat[2048];
1211 int statlen;
1213 fp = fopen("/proc/self/stat", "r");
1214 if (fp) {
1215 statlen = fread(stat, 1, 2047, fp);
1216 stat[statlen] = '\0';
1217 fclose(fp);
1219 // Skip pid and the command string. Note that we could be dealing with
1220 // weird command names, e.g. user could decide to rename java launcher
1221 // to "java 1.4.2 :)", then the stat file would look like
1222 // 1234 (java 1.4.2 :)) R ... ...
1223 // We don't really need to know the command string, just find the last
1224 // occurrence of ")" and then start parsing from there. See bug 4726580.
1225 char * s = strrchr(stat, ')');
1227 i = 0;
1228 if (s) {
1229 // Skip blank chars
1230 do s++; while (isspace(*s));
1232 #define _UFM UINTX_FORMAT
1233 #define _DFM INTX_FORMAT
1235 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1236 /* 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 */
1237 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,
1238 &state, /* 3 %c */
1239 &ppid, /* 4 %d */
1240 &pgrp, /* 5 %d */
1241 &session, /* 6 %d */
1242 &nr, /* 7 %d */
1243 &tpgrp, /* 8 %d */
1244 &flags, /* 9 %lu */
1245 &minflt, /* 10 %lu */
1246 &cminflt, /* 11 %lu */
1247 &majflt, /* 12 %lu */
1248 &cmajflt, /* 13 %lu */
1249 &utime, /* 14 %lu */
1250 &stime, /* 15 %lu */
1251 &cutime, /* 16 %ld */
1252 &cstime, /* 17 %ld */
1253 &prio, /* 18 %ld */
1254 &nice, /* 19 %ld */
1255 &junk, /* 20 %ld */
1256 &it_real, /* 21 %ld */
1257 &start, /* 22 UINTX_FORMAT */
1258 &vsize, /* 23 UINTX_FORMAT */
1259 &rss, /* 24 INTX_FORMAT */
1260 &rsslim, /* 25 UINTX_FORMAT */
1261 &scodes, /* 26 UINTX_FORMAT */
1262 &ecode, /* 27 UINTX_FORMAT */
1263 &stack_start); /* 28 UINTX_FORMAT */
1264 }
1266 #undef _UFM
1267 #undef _DFM
1269 if (i != 28 - 2) {
1270 assert(false, "Bad conversion from /proc/self/stat");
1271 // product mode - assume we are the initial thread, good luck in the
1272 // embedded case.
1273 warning("Can't detect initial thread stack location - bad conversion");
1274 stack_start = (uintptr_t) &rlim;
1275 }
1276 } else {
1277 // For some reason we can't open /proc/self/stat (for example, running on
1278 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1279 // most cases, so don't abort:
1280 warning("Can't detect initial thread stack location - no /proc/self/stat");
1281 stack_start = (uintptr_t) &rlim;
1282 }
1283 }
1285 // Now we have a pointer (stack_start) very close to the stack top, the
1286 // next thing to do is to figure out the exact location of stack top. We
1287 // can find out the virtual memory area that contains stack_start by
1288 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1289 // and its upper limit is the real stack top. (again, this would fail if
1290 // running inside chroot, because /proc may not exist.)
1292 uintptr_t stack_top;
1293 address low, high;
1294 if (find_vma((address)stack_start, &low, &high)) {
1295 // success, "high" is the true stack top. (ignore "low", because initial
1296 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1297 stack_top = (uintptr_t)high;
1298 } else {
1299 // failed, likely because /proc/self/maps does not exist
1300 warning("Can't detect initial thread stack location - find_vma failed");
1301 // best effort: stack_start is normally within a few pages below the real
1302 // stack top, use it as stack top, and reduce stack size so we won't put
1303 // guard page outside stack.
1304 stack_top = stack_start;
1305 stack_size -= 16 * page_size();
1306 }
1308 // stack_top could be partially down the page so align it
1309 stack_top = align_size_up(stack_top, page_size());
1311 // Allowed stack value is minimum of max_size and what we derived from rlimit
1312 if (max_size > 0) {
1313 _initial_thread_stack_size = MIN2(max_size, stack_size);
1314 } else {
1315 // Accept the rlimit max, but if stack is unlimited then it will be huge, so
1316 // clamp it at 8MB as we do on Solaris
1317 _initial_thread_stack_size = MIN2(stack_size, 8*M);
1318 }
1320 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1321 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1322 assert(_initial_thread_stack_bottom < (address)stack_top, "overflow!");
1323 }
1325 ////////////////////////////////////////////////////////////////////////////////
1326 // time support
1328 // Time since start-up in seconds to a fine granularity.
1329 // Used by VMSelfDestructTimer and the MemProfiler.
1330 double os::elapsedTime() {
1332 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1333 }
1335 jlong os::elapsed_counter() {
1336 return javaTimeNanos() - initial_time_count;
1337 }
1339 jlong os::elapsed_frequency() {
1340 return NANOSECS_PER_SEC; // nanosecond resolution
1341 }
1343 bool os::supports_vtime() { return true; }
1344 bool os::enable_vtime() { return false; }
1345 bool os::vtime_enabled() { return false; }
1347 double os::elapsedVTime() {
1348 struct rusage usage;
1349 int retval = getrusage(RUSAGE_THREAD, &usage);
1350 if (retval == 0) {
1351 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);
1352 } else {
1353 // better than nothing, but not much
1354 return elapsedTime();
1355 }
1356 }
1358 jlong os::javaTimeMillis() {
1359 timeval time;
1360 int status = gettimeofday(&time, NULL);
1361 assert(status != -1, "linux error");
1362 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1363 }
1365 #ifndef CLOCK_MONOTONIC
1366 #define CLOCK_MONOTONIC (1)
1367 #endif
1369 void os::Linux::clock_init() {
1370 // we do dlopen's in this particular order due to bug in linux
1371 // dynamical loader (see 6348968) leading to crash on exit
1372 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1373 if (handle == NULL) {
1374 handle = dlopen("librt.so", RTLD_LAZY);
1375 }
1377 if (handle) {
1378 int (*clock_getres_func)(clockid_t, struct timespec*) =
1379 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1380 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1381 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1382 if (clock_getres_func && clock_gettime_func) {
1383 // See if monotonic clock is supported by the kernel. Note that some
1384 // early implementations simply return kernel jiffies (updated every
1385 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1386 // for nano time (though the monotonic property is still nice to have).
1387 // It's fixed in newer kernels, however clock_getres() still returns
1388 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1389 // resolution for now. Hopefully as people move to new kernels, this
1390 // won't be a problem.
1391 struct timespec res;
1392 struct timespec tp;
1393 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1394 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1395 // yes, monotonic clock is supported
1396 _clock_gettime = clock_gettime_func;
1397 return;
1398 } else {
1399 // close librt if there is no monotonic clock
1400 dlclose(handle);
1401 }
1402 }
1403 }
1404 warning("No monotonic clock was available - timed services may " \
1405 "be adversely affected if the time-of-day clock changes");
1406 }
1408 #ifndef SYS_clock_getres
1410 #if defined(IA32) || defined(AMD64)
1411 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1412 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1413 #else
1414 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1415 #define sys_clock_getres(x,y) -1
1416 #endif
1418 #else
1419 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1420 #endif
1422 void os::Linux::fast_thread_clock_init() {
1423 if (!UseLinuxPosixThreadCPUClocks) {
1424 return;
1425 }
1426 clockid_t clockid;
1427 struct timespec tp;
1428 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1429 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1431 // Switch to using fast clocks for thread cpu time if
1432 // the sys_clock_getres() returns 0 error code.
1433 // Note, that some kernels may support the current thread
1434 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1435 // returned by the pthread_getcpuclockid().
1436 // If the fast Posix clocks are supported then the sys_clock_getres()
1437 // must return at least tp.tv_sec == 0 which means a resolution
1438 // better than 1 sec. This is extra check for reliability.
1440 if(pthread_getcpuclockid_func &&
1441 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1442 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1444 _supports_fast_thread_cpu_time = true;
1445 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1446 }
1447 }
1449 jlong os::javaTimeNanos() {
1450 if (Linux::supports_monotonic_clock()) {
1451 struct timespec tp;
1452 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1453 assert(status == 0, "gettime error");
1454 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1455 return result;
1456 } else {
1457 timeval time;
1458 int status = gettimeofday(&time, NULL);
1459 assert(status != -1, "linux error");
1460 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1461 return 1000 * usecs;
1462 }
1463 }
1465 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1466 if (Linux::supports_monotonic_clock()) {
1467 info_ptr->max_value = ALL_64_BITS;
1469 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1470 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1471 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1472 } else {
1473 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1474 info_ptr->max_value = ALL_64_BITS;
1476 // gettimeofday is a real time clock so it skips
1477 info_ptr->may_skip_backward = true;
1478 info_ptr->may_skip_forward = true;
1479 }
1481 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1482 }
1484 // Return the real, user, and system times in seconds from an
1485 // arbitrary fixed point in the past.
1486 bool os::getTimesSecs(double* process_real_time,
1487 double* process_user_time,
1488 double* process_system_time) {
1489 struct tms ticks;
1490 clock_t real_ticks = times(&ticks);
1492 if (real_ticks == (clock_t) (-1)) {
1493 return false;
1494 } else {
1495 double ticks_per_second = (double) clock_tics_per_sec;
1496 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1497 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1498 *process_real_time = ((double) real_ticks) / ticks_per_second;
1500 return true;
1501 }
1502 }
1505 char * os::local_time_string(char *buf, size_t buflen) {
1506 struct tm t;
1507 time_t long_time;
1508 time(&long_time);
1509 localtime_r(&long_time, &t);
1510 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1511 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1512 t.tm_hour, t.tm_min, t.tm_sec);
1513 return buf;
1514 }
1516 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1517 return localtime_r(clock, res);
1518 }
1520 ////////////////////////////////////////////////////////////////////////////////
1521 // runtime exit support
1523 // Note: os::shutdown() might be called very early during initialization, or
1524 // called from signal handler. Before adding something to os::shutdown(), make
1525 // sure it is async-safe and can handle partially initialized VM.
1526 void os::shutdown() {
1528 // allow PerfMemory to attempt cleanup of any persistent resources
1529 perfMemory_exit();
1531 // needs to remove object in file system
1532 AttachListener::abort();
1534 // flush buffered output, finish log files
1535 ostream_abort();
1537 // Check for abort hook
1538 abort_hook_t abort_hook = Arguments::abort_hook();
1539 if (abort_hook != NULL) {
1540 abort_hook();
1541 }
1543 }
1545 // Note: os::abort() might be called very early during initialization, or
1546 // called from signal handler. Before adding something to os::abort(), make
1547 // sure it is async-safe and can handle partially initialized VM.
1548 void os::abort(bool dump_core) {
1549 os::shutdown();
1550 if (dump_core) {
1551 #ifndef PRODUCT
1552 fdStream out(defaultStream::output_fd());
1553 out.print_raw("Current thread is ");
1554 char buf[16];
1555 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1556 out.print_raw_cr(buf);
1557 out.print_raw_cr("Dumping core ...");
1558 #endif
1559 ::abort(); // dump core
1560 }
1562 ::exit(1);
1563 }
1565 // Die immediately, no exit hook, no abort hook, no cleanup.
1566 void os::die() {
1567 // _exit() on LinuxThreads only kills current thread
1568 ::abort();
1569 }
1572 // This method is a copy of JDK's sysGetLastErrorString
1573 // from src/solaris/hpi/src/system_md.c
1575 size_t os::lasterror(char *buf, size_t len) {
1577 if (errno == 0) return 0;
1579 const char *s = ::strerror(errno);
1580 size_t n = ::strlen(s);
1581 if (n >= len) {
1582 n = len - 1;
1583 }
1584 ::strncpy(buf, s, n);
1585 buf[n] = '\0';
1586 return n;
1587 }
1589 intx os::current_thread_id() { return (intx)pthread_self(); }
1590 int os::current_process_id() {
1592 // Under the old linux thread library, linux gives each thread
1593 // its own process id. Because of this each thread will return
1594 // a different pid if this method were to return the result
1595 // of getpid(2). Linux provides no api that returns the pid
1596 // of the launcher thread for the vm. This implementation
1597 // returns a unique pid, the pid of the launcher thread
1598 // that starts the vm 'process'.
1600 // Under the NPTL, getpid() returns the same pid as the
1601 // launcher thread rather than a unique pid per thread.
1602 // Use gettid() if you want the old pre NPTL behaviour.
1604 // if you are looking for the result of a call to getpid() that
1605 // returns a unique pid for the calling thread, then look at the
1606 // OSThread::thread_id() method in osThread_linux.hpp file
1608 return (int)(_initial_pid ? _initial_pid : getpid());
1609 }
1611 // DLL functions
1613 const char* os::dll_file_extension() { return ".so"; }
1615 // This must be hard coded because it's the system's temporary
1616 // directory not the java application's temp directory, ala java.io.tmpdir.
1617 const char* os::get_temp_directory() { return "/tmp"; }
1619 static bool file_exists(const char* filename) {
1620 struct stat statbuf;
1621 if (filename == NULL || strlen(filename) == 0) {
1622 return false;
1623 }
1624 return os::stat(filename, &statbuf) == 0;
1625 }
1627 bool os::dll_build_name(char* buffer, size_t buflen,
1628 const char* pname, const char* fname) {
1629 bool retval = false;
1630 // Copied from libhpi
1631 const size_t pnamelen = pname ? strlen(pname) : 0;
1633 // Return error on buffer overflow.
1634 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1635 return retval;
1636 }
1638 if (pnamelen == 0) {
1639 snprintf(buffer, buflen, "lib%s.so", fname);
1640 retval = true;
1641 } else if (strchr(pname, *os::path_separator()) != NULL) {
1642 int n;
1643 char** pelements = split_path(pname, &n);
1644 if (pelements == NULL) {
1645 return false;
1646 }
1647 for (int i = 0 ; i < n ; i++) {
1648 // Really shouldn't be NULL, but check can't hurt
1649 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1650 continue; // skip the empty path values
1651 }
1652 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1653 if (file_exists(buffer)) {
1654 retval = true;
1655 break;
1656 }
1657 }
1658 // release the storage
1659 for (int i = 0 ; i < n ; i++) {
1660 if (pelements[i] != NULL) {
1661 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1662 }
1663 }
1664 if (pelements != NULL) {
1665 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1666 }
1667 } else {
1668 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1669 retval = true;
1670 }
1671 return retval;
1672 }
1674 // check if addr is inside libjvm.so
1675 bool os::address_is_in_vm(address addr) {
1676 static address libjvm_base_addr;
1677 Dl_info dlinfo;
1679 if (libjvm_base_addr == NULL) {
1680 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1681 libjvm_base_addr = (address)dlinfo.dli_fbase;
1682 }
1683 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1684 }
1686 if (dladdr((void *)addr, &dlinfo) != 0) {
1687 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1688 }
1690 return false;
1691 }
1693 bool os::dll_address_to_function_name(address addr, char *buf,
1694 int buflen, int *offset) {
1695 // buf is not optional, but offset is optional
1696 assert(buf != NULL, "sanity check");
1698 Dl_info dlinfo;
1700 if (dladdr((void*)addr, &dlinfo) != 0) {
1701 // see if we have a matching symbol
1702 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1703 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1704 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1705 }
1706 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1707 return true;
1708 }
1709 // no matching symbol so try for just file info
1710 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1711 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1712 buf, buflen, offset, dlinfo.dli_fname)) {
1713 return true;
1714 }
1715 }
1716 }
1718 buf[0] = '\0';
1719 if (offset != NULL) *offset = -1;
1720 return false;
1721 }
1723 struct _address_to_library_name {
1724 address addr; // input : memory address
1725 size_t buflen; // size of fname
1726 char* fname; // output: library name
1727 address base; // library base addr
1728 };
1730 static int address_to_library_name_callback(struct dl_phdr_info *info,
1731 size_t size, void *data) {
1732 int i;
1733 bool found = false;
1734 address libbase = NULL;
1735 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1737 // iterate through all loadable segments
1738 for (i = 0; i < info->dlpi_phnum; i++) {
1739 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1740 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1741 // base address of a library is the lowest address of its loaded
1742 // segments.
1743 if (libbase == NULL || libbase > segbase) {
1744 libbase = segbase;
1745 }
1746 // see if 'addr' is within current segment
1747 if (segbase <= d->addr &&
1748 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1749 found = true;
1750 }
1751 }
1752 }
1754 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1755 // so dll_address_to_library_name() can fall through to use dladdr() which
1756 // can figure out executable name from argv[0].
1757 if (found && info->dlpi_name && info->dlpi_name[0]) {
1758 d->base = libbase;
1759 if (d->fname) {
1760 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1761 }
1762 return 1;
1763 }
1764 return 0;
1765 }
1767 bool os::dll_address_to_library_name(address addr, char* buf,
1768 int buflen, int* offset) {
1769 // buf is not optional, but offset is optional
1770 assert(buf != NULL, "sanity check");
1772 Dl_info dlinfo;
1773 struct _address_to_library_name data;
1775 // There is a bug in old glibc dladdr() implementation that it could resolve
1776 // to wrong library name if the .so file has a base address != NULL. Here
1777 // we iterate through the program headers of all loaded libraries to find
1778 // out which library 'addr' really belongs to. This workaround can be
1779 // removed once the minimum requirement for glibc is moved to 2.3.x.
1780 data.addr = addr;
1781 data.fname = buf;
1782 data.buflen = buflen;
1783 data.base = NULL;
1784 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1786 if (rslt) {
1787 // buf already contains library name
1788 if (offset) *offset = addr - data.base;
1789 return true;
1790 }
1791 if (dladdr((void*)addr, &dlinfo) != 0) {
1792 if (dlinfo.dli_fname != NULL) {
1793 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1794 }
1795 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1796 *offset = addr - (address)dlinfo.dli_fbase;
1797 }
1798 return true;
1799 }
1801 buf[0] = '\0';
1802 if (offset) *offset = -1;
1803 return false;
1804 }
1806 // Loads .dll/.so and
1807 // in case of error it checks if .dll/.so was built for the
1808 // same architecture as Hotspot is running on
1811 // Remember the stack's state. The Linux dynamic linker will change
1812 // the stack to 'executable' at most once, so we must safepoint only once.
1813 bool os::Linux::_stack_is_executable = false;
1815 // VM operation that loads a library. This is necessary if stack protection
1816 // of the Java stacks can be lost during loading the library. If we
1817 // do not stop the Java threads, they can stack overflow before the stacks
1818 // are protected again.
1819 class VM_LinuxDllLoad: public VM_Operation {
1820 private:
1821 const char *_filename;
1822 char *_ebuf;
1823 int _ebuflen;
1824 void *_lib;
1825 public:
1826 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1827 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1828 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1829 void doit() {
1830 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1831 os::Linux::_stack_is_executable = true;
1832 }
1833 void* loaded_library() { return _lib; }
1834 };
1836 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1837 {
1838 void * result = NULL;
1839 bool load_attempted = false;
1841 // Check whether the library to load might change execution rights
1842 // of the stack. If they are changed, the protection of the stack
1843 // guard pages will be lost. We need a safepoint to fix this.
1844 //
1845 // See Linux man page execstack(8) for more info.
1846 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1847 ElfFile ef(filename);
1848 if (!ef.specifies_noexecstack()) {
1849 if (!is_init_completed()) {
1850 os::Linux::_stack_is_executable = true;
1851 // This is OK - No Java threads have been created yet, and hence no
1852 // stack guard pages to fix.
1853 //
1854 // This should happen only when you are building JDK7 using a very
1855 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1856 //
1857 // Dynamic loader will make all stacks executable after
1858 // this function returns, and will not do that again.
1859 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1860 } else {
1861 warning("You have loaded library %s which might have disabled stack guard. "
1862 "The VM will try to fix the stack guard now.\n"
1863 "It's highly recommended that you fix the library with "
1864 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1865 filename);
1867 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1868 JavaThread *jt = JavaThread::current();
1869 if (jt->thread_state() != _thread_in_native) {
1870 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1871 // that requires ExecStack. Cannot enter safe point. Let's give up.
1872 warning("Unable to fix stack guard. Giving up.");
1873 } else {
1874 if (!LoadExecStackDllInVMThread) {
1875 // This is for the case where the DLL has an static
1876 // constructor function that executes JNI code. We cannot
1877 // load such DLLs in the VMThread.
1878 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1879 }
1881 ThreadInVMfromNative tiv(jt);
1882 debug_only(VMNativeEntryWrapper vew;)
1884 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1885 VMThread::execute(&op);
1886 if (LoadExecStackDllInVMThread) {
1887 result = op.loaded_library();
1888 }
1889 load_attempted = true;
1890 }
1891 }
1892 }
1893 }
1895 if (!load_attempted) {
1896 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1897 }
1899 if (result != NULL) {
1900 // Successful loading
1901 return result;
1902 }
1904 Elf32_Ehdr elf_head;
1905 int diag_msg_max_length=ebuflen-strlen(ebuf);
1906 char* diag_msg_buf=ebuf+strlen(ebuf);
1908 if (diag_msg_max_length==0) {
1909 // No more space in ebuf for additional diagnostics message
1910 return NULL;
1911 }
1914 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1916 if (file_descriptor < 0) {
1917 // Can't open library, report dlerror() message
1918 return NULL;
1919 }
1921 bool failed_to_read_elf_head=
1922 (sizeof(elf_head)!=
1923 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1925 ::close(file_descriptor);
1926 if (failed_to_read_elf_head) {
1927 // file i/o error - report dlerror() msg
1928 return NULL;
1929 }
1931 typedef struct {
1932 Elf32_Half code; // Actual value as defined in elf.h
1933 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1934 char elf_class; // 32 or 64 bit
1935 char endianess; // MSB or LSB
1936 char* name; // String representation
1937 } arch_t;
1939 #ifndef EM_486
1940 #define EM_486 6 /* Intel 80486 */
1941 #endif
1943 static const arch_t arch_array[]={
1944 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1945 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1946 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1947 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1948 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1949 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1950 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1951 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1952 #if defined(VM_LITTLE_ENDIAN)
1953 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1954 #else
1955 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1956 #endif
1957 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1958 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1959 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1960 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1961 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1962 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1963 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1964 };
1966 #if (defined IA32)
1967 static Elf32_Half running_arch_code=EM_386;
1968 #elif (defined AMD64)
1969 static Elf32_Half running_arch_code=EM_X86_64;
1970 #elif (defined IA64)
1971 static Elf32_Half running_arch_code=EM_IA_64;
1972 #elif (defined __sparc) && (defined _LP64)
1973 static Elf32_Half running_arch_code=EM_SPARCV9;
1974 #elif (defined __sparc) && (!defined _LP64)
1975 static Elf32_Half running_arch_code=EM_SPARC;
1976 #elif (defined __powerpc64__)
1977 static Elf32_Half running_arch_code=EM_PPC64;
1978 #elif (defined __powerpc__)
1979 static Elf32_Half running_arch_code=EM_PPC;
1980 #elif (defined ARM)
1981 static Elf32_Half running_arch_code=EM_ARM;
1982 #elif (defined S390)
1983 static Elf32_Half running_arch_code=EM_S390;
1984 #elif (defined ALPHA)
1985 static Elf32_Half running_arch_code=EM_ALPHA;
1986 #elif (defined MIPSEL)
1987 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1988 #elif (defined PARISC)
1989 static Elf32_Half running_arch_code=EM_PARISC;
1990 #elif (defined MIPS)
1991 static Elf32_Half running_arch_code=EM_MIPS;
1992 #elif (defined M68K)
1993 static Elf32_Half running_arch_code=EM_68K;
1994 #else
1995 #error Method os::dll_load requires that one of following is defined:\
1996 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1997 #endif
1999 // Identify compatability class for VM's architecture and library's architecture
2000 // Obtain string descriptions for architectures
2002 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2003 int running_arch_index=-1;
2005 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2006 if (running_arch_code == arch_array[i].code) {
2007 running_arch_index = i;
2008 }
2009 if (lib_arch.code == arch_array[i].code) {
2010 lib_arch.compat_class = arch_array[i].compat_class;
2011 lib_arch.name = arch_array[i].name;
2012 }
2013 }
2015 assert(running_arch_index != -1,
2016 "Didn't find running architecture code (running_arch_code) in arch_array");
2017 if (running_arch_index == -1) {
2018 // Even though running architecture detection failed
2019 // we may still continue with reporting dlerror() message
2020 return NULL;
2021 }
2023 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2024 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2025 return NULL;
2026 }
2028 #ifndef S390
2029 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2030 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2031 return NULL;
2032 }
2033 #endif // !S390
2035 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2036 if ( lib_arch.name!=NULL ) {
2037 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2038 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2039 lib_arch.name, arch_array[running_arch_index].name);
2040 } else {
2041 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2042 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2043 lib_arch.code,
2044 arch_array[running_arch_index].name);
2045 }
2046 }
2048 return NULL;
2049 }
2051 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2052 void * result = ::dlopen(filename, RTLD_LAZY);
2053 if (result == NULL) {
2054 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2055 ebuf[ebuflen-1] = '\0';
2056 }
2057 return result;
2058 }
2060 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2061 void * result = NULL;
2062 if (LoadExecStackDllInVMThread) {
2063 result = dlopen_helper(filename, ebuf, ebuflen);
2064 }
2066 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2067 // library that requires an executable stack, or which does not have this
2068 // stack attribute set, dlopen changes the stack attribute to executable. The
2069 // read protection of the guard pages gets lost.
2070 //
2071 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2072 // may have been queued at the same time.
2074 if (!_stack_is_executable) {
2075 JavaThread *jt = Threads::first();
2077 while (jt) {
2078 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2079 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2080 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2081 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2082 warning("Attempt to reguard stack yellow zone failed.");
2083 }
2084 }
2085 jt = jt->next();
2086 }
2087 }
2089 return result;
2090 }
2092 /*
2093 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2094 * chances are you might want to run the generated bits against glibc-2.0
2095 * libdl.so, so always use locking for any version of glibc.
2096 */
2097 void* os::dll_lookup(void* handle, const char* name) {
2098 pthread_mutex_lock(&dl_mutex);
2099 void* res = dlsym(handle, name);
2100 pthread_mutex_unlock(&dl_mutex);
2101 return res;
2102 }
2104 void* os::get_default_process_handle() {
2105 return (void*)::dlopen(NULL, RTLD_LAZY);
2106 }
2108 static bool _print_ascii_file(const char* filename, outputStream* st) {
2109 int fd = ::open(filename, O_RDONLY);
2110 if (fd == -1) {
2111 return false;
2112 }
2114 char buf[32];
2115 int bytes;
2116 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2117 st->print_raw(buf, bytes);
2118 }
2120 ::close(fd);
2122 return true;
2123 }
2125 void os::print_dll_info(outputStream *st) {
2126 st->print_cr("Dynamic libraries:");
2128 char fname[32];
2129 pid_t pid = os::Linux::gettid();
2131 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2133 if (!_print_ascii_file(fname, st)) {
2134 st->print("Can not get library information for pid = %d\n", pid);
2135 }
2136 }
2138 void os::print_os_info_brief(outputStream* st) {
2139 os::Linux::print_distro_info(st);
2141 os::Posix::print_uname_info(st);
2143 os::Linux::print_libversion_info(st);
2145 }
2147 void os::print_os_info(outputStream* st) {
2148 st->print("OS:");
2150 os::Linux::print_distro_info(st);
2152 os::Posix::print_uname_info(st);
2154 // Print warning if unsafe chroot environment detected
2155 if (unsafe_chroot_detected) {
2156 st->print("WARNING!! ");
2157 st->print_cr("%s", unstable_chroot_error);
2158 }
2160 os::Linux::print_libversion_info(st);
2162 os::Posix::print_rlimit_info(st);
2164 os::Posix::print_load_average(st);
2166 os::Linux::print_full_memory_info(st);
2168 os::Linux::print_container_info(st);
2169 }
2171 // Try to identify popular distros.
2172 // Most Linux distributions have a /etc/XXX-release file, which contains
2173 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2174 // file that also contains the OS version string. Some have more than one
2175 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2176 // /etc/redhat-release.), so the order is important.
2177 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2178 // their own specific XXX-release file as well as a redhat-release file.
2179 // Because of this the XXX-release file needs to be searched for before the
2180 // redhat-release file.
2181 // Since Red Hat has a lsb-release file that is not very descriptive the
2182 // search for redhat-release needs to be before lsb-release.
2183 // Since the lsb-release file is the new standard it needs to be searched
2184 // before the older style release files.
2185 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2186 // next to last resort. The os-release file is a new standard that contains
2187 // distribution information and the system-release file seems to be an old
2188 // standard that has been replaced by the lsb-release and os-release files.
2189 // Searching for the debian_version file is the last resort. It contains
2190 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2191 // "Debian " is printed before the contents of the debian_version file.
2192 void os::Linux::print_distro_info(outputStream* st) {
2193 if (!_print_ascii_file("/etc/oracle-release", st) &&
2194 !_print_ascii_file("/etc/mandriva-release", st) &&
2195 !_print_ascii_file("/etc/mandrake-release", st) &&
2196 !_print_ascii_file("/etc/sun-release", st) &&
2197 !_print_ascii_file("/etc/redhat-release", st) &&
2198 !_print_ascii_file("/etc/lsb-release", st) &&
2199 !_print_ascii_file("/etc/SuSE-release", st) &&
2200 !_print_ascii_file("/etc/turbolinux-release", st) &&
2201 !_print_ascii_file("/etc/gentoo-release", st) &&
2202 !_print_ascii_file("/etc/ltib-release", st) &&
2203 !_print_ascii_file("/etc/angstrom-version", st) &&
2204 !_print_ascii_file("/etc/system-release", st) &&
2205 !_print_ascii_file("/etc/os-release", st)) {
2207 if (file_exists("/etc/debian_version")) {
2208 st->print("Debian ");
2209 _print_ascii_file("/etc/debian_version", st);
2210 } else {
2211 st->print("Linux");
2212 }
2213 }
2214 st->cr();
2215 }
2217 void os::Linux::print_libversion_info(outputStream* st) {
2218 // libc, pthread
2219 st->print("libc:");
2220 st->print("%s ", os::Linux::glibc_version());
2221 st->print("%s ", os::Linux::libpthread_version());
2222 if (os::Linux::is_LinuxThreads()) {
2223 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2224 }
2225 st->cr();
2226 }
2228 void os::Linux::print_full_memory_info(outputStream* st) {
2229 st->print("\n/proc/meminfo:\n");
2230 _print_ascii_file("/proc/meminfo", st);
2231 st->cr();
2232 }
2234 void os::Linux::print_container_info(outputStream* st) {
2235 if (!OSContainer::is_containerized()) {
2236 return;
2237 }
2239 st->print("container (cgroup) information:\n");
2241 const char *p_ct = OSContainer::container_type();
2242 st->print("container_type: %s\n", p_ct != NULL ? p_ct : "failed");
2244 char *p = OSContainer::cpu_cpuset_cpus();
2245 st->print("cpu_cpuset_cpus: %s\n", p != NULL ? p : "failed");
2246 free(p);
2248 p = OSContainer::cpu_cpuset_memory_nodes();
2249 st->print("cpu_memory_nodes: %s\n", p != NULL ? p : "failed");
2250 free(p);
2252 int i = OSContainer::active_processor_count();
2253 if (i > 0) {
2254 st->print("active_processor_count: %d\n", i);
2255 } else {
2256 st->print("active_processor_count: failed\n");
2257 }
2259 i = OSContainer::cpu_quota();
2260 st->print("cpu_quota: %d\n", i);
2262 i = OSContainer::cpu_period();
2263 st->print("cpu_period: %d\n", i);
2265 i = OSContainer::cpu_shares();
2266 st->print("cpu_shares: %d\n", i);
2268 jlong j = OSContainer::memory_limit_in_bytes();
2269 st->print("memory_limit_in_bytes: " JLONG_FORMAT "\n", j);
2271 j = OSContainer::memory_and_swap_limit_in_bytes();
2272 st->print("memory_and_swap_limit_in_bytes: " JLONG_FORMAT "\n", j);
2274 j = OSContainer::memory_soft_limit_in_bytes();
2275 st->print("memory_soft_limit_in_bytes: " JLONG_FORMAT "\n", j);
2277 j = OSContainer::OSContainer::memory_usage_in_bytes();
2278 st->print("memory_usage_in_bytes: " JLONG_FORMAT "\n", j);
2280 j = OSContainer::OSContainer::memory_max_usage_in_bytes();
2281 st->print("memory_max_usage_in_bytes: " JLONG_FORMAT "\n", j);
2282 st->cr();
2283 }
2285 void os::print_memory_info(outputStream* st) {
2287 st->print("Memory:");
2288 st->print(" %dk page", os::vm_page_size()>>10);
2290 // values in struct sysinfo are "unsigned long"
2291 struct sysinfo si;
2292 sysinfo(&si);
2294 st->print(", physical " UINT64_FORMAT "k",
2295 os::physical_memory() >> 10);
2296 st->print("(" UINT64_FORMAT "k free)",
2297 os::available_memory() >> 10);
2298 st->print(", swap " UINT64_FORMAT "k",
2299 ((jlong)si.totalswap * si.mem_unit) >> 10);
2300 st->print("(" UINT64_FORMAT "k free)",
2301 ((jlong)si.freeswap * si.mem_unit) >> 10);
2302 st->cr();
2303 }
2305 void os::pd_print_cpu_info(outputStream* st) {
2306 st->print("\n/proc/cpuinfo:\n");
2307 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2308 st->print(" <Not Available>");
2309 }
2310 st->cr();
2311 }
2313 void os::print_siginfo(outputStream* st, void* siginfo) {
2314 const siginfo_t* si = (const siginfo_t*)siginfo;
2316 os::Posix::print_siginfo_brief(st, si);
2317 #if INCLUDE_CDS
2318 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2319 UseSharedSpaces) {
2320 FileMapInfo* mapinfo = FileMapInfo::current_info();
2321 if (mapinfo->is_in_shared_space(si->si_addr)) {
2322 st->print("\n\nError accessing class data sharing archive." \
2323 " Mapped file inaccessible during execution, " \
2324 " possible disk/network problem.");
2325 }
2326 }
2327 #endif
2328 st->cr();
2329 }
2332 static void print_signal_handler(outputStream* st, int sig,
2333 char* buf, size_t buflen);
2335 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2336 st->print_cr("Signal Handlers:");
2337 print_signal_handler(st, SIGSEGV, buf, buflen);
2338 print_signal_handler(st, SIGBUS , buf, buflen);
2339 print_signal_handler(st, SIGFPE , buf, buflen);
2340 print_signal_handler(st, SIGPIPE, buf, buflen);
2341 print_signal_handler(st, SIGXFSZ, buf, buflen);
2342 print_signal_handler(st, SIGILL , buf, buflen);
2343 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2344 print_signal_handler(st, SR_signum, buf, buflen);
2345 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2346 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2347 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2348 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2349 #if defined(PPC64)
2350 print_signal_handler(st, SIGTRAP, buf, buflen);
2351 #endif
2352 }
2354 static char saved_jvm_path[MAXPATHLEN] = {0};
2356 // Find the full path to the current module, libjvm.so
2357 void os::jvm_path(char *buf, jint buflen) {
2358 // Error checking.
2359 if (buflen < MAXPATHLEN) {
2360 assert(false, "must use a large-enough buffer");
2361 buf[0] = '\0';
2362 return;
2363 }
2364 // Lazy resolve the path to current module.
2365 if (saved_jvm_path[0] != 0) {
2366 strcpy(buf, saved_jvm_path);
2367 return;
2368 }
2370 char dli_fname[MAXPATHLEN];
2371 bool ret = dll_address_to_library_name(
2372 CAST_FROM_FN_PTR(address, os::jvm_path),
2373 dli_fname, sizeof(dli_fname), NULL);
2374 assert(ret, "cannot locate libjvm");
2375 char *rp = NULL;
2376 if (ret && dli_fname[0] != '\0') {
2377 rp = realpath(dli_fname, buf);
2378 }
2379 if (rp == NULL)
2380 return;
2382 if (Arguments::created_by_gamma_launcher()) {
2383 // Support for the gamma launcher. Typical value for buf is
2384 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2385 // the right place in the string, then assume we are installed in a JDK and
2386 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2387 // up the path so it looks like libjvm.so is installed there (append a
2388 // fake suffix hotspot/libjvm.so).
2389 const char *p = buf + strlen(buf) - 1;
2390 for (int count = 0; p > buf && count < 5; ++count) {
2391 for (--p; p > buf && *p != '/'; --p)
2392 /* empty */ ;
2393 }
2395 if (strncmp(p, "/jre/lib/", 9) != 0) {
2396 // Look for JAVA_HOME in the environment.
2397 char* java_home_var = ::getenv("JAVA_HOME");
2398 if (java_home_var != NULL && java_home_var[0] != 0) {
2399 char* jrelib_p;
2400 int len;
2402 // Check the current module name "libjvm.so".
2403 p = strrchr(buf, '/');
2404 assert(strstr(p, "/libjvm") == p, "invalid library name");
2406 rp = realpath(java_home_var, buf);
2407 if (rp == NULL)
2408 return;
2410 // determine if this is a legacy image or modules image
2411 // modules image doesn't have "jre" subdirectory
2412 len = strlen(buf);
2413 assert(len < buflen, "Ran out of buffer room");
2414 jrelib_p = buf + len;
2415 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2416 if (0 != access(buf, F_OK)) {
2417 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2418 }
2420 if (0 == access(buf, F_OK)) {
2421 // Use current module name "libjvm.so"
2422 len = strlen(buf);
2423 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2424 } else {
2425 // Go back to path of .so
2426 rp = realpath(dli_fname, buf);
2427 if (rp == NULL)
2428 return;
2429 }
2430 }
2431 }
2432 }
2434 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2435 }
2437 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2438 // no prefix required, not even "_"
2439 }
2441 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2442 // no suffix required
2443 }
2445 ////////////////////////////////////////////////////////////////////////////////
2446 // sun.misc.Signal support
2448 static volatile jint sigint_count = 0;
2450 static void
2451 UserHandler(int sig, void *siginfo, void *context) {
2452 // 4511530 - sem_post is serialized and handled by the manager thread. When
2453 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2454 // don't want to flood the manager thread with sem_post requests.
2455 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2456 return;
2458 // Ctrl-C is pressed during error reporting, likely because the error
2459 // handler fails to abort. Let VM die immediately.
2460 if (sig == SIGINT && is_error_reported()) {
2461 os::die();
2462 }
2464 os::signal_notify(sig);
2465 }
2467 void* os::user_handler() {
2468 return CAST_FROM_FN_PTR(void*, UserHandler);
2469 }
2471 class Semaphore : public StackObj {
2472 public:
2473 Semaphore();
2474 ~Semaphore();
2475 void signal();
2476 void wait();
2477 bool trywait();
2478 bool timedwait(unsigned int sec, int nsec);
2479 private:
2480 sem_t _semaphore;
2481 };
2483 Semaphore::Semaphore() {
2484 sem_init(&_semaphore, 0, 0);
2485 }
2487 Semaphore::~Semaphore() {
2488 sem_destroy(&_semaphore);
2489 }
2491 void Semaphore::signal() {
2492 sem_post(&_semaphore);
2493 }
2495 void Semaphore::wait() {
2496 sem_wait(&_semaphore);
2497 }
2499 bool Semaphore::trywait() {
2500 return sem_trywait(&_semaphore) == 0;
2501 }
2503 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2505 struct timespec ts;
2506 // Semaphore's are always associated with CLOCK_REALTIME
2507 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2508 // see unpackTime for discussion on overflow checking
2509 if (sec >= MAX_SECS) {
2510 ts.tv_sec += MAX_SECS;
2511 ts.tv_nsec = 0;
2512 } else {
2513 ts.tv_sec += sec;
2514 ts.tv_nsec += nsec;
2515 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2516 ts.tv_nsec -= NANOSECS_PER_SEC;
2517 ++ts.tv_sec; // note: this must be <= max_secs
2518 }
2519 }
2521 while (1) {
2522 int result = sem_timedwait(&_semaphore, &ts);
2523 if (result == 0) {
2524 return true;
2525 } else if (errno == EINTR) {
2526 continue;
2527 } else if (errno == ETIMEDOUT) {
2528 return false;
2529 } else {
2530 return false;
2531 }
2532 }
2533 }
2535 extern "C" {
2536 typedef void (*sa_handler_t)(int);
2537 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2538 }
2540 void* os::signal(int signal_number, void* handler) {
2541 struct sigaction sigAct, oldSigAct;
2543 sigfillset(&(sigAct.sa_mask));
2544 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2545 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2547 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2548 // -1 means registration failed
2549 return (void *)-1;
2550 }
2552 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2553 }
2555 void os::signal_raise(int signal_number) {
2556 ::raise(signal_number);
2557 }
2559 /*
2560 * The following code is moved from os.cpp for making this
2561 * code platform specific, which it is by its very nature.
2562 */
2564 // Will be modified when max signal is changed to be dynamic
2565 int os::sigexitnum_pd() {
2566 return NSIG;
2567 }
2569 // a counter for each possible signal value
2570 static volatile jint pending_signals[NSIG+1] = { 0 };
2572 // Linux(POSIX) specific hand shaking semaphore.
2573 static sem_t sig_sem;
2574 static Semaphore sr_semaphore;
2576 void os::signal_init_pd() {
2577 // Initialize signal structures
2578 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2580 // Initialize signal semaphore
2581 ::sem_init(&sig_sem, 0, 0);
2582 }
2584 void os::signal_notify(int sig) {
2585 Atomic::inc(&pending_signals[sig]);
2586 ::sem_post(&sig_sem);
2587 }
2589 static int check_pending_signals(bool wait) {
2590 Atomic::store(0, &sigint_count);
2591 for (;;) {
2592 for (int i = 0; i < NSIG + 1; i++) {
2593 jint n = pending_signals[i];
2594 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2595 return i;
2596 }
2597 }
2598 if (!wait) {
2599 return -1;
2600 }
2601 JavaThread *thread = JavaThread::current();
2602 ThreadBlockInVM tbivm(thread);
2604 bool threadIsSuspended;
2605 do {
2606 thread->set_suspend_equivalent();
2607 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2608 ::sem_wait(&sig_sem);
2610 // were we externally suspended while we were waiting?
2611 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2612 if (threadIsSuspended) {
2613 //
2614 // The semaphore has been incremented, but while we were waiting
2615 // another thread suspended us. We don't want to continue running
2616 // while suspended because that would surprise the thread that
2617 // suspended us.
2618 //
2619 ::sem_post(&sig_sem);
2621 thread->java_suspend_self();
2622 }
2623 } while (threadIsSuspended);
2624 }
2625 }
2627 int os::signal_lookup() {
2628 return check_pending_signals(false);
2629 }
2631 int os::signal_wait() {
2632 return check_pending_signals(true);
2633 }
2635 ////////////////////////////////////////////////////////////////////////////////
2636 // Virtual Memory
2638 int os::vm_page_size() {
2639 // Seems redundant as all get out
2640 assert(os::Linux::page_size() != -1, "must call os::init");
2641 return os::Linux::page_size();
2642 }
2644 // Solaris allocates memory by pages.
2645 int os::vm_allocation_granularity() {
2646 assert(os::Linux::page_size() != -1, "must call os::init");
2647 return os::Linux::page_size();
2648 }
2650 // Rationale behind this function:
2651 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2652 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2653 // samples for JITted code. Here we create private executable mapping over the code cache
2654 // and then we can use standard (well, almost, as mapping can change) way to provide
2655 // info for the reporting script by storing timestamp and location of symbol
2656 void linux_wrap_code(char* base, size_t size) {
2657 static volatile jint cnt = 0;
2659 if (!UseOprofile) {
2660 return;
2661 }
2663 char buf[PATH_MAX+1];
2664 int num = Atomic::add(1, &cnt);
2666 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2667 os::get_temp_directory(), os::current_process_id(), num);
2668 unlink(buf);
2670 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2672 if (fd != -1) {
2673 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2674 if (rv != (off_t)-1) {
2675 if (::write(fd, "", 1) == 1) {
2676 mmap(base, size,
2677 PROT_READ|PROT_WRITE|PROT_EXEC,
2678 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2679 }
2680 }
2681 ::close(fd);
2682 unlink(buf);
2683 }
2684 }
2686 static bool recoverable_mmap_error(int err) {
2687 // See if the error is one we can let the caller handle. This
2688 // list of errno values comes from JBS-6843484. I can't find a
2689 // Linux man page that documents this specific set of errno
2690 // values so while this list currently matches Solaris, it may
2691 // change as we gain experience with this failure mode.
2692 switch (err) {
2693 case EBADF:
2694 case EINVAL:
2695 case ENOTSUP:
2696 // let the caller deal with these errors
2697 return true;
2699 default:
2700 // Any remaining errors on this OS can cause our reserved mapping
2701 // to be lost. That can cause confusion where different data
2702 // structures think they have the same memory mapped. The worst
2703 // scenario is if both the VM and a library think they have the
2704 // same memory mapped.
2705 return false;
2706 }
2707 }
2709 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2710 int err) {
2711 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2712 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2713 strerror(err), err);
2714 }
2716 static void warn_fail_commit_memory(char* addr, size_t size,
2717 size_t alignment_hint, bool exec,
2718 int err) {
2719 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2720 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2721 alignment_hint, exec, strerror(err), err);
2722 }
2724 // NOTE: Linux kernel does not really reserve the pages for us.
2725 // All it does is to check if there are enough free pages
2726 // left at the time of mmap(). This could be a potential
2727 // problem.
2728 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2729 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2730 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2731 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2732 if (res != (uintptr_t) MAP_FAILED) {
2733 if (UseNUMAInterleaving) {
2734 numa_make_global(addr, size);
2735 }
2736 return 0;
2737 }
2739 int err = errno; // save errno from mmap() call above
2741 if (!recoverable_mmap_error(err)) {
2742 warn_fail_commit_memory(addr, size, exec, err);
2743 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2744 }
2746 return err;
2747 }
2749 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2750 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2751 }
2753 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2754 const char* mesg) {
2755 assert(mesg != NULL, "mesg must be specified");
2756 int err = os::Linux::commit_memory_impl(addr, size, exec);
2757 if (err != 0) {
2758 // the caller wants all commit errors to exit with the specified mesg:
2759 warn_fail_commit_memory(addr, size, exec, err);
2760 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2761 }
2762 }
2764 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2765 #ifndef MAP_HUGETLB
2766 #define MAP_HUGETLB 0x40000
2767 #endif
2769 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2770 #ifndef MADV_HUGEPAGE
2771 #define MADV_HUGEPAGE 14
2772 #endif
2774 int os::Linux::commit_memory_impl(char* addr, size_t size,
2775 size_t alignment_hint, bool exec) {
2776 int err = os::Linux::commit_memory_impl(addr, size, exec);
2777 if (err == 0) {
2778 realign_memory(addr, size, alignment_hint);
2779 }
2780 return err;
2781 }
2783 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2784 bool exec) {
2785 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2786 }
2788 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2789 size_t alignment_hint, bool exec,
2790 const char* mesg) {
2791 assert(mesg != NULL, "mesg must be specified");
2792 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2793 if (err != 0) {
2794 // the caller wants all commit errors to exit with the specified mesg:
2795 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2796 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2797 }
2798 }
2800 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2801 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2802 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2803 // be supported or the memory may already be backed by huge pages.
2804 ::madvise(addr, bytes, MADV_HUGEPAGE);
2805 }
2806 }
2808 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2809 // This method works by doing an mmap over an existing mmaping and effectively discarding
2810 // the existing pages. However it won't work for SHM-based large pages that cannot be
2811 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2812 // small pages on top of the SHM segment. This method always works for small pages, so we
2813 // allow that in any case.
2814 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2815 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2816 }
2817 }
2819 void os::numa_make_global(char *addr, size_t bytes) {
2820 Linux::numa_interleave_memory(addr, bytes);
2821 }
2823 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2824 // bind policy to MPOL_PREFERRED for the current thread.
2825 #define USE_MPOL_PREFERRED 0
2827 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2828 // To make NUMA and large pages more robust when both enabled, we need to ease
2829 // the requirements on where the memory should be allocated. MPOL_BIND is the
2830 // default policy and it will force memory to be allocated on the specified
2831 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2832 // the specified node, but will not force it. Using this policy will prevent
2833 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2834 // free large pages.
2835 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2836 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2837 }
2839 bool os::numa_topology_changed() { return false; }
2841 size_t os::numa_get_groups_num() {
2842 // Return just the number of nodes in which it's possible to allocate memory
2843 // (in numa terminology, configured nodes).
2844 return Linux::numa_num_configured_nodes();
2845 }
2847 int os::numa_get_group_id() {
2848 int cpu_id = Linux::sched_getcpu();
2849 if (cpu_id != -1) {
2850 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2851 if (lgrp_id != -1) {
2852 return lgrp_id;
2853 }
2854 }
2855 return 0;
2856 }
2858 int os::Linux::get_existing_num_nodes() {
2859 size_t node;
2860 size_t highest_node_number = Linux::numa_max_node();
2861 int num_nodes = 0;
2863 // Get the total number of nodes in the system including nodes without memory.
2864 for (node = 0; node <= highest_node_number; node++) {
2865 if (isnode_in_existing_nodes(node)) {
2866 num_nodes++;
2867 }
2868 }
2869 return num_nodes;
2870 }
2872 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2873 size_t highest_node_number = Linux::numa_max_node();
2874 size_t i = 0;
2876 // Map all node ids in which is possible to allocate memory. Also nodes are
2877 // not always consecutively available, i.e. available from 0 to the highest
2878 // node number.
2879 for (size_t node = 0; node <= highest_node_number; node++) {
2880 if (Linux::isnode_in_configured_nodes(node)) {
2881 ids[i++] = node;
2882 }
2883 }
2884 return i;
2885 }
2887 bool os::get_page_info(char *start, page_info* info) {
2888 return false;
2889 }
2891 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2892 return end;
2893 }
2896 int os::Linux::sched_getcpu_syscall(void) {
2897 unsigned int cpu = 0;
2898 int retval = -1;
2900 #if defined(IA32)
2901 # ifndef SYS_getcpu
2902 # define SYS_getcpu 318
2903 # endif
2904 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2905 #elif defined(AMD64)
2906 // Unfortunately we have to bring all these macros here from vsyscall.h
2907 // to be able to compile on old linuxes.
2908 # define __NR_vgetcpu 2
2909 # define VSYSCALL_START (-10UL << 20)
2910 # define VSYSCALL_SIZE 1024
2911 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2912 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2913 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2914 retval = vgetcpu(&cpu, NULL, NULL);
2915 #endif
2917 return (retval == -1) ? retval : cpu;
2918 }
2920 // Something to do with the numa-aware allocator needs these symbols
2921 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2922 extern "C" JNIEXPORT void numa_error(char *where) { }
2923 extern "C" JNIEXPORT int fork1() { return fork(); }
2925 // Handle request to load libnuma symbol version 1.1 (API v1). If it fails
2926 // load symbol from base version instead.
2927 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2928 void *f = dlvsym(handle, name, "libnuma_1.1");
2929 if (f == NULL) {
2930 f = dlsym(handle, name);
2931 }
2932 return f;
2933 }
2935 // Handle request to load libnuma symbol version 1.2 (API v2) only.
2936 // Return NULL if the symbol is not defined in this particular version.
2937 void* os::Linux::libnuma_v2_dlsym(void* handle, const char* name) {
2938 return dlvsym(handle, name, "libnuma_1.2");
2939 }
2941 bool os::Linux::libnuma_init() {
2942 // sched_getcpu() should be in libc.
2943 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2944 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2946 // If it's not, try a direct syscall.
2947 if (sched_getcpu() == -1)
2948 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2950 if (sched_getcpu() != -1) { // Does it work?
2951 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2952 if (handle != NULL) {
2953 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2954 libnuma_dlsym(handle, "numa_node_to_cpus")));
2955 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2956 libnuma_dlsym(handle, "numa_max_node")));
2957 set_numa_num_configured_nodes(CAST_TO_FN_PTR(numa_num_configured_nodes_func_t,
2958 libnuma_dlsym(handle, "numa_num_configured_nodes")));
2959 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2960 libnuma_dlsym(handle, "numa_available")));
2961 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2962 libnuma_dlsym(handle, "numa_tonode_memory")));
2963 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2964 libnuma_dlsym(handle, "numa_interleave_memory")));
2965 set_numa_interleave_memory_v2(CAST_TO_FN_PTR(numa_interleave_memory_v2_func_t,
2966 libnuma_v2_dlsym(handle, "numa_interleave_memory")));
2967 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2968 libnuma_dlsym(handle, "numa_set_bind_policy")));
2969 set_numa_bitmask_isbitset(CAST_TO_FN_PTR(numa_bitmask_isbitset_func_t,
2970 libnuma_dlsym(handle, "numa_bitmask_isbitset")));
2971 set_numa_distance(CAST_TO_FN_PTR(numa_distance_func_t,
2972 libnuma_dlsym(handle, "numa_distance")));
2974 if (numa_available() != -1) {
2975 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2976 set_numa_all_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_all_nodes_ptr"));
2977 set_numa_nodes_ptr((struct bitmask **)libnuma_dlsym(handle, "numa_nodes_ptr"));
2978 // Create an index -> node mapping, since nodes are not always consecutive
2979 _nindex_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2980 rebuild_nindex_to_node_map();
2981 // Create a cpu -> node mapping
2982 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2983 rebuild_cpu_to_node_map();
2984 return true;
2985 }
2986 }
2987 }
2988 return false;
2989 }
2991 void os::Linux::rebuild_nindex_to_node_map() {
2992 int highest_node_number = Linux::numa_max_node();
2994 nindex_to_node()->clear();
2995 for (int node = 0; node <= highest_node_number; node++) {
2996 if (Linux::isnode_in_existing_nodes(node)) {
2997 nindex_to_node()->append(node);
2998 }
2999 }
3000 }
3002 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
3003 // The table is later used in get_node_by_cpu().
3004 void os::Linux::rebuild_cpu_to_node_map() {
3005 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
3006 // in libnuma (possible values are starting from 16,
3007 // and continuing up with every other power of 2, but less
3008 // than the maximum number of CPUs supported by kernel), and
3009 // is a subject to change (in libnuma version 2 the requirements
3010 // are more reasonable) we'll just hardcode the number they use
3011 // in the library.
3012 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
3014 size_t cpu_num = processor_count();
3015 size_t cpu_map_size = NCPUS / BitsPerCLong;
3016 size_t cpu_map_valid_size =
3017 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
3019 cpu_to_node()->clear();
3020 cpu_to_node()->at_grow(cpu_num - 1);
3022 size_t node_num = get_existing_num_nodes();
3024 int distance = 0;
3025 int closest_distance = INT_MAX;
3026 int closest_node = 0;
3027 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
3028 for (size_t i = 0; i < node_num; i++) {
3029 // Check if node is configured (not a memory-less node). If it is not, find
3030 // the closest configured node.
3031 if (!isnode_in_configured_nodes(nindex_to_node()->at(i))) {
3032 closest_distance = INT_MAX;
3033 // Check distance from all remaining nodes in the system. Ignore distance
3034 // from itself and from another non-configured node.
3035 for (size_t m = 0; m < node_num; m++) {
3036 if (m != i && isnode_in_configured_nodes(nindex_to_node()->at(m))) {
3037 distance = numa_distance(nindex_to_node()->at(i), nindex_to_node()->at(m));
3038 // If a closest node is found, update. There is always at least one
3039 // configured node in the system so there is always at least one node
3040 // close.
3041 if (distance != 0 && distance < closest_distance) {
3042 closest_distance = distance;
3043 closest_node = nindex_to_node()->at(m);
3044 }
3045 }
3046 }
3047 } else {
3048 // Current node is already a configured node.
3049 closest_node = nindex_to_node()->at(i);
3050 }
3052 // Get cpus from the original node and map them to the closest node. If node
3053 // is a configured node (not a memory-less node), then original node and
3054 // closest node are the same.
3055 if (numa_node_to_cpus(nindex_to_node()->at(i), cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
3056 for (size_t j = 0; j < cpu_map_valid_size; j++) {
3057 if (cpu_map[j] != 0) {
3058 for (size_t k = 0; k < BitsPerCLong; k++) {
3059 if (cpu_map[j] & (1UL << k)) {
3060 cpu_to_node()->at_put(j * BitsPerCLong + k, closest_node);
3061 }
3062 }
3063 }
3064 }
3065 }
3066 }
3067 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
3068 }
3070 int os::Linux::get_node_by_cpu(int cpu_id) {
3071 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
3072 return cpu_to_node()->at(cpu_id);
3073 }
3074 return -1;
3075 }
3077 GrowableArray<int>* os::Linux::_cpu_to_node;
3078 GrowableArray<int>* os::Linux::_nindex_to_node;
3079 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
3080 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
3081 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
3082 os::Linux::numa_num_configured_nodes_func_t os::Linux::_numa_num_configured_nodes;
3083 os::Linux::numa_available_func_t os::Linux::_numa_available;
3084 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
3085 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
3086 os::Linux::numa_interleave_memory_v2_func_t os::Linux::_numa_interleave_memory_v2;
3087 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
3088 os::Linux::numa_bitmask_isbitset_func_t os::Linux::_numa_bitmask_isbitset;
3089 os::Linux::numa_distance_func_t os::Linux::_numa_distance;
3090 unsigned long* os::Linux::_numa_all_nodes;
3091 struct bitmask* os::Linux::_numa_all_nodes_ptr;
3092 struct bitmask* os::Linux::_numa_nodes_ptr;
3094 bool os::pd_uncommit_memory(char* addr, size_t size) {
3095 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
3096 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
3097 return res != (uintptr_t) MAP_FAILED;
3098 }
3100 static
3101 address get_stack_commited_bottom(address bottom, size_t size) {
3102 address nbot = bottom;
3103 address ntop = bottom + size;
3105 size_t page_sz = os::vm_page_size();
3106 unsigned pages = size / page_sz;
3108 unsigned char vec[1];
3109 unsigned imin = 1, imax = pages + 1, imid;
3110 int mincore_return_value = 0;
3112 assert(imin <= imax, "Unexpected page size");
3114 while (imin < imax) {
3115 imid = (imax + imin) / 2;
3116 nbot = ntop - (imid * page_sz);
3118 // Use a trick with mincore to check whether the page is mapped or not.
3119 // mincore sets vec to 1 if page resides in memory and to 0 if page
3120 // is swapped output but if page we are asking for is unmapped
3121 // it returns -1,ENOMEM
3122 mincore_return_value = mincore(nbot, page_sz, vec);
3124 if (mincore_return_value == -1) {
3125 // Page is not mapped go up
3126 // to find first mapped page
3127 if (errno != EAGAIN) {
3128 assert(errno == ENOMEM, "Unexpected mincore errno");
3129 imax = imid;
3130 }
3131 } else {
3132 // Page is mapped go down
3133 // to find first not mapped page
3134 imin = imid + 1;
3135 }
3136 }
3138 nbot = nbot + page_sz;
3140 // Adjust stack bottom one page up if last checked page is not mapped
3141 if (mincore_return_value == -1) {
3142 nbot = nbot + page_sz;
3143 }
3145 return nbot;
3146 }
3149 // Linux uses a growable mapping for the stack, and if the mapping for
3150 // the stack guard pages is not removed when we detach a thread the
3151 // stack cannot grow beyond the pages where the stack guard was
3152 // mapped. If at some point later in the process the stack expands to
3153 // that point, the Linux kernel cannot expand the stack any further
3154 // because the guard pages are in the way, and a segfault occurs.
3155 //
3156 // However, it's essential not to split the stack region by unmapping
3157 // a region (leaving a hole) that's already part of the stack mapping,
3158 // so if the stack mapping has already grown beyond the guard pages at
3159 // the time we create them, we have to truncate the stack mapping.
3160 // So, we need to know the extent of the stack mapping when
3161 // create_stack_guard_pages() is called.
3163 // We only need this for stacks that are growable: at the time of
3164 // writing thread stacks don't use growable mappings (i.e. those
3165 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3166 // only applies to the main thread.
3168 // If the (growable) stack mapping already extends beyond the point
3169 // where we're going to put our guard pages, truncate the mapping at
3170 // that point by munmap()ping it. This ensures that when we later
3171 // munmap() the guard pages we don't leave a hole in the stack
3172 // mapping. This only affects the main/initial thread
3174 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3176 if (os::Linux::is_initial_thread()) {
3177 // As we manually grow stack up to bottom inside create_attached_thread(),
3178 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3179 // we don't need to do anything special.
3180 // Check it first, before calling heavy function.
3181 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3182 unsigned char vec[1];
3184 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3185 // Fallback to slow path on all errors, including EAGAIN
3186 stack_extent = (uintptr_t) get_stack_commited_bottom(
3187 os::Linux::initial_thread_stack_bottom(),
3188 (size_t)addr - stack_extent);
3189 }
3191 if (stack_extent < (uintptr_t)addr) {
3192 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3193 }
3194 }
3196 return os::commit_memory(addr, size, !ExecMem);
3197 }
3199 // If this is a growable mapping, remove the guard pages entirely by
3200 // munmap()ping them. If not, just call uncommit_memory(). This only
3201 // affects the main/initial thread, but guard against future OS changes
3202 // It's safe to always unmap guard pages for initial thread because we
3203 // always place it right after end of the mapped region
3205 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3206 uintptr_t stack_extent, stack_base;
3208 if (os::Linux::is_initial_thread()) {
3209 return ::munmap(addr, size) == 0;
3210 }
3212 return os::uncommit_memory(addr, size);
3213 }
3215 static address _highest_vm_reserved_address = NULL;
3217 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3218 // at 'requested_addr'. If there are existing memory mappings at the same
3219 // location, however, they will be overwritten. If 'fixed' is false,
3220 // 'requested_addr' is only treated as a hint, the return value may or
3221 // may not start from the requested address. Unlike Linux mmap(), this
3222 // function returns NULL to indicate failure.
3223 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3224 char * addr;
3225 int flags;
3227 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3228 if (fixed) {
3229 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3230 flags |= MAP_FIXED;
3231 }
3233 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3234 // touch an uncommitted page. Otherwise, the read/write might
3235 // succeed if we have enough swap space to back the physical page.
3236 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3237 flags, -1, 0);
3239 if (addr != MAP_FAILED) {
3240 // anon_mmap() should only get called during VM initialization,
3241 // don't need lock (actually we can skip locking even it can be called
3242 // from multiple threads, because _highest_vm_reserved_address is just a
3243 // hint about the upper limit of non-stack memory regions.)
3244 if ((address)addr + bytes > _highest_vm_reserved_address) {
3245 _highest_vm_reserved_address = (address)addr + bytes;
3246 }
3247 }
3249 return addr == MAP_FAILED ? NULL : addr;
3250 }
3252 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3253 // (req_addr != NULL) or with a given alignment.
3254 // - bytes shall be a multiple of alignment.
3255 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3256 // - alignment sets the alignment at which memory shall be allocated.
3257 // It must be a multiple of allocation granularity.
3258 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3259 // req_addr or NULL.
3260 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3262 size_t extra_size = bytes;
3263 if (req_addr == NULL && alignment > 0) {
3264 extra_size += alignment;
3265 }
3267 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3268 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3269 -1, 0);
3270 if (start == MAP_FAILED) {
3271 start = NULL;
3272 } else {
3273 if (req_addr != NULL) {
3274 if (start != req_addr) {
3275 ::munmap(start, extra_size);
3276 start = NULL;
3277 }
3278 } else {
3279 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3280 char* const end_aligned = start_aligned + bytes;
3281 char* const end = start + extra_size;
3282 if (start_aligned > start) {
3283 ::munmap(start, start_aligned - start);
3284 }
3285 if (end_aligned < end) {
3286 ::munmap(end_aligned, end - end_aligned);
3287 }
3288 start = start_aligned;
3289 }
3290 }
3291 return start;
3292 }
3294 // Don't update _highest_vm_reserved_address, because there might be memory
3295 // regions above addr + size. If so, releasing a memory region only creates
3296 // a hole in the address space, it doesn't help prevent heap-stack collision.
3297 //
3298 static int anon_munmap(char * addr, size_t size) {
3299 return ::munmap(addr, size) == 0;
3300 }
3302 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3303 size_t alignment_hint) {
3304 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3305 }
3307 bool os::pd_release_memory(char* addr, size_t size) {
3308 return anon_munmap(addr, size);
3309 }
3311 static address highest_vm_reserved_address() {
3312 return _highest_vm_reserved_address;
3313 }
3315 static bool linux_mprotect(char* addr, size_t size, int prot) {
3316 // Linux wants the mprotect address argument to be page aligned.
3317 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3319 // According to SUSv3, mprotect() should only be used with mappings
3320 // established by mmap(), and mmap() always maps whole pages. Unaligned
3321 // 'addr' likely indicates problem in the VM (e.g. trying to change
3322 // protection of malloc'ed or statically allocated memory). Check the
3323 // caller if you hit this assert.
3324 assert(addr == bottom, "sanity check");
3326 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3327 return ::mprotect(bottom, size, prot) == 0;
3328 }
3330 // Set protections specified
3331 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3332 bool is_committed) {
3333 unsigned int p = 0;
3334 switch (prot) {
3335 case MEM_PROT_NONE: p = PROT_NONE; break;
3336 case MEM_PROT_READ: p = PROT_READ; break;
3337 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3338 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3339 default:
3340 ShouldNotReachHere();
3341 }
3342 // is_committed is unused.
3343 return linux_mprotect(addr, bytes, p);
3344 }
3346 bool os::guard_memory(char* addr, size_t size) {
3347 return linux_mprotect(addr, size, PROT_NONE);
3348 }
3350 bool os::unguard_memory(char* addr, size_t size) {
3351 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3352 }
3354 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3355 bool result = false;
3356 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3357 MAP_ANONYMOUS|MAP_PRIVATE,
3358 -1, 0);
3359 if (p != MAP_FAILED) {
3360 void *aligned_p = align_ptr_up(p, page_size);
3362 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3364 munmap(p, page_size * 2);
3365 }
3367 if (warn && !result) {
3368 warning("TransparentHugePages is not supported by the operating system.");
3369 }
3371 return result;
3372 }
3374 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3375 bool result = false;
3376 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3377 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3378 -1, 0);
3380 if (p != MAP_FAILED) {
3381 // We don't know if this really is a huge page or not.
3382 FILE *fp = fopen("/proc/self/maps", "r");
3383 if (fp) {
3384 while (!feof(fp)) {
3385 char chars[257];
3386 long x = 0;
3387 if (fgets(chars, sizeof(chars), fp)) {
3388 if (sscanf(chars, "%lx-%*x", &x) == 1
3389 && x == (long)p) {
3390 if (strstr (chars, "hugepage")) {
3391 result = true;
3392 break;
3393 }
3394 }
3395 }
3396 }
3397 fclose(fp);
3398 }
3399 munmap(p, page_size);
3400 }
3402 if (warn && !result) {
3403 warning("HugeTLBFS is not supported by the operating system.");
3404 }
3406 return result;
3407 }
3409 /*
3410 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3411 *
3412 * From the coredump_filter documentation:
3413 *
3414 * - (bit 0) anonymous private memory
3415 * - (bit 1) anonymous shared memory
3416 * - (bit 2) file-backed private memory
3417 * - (bit 3) file-backed shared memory
3418 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3419 * effective only if the bit 2 is cleared)
3420 * - (bit 5) hugetlb private memory
3421 * - (bit 6) hugetlb shared memory
3422 */
3423 static void set_coredump_filter(void) {
3424 FILE *f;
3425 long cdm;
3427 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3428 return;
3429 }
3431 if (fscanf(f, "%lx", &cdm) != 1) {
3432 fclose(f);
3433 return;
3434 }
3436 rewind(f);
3438 if ((cdm & LARGEPAGES_BIT) == 0) {
3439 cdm |= LARGEPAGES_BIT;
3440 fprintf(f, "%#lx", cdm);
3441 }
3443 fclose(f);
3444 }
3446 // Large page support
3448 static size_t _large_page_size = 0;
3450 size_t os::Linux::find_large_page_size() {
3451 size_t large_page_size = 0;
3453 // large_page_size on Linux is used to round up heap size. x86 uses either
3454 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3455 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3456 // page as large as 256M.
3457 //
3458 // Here we try to figure out page size by parsing /proc/meminfo and looking
3459 // for a line with the following format:
3460 // Hugepagesize: 2048 kB
3461 //
3462 // If we can't determine the value (e.g. /proc is not mounted, or the text
3463 // format has been changed), we'll use the largest page size supported by
3464 // the processor.
3466 #ifndef ZERO
3467 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3468 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3469 #endif // ZERO
3471 FILE *fp = fopen("/proc/meminfo", "r");
3472 if (fp) {
3473 while (!feof(fp)) {
3474 int x = 0;
3475 char buf[16];
3476 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3477 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3478 large_page_size = x * K;
3479 break;
3480 }
3481 } else {
3482 // skip to next line
3483 for (;;) {
3484 int ch = fgetc(fp);
3485 if (ch == EOF || ch == (int)'\n') break;
3486 }
3487 }
3488 }
3489 fclose(fp);
3490 }
3492 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3493 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3494 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3495 proper_unit_for_byte_size(large_page_size));
3496 }
3498 return large_page_size;
3499 }
3501 size_t os::Linux::setup_large_page_size() {
3502 _large_page_size = Linux::find_large_page_size();
3503 const size_t default_page_size = (size_t)Linux::page_size();
3504 if (_large_page_size > default_page_size) {
3505 _page_sizes[0] = _large_page_size;
3506 _page_sizes[1] = default_page_size;
3507 _page_sizes[2] = 0;
3508 }
3510 return _large_page_size;
3511 }
3513 bool os::Linux::setup_large_page_type(size_t page_size) {
3514 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3515 FLAG_IS_DEFAULT(UseSHM) &&
3516 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3518 // The type of large pages has not been specified by the user.
3520 // Try UseHugeTLBFS and then UseSHM.
3521 UseHugeTLBFS = UseSHM = true;
3523 // Don't try UseTransparentHugePages since there are known
3524 // performance issues with it turned on. This might change in the future.
3525 UseTransparentHugePages = false;
3526 }
3528 if (UseTransparentHugePages) {
3529 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3530 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3531 UseHugeTLBFS = false;
3532 UseSHM = false;
3533 return true;
3534 }
3535 UseTransparentHugePages = false;
3536 }
3538 if (UseHugeTLBFS) {
3539 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3540 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3541 UseSHM = false;
3542 return true;
3543 }
3544 UseHugeTLBFS = false;
3545 }
3547 return UseSHM;
3548 }
3550 void os::large_page_init() {
3551 if (!UseLargePages &&
3552 !UseTransparentHugePages &&
3553 !UseHugeTLBFS &&
3554 !UseSHM) {
3555 // Not using large pages.
3556 return;
3557 }
3559 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3560 // The user explicitly turned off large pages.
3561 // Ignore the rest of the large pages flags.
3562 UseTransparentHugePages = false;
3563 UseHugeTLBFS = false;
3564 UseSHM = false;
3565 return;
3566 }
3568 size_t large_page_size = Linux::setup_large_page_size();
3569 UseLargePages = Linux::setup_large_page_type(large_page_size);
3571 set_coredump_filter();
3572 }
3574 #ifndef SHM_HUGETLB
3575 #define SHM_HUGETLB 04000
3576 #endif
3578 #define shm_warning_format(format, ...) \
3579 do { \
3580 if (UseLargePages && \
3581 (!FLAG_IS_DEFAULT(UseLargePages) || \
3582 !FLAG_IS_DEFAULT(UseSHM) || \
3583 !FLAG_IS_DEFAULT(LargePageSizeInBytes))) { \
3584 warning(format, __VA_ARGS__); \
3585 } \
3586 } while (0)
3588 #define shm_warning(str) shm_warning_format("%s", str)
3590 #define shm_warning_with_errno(str) \
3591 do { \
3592 int err = errno; \
3593 shm_warning_format(str " (error = %d)", err); \
3594 } while (0)
3596 static char* shmat_with_alignment(int shmid, size_t bytes, size_t alignment) {
3597 assert(is_size_aligned(bytes, alignment), "Must be divisible by the alignment");
3599 if (!is_size_aligned(alignment, SHMLBA)) {
3600 assert(false, "Code below assumes that alignment is at least SHMLBA aligned");
3601 return NULL;
3602 }
3604 // To ensure that we get 'alignment' aligned memory from shmat,
3605 // we pre-reserve aligned virtual memory and then attach to that.
3607 char* pre_reserved_addr = anon_mmap_aligned(bytes, alignment, NULL);
3608 if (pre_reserved_addr == NULL) {
3609 // Couldn't pre-reserve aligned memory.
3610 shm_warning("Failed to pre-reserve aligned memory for shmat.");
3611 return NULL;
3612 }
3614 // SHM_REMAP is needed to allow shmat to map over an existing mapping.
3615 char* addr = (char*)shmat(shmid, pre_reserved_addr, SHM_REMAP);
3617 if ((intptr_t)addr == -1) {
3618 int err = errno;
3619 shm_warning_with_errno("Failed to attach shared memory.");
3621 assert(err != EACCES, "Unexpected error");
3622 assert(err != EIDRM, "Unexpected error");
3623 assert(err != EINVAL, "Unexpected error");
3625 // Since we don't know if the kernel unmapped the pre-reserved memory area
3626 // we can't unmap it, since that would potentially unmap memory that was
3627 // mapped from other threads.
3628 return NULL;
3629 }
3631 return addr;
3632 }
3634 static char* shmat_at_address(int shmid, char* req_addr) {
3635 if (!is_ptr_aligned(req_addr, SHMLBA)) {
3636 assert(false, "Requested address needs to be SHMLBA aligned");
3637 return NULL;
3638 }
3640 char* addr = (char*)shmat(shmid, req_addr, 0);
3642 if ((intptr_t)addr == -1) {
3643 shm_warning_with_errno("Failed to attach shared memory.");
3644 return NULL;
3645 }
3647 return addr;
3648 }
3650 static char* shmat_large_pages(int shmid, size_t bytes, size_t alignment, char* req_addr) {
3651 // If a req_addr has been provided, we assume that the caller has already aligned the address.
3652 if (req_addr != NULL) {
3653 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Must be divisible by the large page size");
3654 assert(is_ptr_aligned(req_addr, alignment), "Must be divisible by given alignment");
3655 return shmat_at_address(shmid, req_addr);
3656 }
3658 // Since shmid has been setup with SHM_HUGETLB, shmat will automatically
3659 // return large page size aligned memory addresses when req_addr == NULL.
3660 // However, if the alignment is larger than the large page size, we have
3661 // to manually ensure that the memory returned is 'alignment' aligned.
3662 if (alignment > os::large_page_size()) {
3663 assert(is_size_aligned(alignment, os::large_page_size()), "Must be divisible by the large page size");
3664 return shmat_with_alignment(shmid, bytes, alignment);
3665 } else {
3666 return shmat_at_address(shmid, NULL);
3667 }
3668 }
3670 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3671 // "exec" is passed in but not used. Creating the shared image for
3672 // the code cache doesn't have an SHM_X executable permission to check.
3673 assert(UseLargePages && UseSHM, "only for SHM large pages");
3674 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3675 assert(is_ptr_aligned(req_addr, alignment), "Unaligned address");
3677 if (!is_size_aligned(bytes, os::large_page_size())) {
3678 return NULL; // Fallback to small pages.
3679 }
3681 // Create a large shared memory region to attach to based on size.
3682 // Currently, size is the total size of the heap.
3683 int shmid = shmget(IPC_PRIVATE, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3684 if (shmid == -1) {
3685 // Possible reasons for shmget failure:
3686 // 1. shmmax is too small for Java heap.
3687 // > check shmmax value: cat /proc/sys/kernel/shmmax
3688 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3689 // 2. not enough large page memory.
3690 // > check available large pages: cat /proc/meminfo
3691 // > increase amount of large pages:
3692 // echo new_value > /proc/sys/vm/nr_hugepages
3693 // Note 1: different Linux may use different name for this property,
3694 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3695 // Note 2: it's possible there's enough physical memory available but
3696 // they are so fragmented after a long run that they can't
3697 // coalesce into large pages. Try to reserve large pages when
3698 // the system is still "fresh".
3699 shm_warning_with_errno("Failed to reserve shared memory.");
3700 return NULL;
3701 }
3703 // Attach to the region.
3704 char* addr = shmat_large_pages(shmid, bytes, alignment, req_addr);
3706 // Remove shmid. If shmat() is successful, the actual shared memory segment
3707 // will be deleted when it's detached by shmdt() or when the process
3708 // terminates. If shmat() is not successful this will remove the shared
3709 // segment immediately.
3710 shmctl(shmid, IPC_RMID, NULL);
3712 return addr;
3713 }
3715 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3716 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3718 bool warn_on_failure = UseLargePages &&
3719 (!FLAG_IS_DEFAULT(UseLargePages) ||
3720 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3721 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3723 if (warn_on_failure) {
3724 char msg[128];
3725 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3726 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3727 warning("%s", msg);
3728 }
3729 }
3731 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3732 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3733 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3734 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3736 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3737 char* addr = (char*)::mmap(req_addr, bytes, prot,
3738 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3739 -1, 0);
3741 if (addr == MAP_FAILED) {
3742 warn_on_large_pages_failure(req_addr, bytes, errno);
3743 return NULL;
3744 }
3746 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3748 return addr;
3749 }
3751 // Reserve memory using mmap(MAP_HUGETLB).
3752 // - bytes shall be a multiple of alignment.
3753 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3754 // - alignment sets the alignment at which memory shall be allocated.
3755 // It must be a multiple of allocation granularity.
3756 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3757 // req_addr or NULL.
3758 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3759 size_t large_page_size = os::large_page_size();
3760 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3762 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3763 assert(is_size_aligned(bytes, alignment), "Must be");
3765 // First reserve - but not commit - the address range in small pages.
3766 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3768 if (start == NULL) {
3769 return NULL;
3770 }
3772 assert(is_ptr_aligned(start, alignment), "Must be");
3774 char* end = start + bytes;
3776 // Find the regions of the allocated chunk that can be promoted to large pages.
3777 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3778 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3780 size_t lp_bytes = lp_end - lp_start;
3782 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3784 if (lp_bytes == 0) {
3785 // The mapped region doesn't even span the start and the end of a large page.
3786 // Fall back to allocate a non-special area.
3787 ::munmap(start, end - start);
3788 return NULL;
3789 }
3791 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3793 void* result;
3795 // Commit small-paged leading area.
3796 if (start != lp_start) {
3797 result = ::mmap(start, lp_start - start, prot,
3798 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3799 -1, 0);
3800 if (result == MAP_FAILED) {
3801 ::munmap(lp_start, end - lp_start);
3802 return NULL;
3803 }
3804 }
3806 // Commit large-paged area.
3807 result = ::mmap(lp_start, lp_bytes, prot,
3808 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3809 -1, 0);
3810 if (result == MAP_FAILED) {
3811 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3812 // If the mmap above fails, the large pages region will be unmapped and we
3813 // have regions before and after with small pages. Release these regions.
3814 //
3815 // | mapped | unmapped | mapped |
3816 // ^ ^ ^ ^
3817 // start lp_start lp_end end
3818 //
3819 ::munmap(start, lp_start - start);
3820 ::munmap(lp_end, end - lp_end);
3821 return NULL;
3822 }
3824 // Commit small-paged trailing area.
3825 if (lp_end != end) {
3826 result = ::mmap(lp_end, end - lp_end, prot,
3827 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3828 -1, 0);
3829 if (result == MAP_FAILED) {
3830 ::munmap(start, lp_end - start);
3831 return NULL;
3832 }
3833 }
3835 return start;
3836 }
3838 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3839 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3840 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3841 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3842 assert(is_power_of_2(os::large_page_size()), "Must be");
3843 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3845 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3846 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3847 } else {
3848 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3849 }
3850 }
3852 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3853 assert(UseLargePages, "only for large pages");
3855 char* addr;
3856 if (UseSHM) {
3857 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3858 } else {
3859 assert(UseHugeTLBFS, "must be");
3860 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3861 }
3863 if (addr != NULL) {
3864 if (UseNUMAInterleaving) {
3865 numa_make_global(addr, bytes);
3866 }
3868 // The memory is committed
3869 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3870 }
3872 return addr;
3873 }
3875 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3876 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3877 return shmdt(base) == 0;
3878 }
3880 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3881 return pd_release_memory(base, bytes);
3882 }
3884 bool os::release_memory_special(char* base, size_t bytes) {
3885 bool res;
3886 if (MemTracker::tracking_level() > NMT_minimal) {
3887 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3888 res = os::Linux::release_memory_special_impl(base, bytes);
3889 if (res) {
3890 tkr.record((address)base, bytes);
3891 }
3893 } else {
3894 res = os::Linux::release_memory_special_impl(base, bytes);
3895 }
3896 return res;
3897 }
3899 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3900 assert(UseLargePages, "only for large pages");
3901 bool res;
3903 if (UseSHM) {
3904 res = os::Linux::release_memory_special_shm(base, bytes);
3905 } else {
3906 assert(UseHugeTLBFS, "must be");
3907 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3908 }
3909 return res;
3910 }
3912 size_t os::large_page_size() {
3913 return _large_page_size;
3914 }
3916 // With SysV SHM the entire memory region must be allocated as shared
3917 // memory.
3918 // HugeTLBFS allows application to commit large page memory on demand.
3919 // However, when committing memory with HugeTLBFS fails, the region
3920 // that was supposed to be committed will lose the old reservation
3921 // and allow other threads to steal that memory region. Because of this
3922 // behavior we can't commit HugeTLBFS memory.
3923 bool os::can_commit_large_page_memory() {
3924 return UseTransparentHugePages;
3925 }
3927 bool os::can_execute_large_page_memory() {
3928 return UseTransparentHugePages || UseHugeTLBFS;
3929 }
3931 // Reserve memory at an arbitrary address, only if that area is
3932 // available (and not reserved for something else).
3934 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3935 const int max_tries = 10;
3936 char* base[max_tries];
3937 size_t size[max_tries];
3938 const size_t gap = 0x000000;
3940 // Assert only that the size is a multiple of the page size, since
3941 // that's all that mmap requires, and since that's all we really know
3942 // about at this low abstraction level. If we need higher alignment,
3943 // we can either pass an alignment to this method or verify alignment
3944 // in one of the methods further up the call chain. See bug 5044738.
3945 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3947 // Repeatedly allocate blocks until the block is allocated at the
3948 // right spot. Give up after max_tries. Note that reserve_memory() will
3949 // automatically update _highest_vm_reserved_address if the call is
3950 // successful. The variable tracks the highest memory address every reserved
3951 // by JVM. It is used to detect heap-stack collision if running with
3952 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3953 // space than needed, it could confuse the collision detecting code. To
3954 // solve the problem, save current _highest_vm_reserved_address and
3955 // calculate the correct value before return.
3956 address old_highest = _highest_vm_reserved_address;
3958 // Linux mmap allows caller to pass an address as hint; give it a try first,
3959 // if kernel honors the hint then we can return immediately.
3960 char * addr = anon_mmap(requested_addr, bytes, false);
3961 if (addr == requested_addr) {
3962 return requested_addr;
3963 }
3965 if (addr != NULL) {
3966 // mmap() is successful but it fails to reserve at the requested address
3967 anon_munmap(addr, bytes);
3968 }
3970 int i;
3971 for (i = 0; i < max_tries; ++i) {
3972 base[i] = reserve_memory(bytes);
3974 if (base[i] != NULL) {
3975 // Is this the block we wanted?
3976 if (base[i] == requested_addr) {
3977 size[i] = bytes;
3978 break;
3979 }
3981 // Does this overlap the block we wanted? Give back the overlapped
3982 // parts and try again.
3984 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3985 if (top_overlap >= 0 && top_overlap < bytes) {
3986 unmap_memory(base[i], top_overlap);
3987 base[i] += top_overlap;
3988 size[i] = bytes - top_overlap;
3989 } else {
3990 size_t bottom_overlap = base[i] + bytes - requested_addr;
3991 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3992 unmap_memory(requested_addr, bottom_overlap);
3993 size[i] = bytes - bottom_overlap;
3994 } else {
3995 size[i] = bytes;
3996 }
3997 }
3998 }
3999 }
4001 // Give back the unused reserved pieces.
4003 for (int j = 0; j < i; ++j) {
4004 if (base[j] != NULL) {
4005 unmap_memory(base[j], size[j]);
4006 }
4007 }
4009 if (i < max_tries) {
4010 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
4011 return requested_addr;
4012 } else {
4013 _highest_vm_reserved_address = old_highest;
4014 return NULL;
4015 }
4016 }
4018 size_t os::read(int fd, void *buf, unsigned int nBytes) {
4019 return ::read(fd, buf, nBytes);
4020 }
4022 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
4023 // Solaris uses poll(), linux uses park().
4024 // Poll() is likely a better choice, assuming that Thread.interrupt()
4025 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
4026 // SIGSEGV, see 4355769.
4028 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
4029 assert(thread == Thread::current(), "thread consistency check");
4031 ParkEvent * const slp = thread->_SleepEvent ;
4032 slp->reset() ;
4033 OrderAccess::fence() ;
4035 if (interruptible) {
4036 jlong prevtime = javaTimeNanos();
4038 for (;;) {
4039 if (os::is_interrupted(thread, true)) {
4040 return OS_INTRPT;
4041 }
4043 jlong newtime = javaTimeNanos();
4045 if (newtime - prevtime < 0) {
4046 // time moving backwards, should only happen if no monotonic clock
4047 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4048 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4049 } else {
4050 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4051 }
4053 if(millis <= 0) {
4054 return OS_OK;
4055 }
4057 prevtime = newtime;
4059 {
4060 assert(thread->is_Java_thread(), "sanity check");
4061 JavaThread *jt = (JavaThread *) thread;
4062 ThreadBlockInVM tbivm(jt);
4063 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
4065 jt->set_suspend_equivalent();
4066 // cleared by handle_special_suspend_equivalent_condition() or
4067 // java_suspend_self() via check_and_wait_while_suspended()
4069 slp->park(millis);
4071 // were we externally suspended while we were waiting?
4072 jt->check_and_wait_while_suspended();
4073 }
4074 }
4075 } else {
4076 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
4077 jlong prevtime = javaTimeNanos();
4079 for (;;) {
4080 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
4081 // the 1st iteration ...
4082 jlong newtime = javaTimeNanos();
4084 if (newtime - prevtime < 0) {
4085 // time moving backwards, should only happen if no monotonic clock
4086 // not a guarantee() because JVM should not abort on kernel/glibc bugs
4087 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
4088 } else {
4089 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
4090 }
4092 if(millis <= 0) break ;
4094 prevtime = newtime;
4095 slp->park(millis);
4096 }
4097 return OS_OK ;
4098 }
4099 }
4101 //
4102 // Short sleep, direct OS call.
4103 //
4104 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
4105 // sched_yield(2) will actually give up the CPU:
4106 //
4107 // * Alone on this pariticular CPU, keeps running.
4108 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
4109 // (pre 2.6.39).
4110 //
4111 // So calling this with 0 is an alternative.
4112 //
4113 void os::naked_short_sleep(jlong ms) {
4114 struct timespec req;
4116 assert(ms < 1000, "Un-interruptable sleep, short time use only");
4117 req.tv_sec = 0;
4118 if (ms > 0) {
4119 req.tv_nsec = (ms % 1000) * 1000000;
4120 }
4121 else {
4122 req.tv_nsec = 1;
4123 }
4125 nanosleep(&req, NULL);
4127 return;
4128 }
4130 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
4131 void os::infinite_sleep() {
4132 while (true) { // sleep forever ...
4133 ::sleep(100); // ... 100 seconds at a time
4134 }
4135 }
4137 // Used to convert frequent JVM_Yield() to nops
4138 bool os::dont_yield() {
4139 return DontYieldALot;
4140 }
4142 void os::yield() {
4143 sched_yield();
4144 }
4146 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
4148 void os::yield_all(int attempts) {
4149 // Yields to all threads, including threads with lower priorities
4150 // Threads on Linux are all with same priority. The Solaris style
4151 // os::yield_all() with nanosleep(1ms) is not necessary.
4152 sched_yield();
4153 }
4155 // Called from the tight loops to possibly influence time-sharing heuristics
4156 void os::loop_breaker(int attempts) {
4157 os::yield_all(attempts);
4158 }
4160 ////////////////////////////////////////////////////////////////////////////////
4161 // thread priority support
4163 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
4164 // only supports dynamic priority, static priority must be zero. For real-time
4165 // applications, Linux supports SCHED_RR which allows static priority (1-99).
4166 // However, for large multi-threaded applications, SCHED_RR is not only slower
4167 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
4168 // of 5 runs - Sep 2005).
4169 //
4170 // The following code actually changes the niceness of kernel-thread/LWP. It
4171 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
4172 // not the entire user process, and user level threads are 1:1 mapped to kernel
4173 // threads. It has always been the case, but could change in the future. For
4174 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
4175 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
4177 int os::java_to_os_priority[CriticalPriority + 1] = {
4178 19, // 0 Entry should never be used
4180 4, // 1 MinPriority
4181 3, // 2
4182 2, // 3
4184 1, // 4
4185 0, // 5 NormPriority
4186 -1, // 6
4188 -2, // 7
4189 -3, // 8
4190 -4, // 9 NearMaxPriority
4192 -5, // 10 MaxPriority
4194 -5 // 11 CriticalPriority
4195 };
4197 static int prio_init() {
4198 if (ThreadPriorityPolicy == 1) {
4199 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
4200 // if effective uid is not root. Perhaps, a more elegant way of doing
4201 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
4202 if (geteuid() != 0) {
4203 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
4204 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
4205 }
4206 ThreadPriorityPolicy = 0;
4207 }
4208 }
4209 if (UseCriticalJavaThreadPriority) {
4210 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
4211 }
4212 return 0;
4213 }
4215 OSReturn os::set_native_priority(Thread* thread, int newpri) {
4216 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
4218 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
4219 return (ret == 0) ? OS_OK : OS_ERR;
4220 }
4222 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
4223 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
4224 *priority_ptr = java_to_os_priority[NormPriority];
4225 return OS_OK;
4226 }
4228 errno = 0;
4229 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
4230 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
4231 }
4233 // Hint to the underlying OS that a task switch would not be good.
4234 // Void return because it's a hint and can fail.
4235 void os::hint_no_preempt() {}
4237 ////////////////////////////////////////////////////////////////////////////////
4238 // suspend/resume support
4240 // the low-level signal-based suspend/resume support is a remnant from the
4241 // old VM-suspension that used to be for java-suspension, safepoints etc,
4242 // within hotspot. Now there is a single use-case for this:
4243 // - calling get_thread_pc() on the VMThread by the flat-profiler task
4244 // that runs in the watcher thread.
4245 // The remaining code is greatly simplified from the more general suspension
4246 // code that used to be used.
4247 //
4248 // The protocol is quite simple:
4249 // - suspend:
4250 // - sends a signal to the target thread
4251 // - polls the suspend state of the osthread using a yield loop
4252 // - target thread signal handler (SR_handler) sets suspend state
4253 // and blocks in sigsuspend until continued
4254 // - resume:
4255 // - sets target osthread state to continue
4256 // - sends signal to end the sigsuspend loop in the SR_handler
4257 //
4258 // Note that the SR_lock plays no role in this suspend/resume protocol.
4259 //
4261 static void resume_clear_context(OSThread *osthread) {
4262 osthread->set_ucontext(NULL);
4263 osthread->set_siginfo(NULL);
4264 }
4266 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
4267 osthread->set_ucontext(context);
4268 osthread->set_siginfo(siginfo);
4269 }
4271 //
4272 // Handler function invoked when a thread's execution is suspended or
4273 // resumed. We have to be careful that only async-safe functions are
4274 // called here (Note: most pthread functions are not async safe and
4275 // should be avoided.)
4276 //
4277 // Note: sigwait() is a more natural fit than sigsuspend() from an
4278 // interface point of view, but sigwait() prevents the signal hander
4279 // from being run. libpthread would get very confused by not having
4280 // its signal handlers run and prevents sigwait()'s use with the
4281 // mutex granting granting signal.
4282 //
4283 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4284 //
4285 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4286 // Save and restore errno to avoid confusing native code with EINTR
4287 // after sigsuspend.
4288 int old_errno = errno;
4290 Thread* thread = Thread::current();
4291 OSThread* osthread = thread->osthread();
4292 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4294 os::SuspendResume::State current = osthread->sr.state();
4295 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4296 suspend_save_context(osthread, siginfo, context);
4298 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4299 os::SuspendResume::State state = osthread->sr.suspended();
4300 if (state == os::SuspendResume::SR_SUSPENDED) {
4301 sigset_t suspend_set; // signals for sigsuspend()
4303 // get current set of blocked signals and unblock resume signal
4304 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4305 sigdelset(&suspend_set, SR_signum);
4307 sr_semaphore.signal();
4308 // wait here until we are resumed
4309 while (1) {
4310 sigsuspend(&suspend_set);
4312 os::SuspendResume::State result = osthread->sr.running();
4313 if (result == os::SuspendResume::SR_RUNNING) {
4314 sr_semaphore.signal();
4315 break;
4316 }
4317 }
4319 } else if (state == os::SuspendResume::SR_RUNNING) {
4320 // request was cancelled, continue
4321 } else {
4322 ShouldNotReachHere();
4323 }
4325 resume_clear_context(osthread);
4326 } else if (current == os::SuspendResume::SR_RUNNING) {
4327 // request was cancelled, continue
4328 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4329 // ignore
4330 } else {
4331 // ignore
4332 }
4334 errno = old_errno;
4335 }
4338 static int SR_initialize() {
4339 struct sigaction act;
4340 char *s;
4341 /* Get signal number to use for suspend/resume */
4342 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4343 int sig = ::strtol(s, 0, 10);
4344 if (sig > 0 || sig < _NSIG) {
4345 SR_signum = sig;
4346 }
4347 }
4349 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4350 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4352 sigemptyset(&SR_sigset);
4353 sigaddset(&SR_sigset, SR_signum);
4355 /* Set up signal handler for suspend/resume */
4356 act.sa_flags = SA_RESTART|SA_SIGINFO;
4357 act.sa_handler = (void (*)(int)) SR_handler;
4359 // SR_signum is blocked by default.
4360 // 4528190 - We also need to block pthread restart signal (32 on all
4361 // supported Linux platforms). Note that LinuxThreads need to block
4362 // this signal for all threads to work properly. So we don't have
4363 // to use hard-coded signal number when setting up the mask.
4364 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4366 if (sigaction(SR_signum, &act, 0) == -1) {
4367 return -1;
4368 }
4370 // Save signal flag
4371 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4372 return 0;
4373 }
4375 static int sr_notify(OSThread* osthread) {
4376 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4377 assert_status(status == 0, status, "pthread_kill");
4378 return status;
4379 }
4381 // "Randomly" selected value for how long we want to spin
4382 // before bailing out on suspending a thread, also how often
4383 // we send a signal to a thread we want to resume
4384 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4385 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4387 // returns true on success and false on error - really an error is fatal
4388 // but this seems the normal response to library errors
4389 static bool do_suspend(OSThread* osthread) {
4390 assert(osthread->sr.is_running(), "thread should be running");
4391 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4393 // mark as suspended and send signal
4394 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4395 // failed to switch, state wasn't running?
4396 ShouldNotReachHere();
4397 return false;
4398 }
4400 if (sr_notify(osthread) != 0) {
4401 ShouldNotReachHere();
4402 }
4404 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4405 while (true) {
4406 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4407 break;
4408 } else {
4409 // timeout
4410 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4411 if (cancelled == os::SuspendResume::SR_RUNNING) {
4412 return false;
4413 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4414 // make sure that we consume the signal on the semaphore as well
4415 sr_semaphore.wait();
4416 break;
4417 } else {
4418 ShouldNotReachHere();
4419 return false;
4420 }
4421 }
4422 }
4424 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4425 return true;
4426 }
4428 static void do_resume(OSThread* osthread) {
4429 assert(osthread->sr.is_suspended(), "thread should be suspended");
4430 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4432 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4433 // failed to switch to WAKEUP_REQUEST
4434 ShouldNotReachHere();
4435 return;
4436 }
4438 while (true) {
4439 if (sr_notify(osthread) == 0) {
4440 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4441 if (osthread->sr.is_running()) {
4442 return;
4443 }
4444 }
4445 } else {
4446 ShouldNotReachHere();
4447 }
4448 }
4450 guarantee(osthread->sr.is_running(), "Must be running!");
4451 }
4453 ////////////////////////////////////////////////////////////////////////////////
4454 // interrupt support
4456 void os::interrupt(Thread* thread) {
4457 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4458 "possibility of dangling Thread pointer");
4460 OSThread* osthread = thread->osthread();
4462 if (!osthread->interrupted()) {
4463 osthread->set_interrupted(true);
4464 // More than one thread can get here with the same value of osthread,
4465 // resulting in multiple notifications. We do, however, want the store
4466 // to interrupted() to be visible to other threads before we execute unpark().
4467 OrderAccess::fence();
4468 ParkEvent * const slp = thread->_SleepEvent ;
4469 if (slp != NULL) slp->unpark() ;
4470 }
4472 // For JSR166. Unpark even if interrupt status already was set
4473 if (thread->is_Java_thread())
4474 ((JavaThread*)thread)->parker()->unpark();
4476 ParkEvent * ev = thread->_ParkEvent ;
4477 if (ev != NULL) ev->unpark() ;
4479 }
4481 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4482 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4483 "possibility of dangling Thread pointer");
4485 OSThread* osthread = thread->osthread();
4487 bool interrupted = osthread->interrupted();
4489 if (interrupted && clear_interrupted) {
4490 osthread->set_interrupted(false);
4491 // consider thread->_SleepEvent->reset() ... optional optimization
4492 }
4494 return interrupted;
4495 }
4497 ///////////////////////////////////////////////////////////////////////////////////
4498 // signal handling (except suspend/resume)
4500 // This routine may be used by user applications as a "hook" to catch signals.
4501 // The user-defined signal handler must pass unrecognized signals to this
4502 // routine, and if it returns true (non-zero), then the signal handler must
4503 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4504 // routine will never retun false (zero), but instead will execute a VM panic
4505 // routine kill the process.
4506 //
4507 // If this routine returns false, it is OK to call it again. This allows
4508 // the user-defined signal handler to perform checks either before or after
4509 // the VM performs its own checks. Naturally, the user code would be making
4510 // a serious error if it tried to handle an exception (such as a null check
4511 // or breakpoint) that the VM was generating for its own correct operation.
4512 //
4513 // This routine may recognize any of the following kinds of signals:
4514 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4515 // It should be consulted by handlers for any of those signals.
4516 //
4517 // The caller of this routine must pass in the three arguments supplied
4518 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4519 // field of the structure passed to sigaction(). This routine assumes that
4520 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4521 //
4522 // Note that the VM will print warnings if it detects conflicting signal
4523 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4524 //
4525 extern "C" JNIEXPORT int
4526 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4527 void* ucontext, int abort_if_unrecognized);
4529 void signalHandler(int sig, siginfo_t* info, void* uc) {
4530 assert(info != NULL && uc != NULL, "it must be old kernel");
4531 int orig_errno = errno; // Preserve errno value over signal handler.
4532 JVM_handle_linux_signal(sig, info, uc, true);
4533 errno = orig_errno;
4534 }
4537 // This boolean allows users to forward their own non-matching signals
4538 // to JVM_handle_linux_signal, harmlessly.
4539 bool os::Linux::signal_handlers_are_installed = false;
4541 // For signal-chaining
4542 struct sigaction os::Linux::sigact[MAXSIGNUM];
4543 unsigned int os::Linux::sigs = 0;
4544 bool os::Linux::libjsig_is_loaded = false;
4545 typedef struct sigaction *(*get_signal_t)(int);
4546 get_signal_t os::Linux::get_signal_action = NULL;
4548 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4549 struct sigaction *actp = NULL;
4551 if (libjsig_is_loaded) {
4552 // Retrieve the old signal handler from libjsig
4553 actp = (*get_signal_action)(sig);
4554 }
4555 if (actp == NULL) {
4556 // Retrieve the preinstalled signal handler from jvm
4557 actp = get_preinstalled_handler(sig);
4558 }
4560 return actp;
4561 }
4563 static bool call_chained_handler(struct sigaction *actp, int sig,
4564 siginfo_t *siginfo, void *context) {
4565 // Call the old signal handler
4566 if (actp->sa_handler == SIG_DFL) {
4567 // It's more reasonable to let jvm treat it as an unexpected exception
4568 // instead of taking the default action.
4569 return false;
4570 } else if (actp->sa_handler != SIG_IGN) {
4571 if ((actp->sa_flags & SA_NODEFER) == 0) {
4572 // automaticlly block the signal
4573 sigaddset(&(actp->sa_mask), sig);
4574 }
4576 sa_handler_t hand = NULL;
4577 sa_sigaction_t sa = NULL;
4578 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4579 // retrieve the chained handler
4580 if (siginfo_flag_set) {
4581 sa = actp->sa_sigaction;
4582 } else {
4583 hand = actp->sa_handler;
4584 }
4586 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4587 actp->sa_handler = SIG_DFL;
4588 }
4590 // try to honor the signal mask
4591 sigset_t oset;
4592 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4594 // call into the chained handler
4595 if (siginfo_flag_set) {
4596 (*sa)(sig, siginfo, context);
4597 } else {
4598 (*hand)(sig);
4599 }
4601 // restore the signal mask
4602 pthread_sigmask(SIG_SETMASK, &oset, 0);
4603 }
4604 // Tell jvm's signal handler the signal is taken care of.
4605 return true;
4606 }
4608 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4609 bool chained = false;
4610 // signal-chaining
4611 if (UseSignalChaining) {
4612 struct sigaction *actp = get_chained_signal_action(sig);
4613 if (actp != NULL) {
4614 chained = call_chained_handler(actp, sig, siginfo, context);
4615 }
4616 }
4617 return chained;
4618 }
4620 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4621 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4622 return &sigact[sig];
4623 }
4624 return NULL;
4625 }
4627 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4628 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4629 sigact[sig] = oldAct;
4630 sigs |= (unsigned int)1 << sig;
4631 }
4633 // for diagnostic
4634 int os::Linux::sigflags[MAXSIGNUM];
4636 int os::Linux::get_our_sigflags(int sig) {
4637 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4638 return sigflags[sig];
4639 }
4641 void os::Linux::set_our_sigflags(int sig, int flags) {
4642 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4643 sigflags[sig] = flags;
4644 }
4646 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4647 // Check for overwrite.
4648 struct sigaction oldAct;
4649 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4651 void* oldhand = oldAct.sa_sigaction
4652 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4653 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4654 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4655 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4656 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4657 if (AllowUserSignalHandlers || !set_installed) {
4658 // Do not overwrite; user takes responsibility to forward to us.
4659 return;
4660 } else if (UseSignalChaining) {
4661 // save the old handler in jvm
4662 save_preinstalled_handler(sig, oldAct);
4663 // libjsig also interposes the sigaction() call below and saves the
4664 // old sigaction on it own.
4665 } else {
4666 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4667 "%#lx for signal %d.", (long)oldhand, sig));
4668 }
4669 }
4671 struct sigaction sigAct;
4672 sigfillset(&(sigAct.sa_mask));
4673 sigAct.sa_handler = SIG_DFL;
4674 if (!set_installed) {
4675 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4676 } else {
4677 sigAct.sa_sigaction = signalHandler;
4678 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4679 }
4680 // Save flags, which are set by ours
4681 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4682 sigflags[sig] = sigAct.sa_flags;
4684 int ret = sigaction(sig, &sigAct, &oldAct);
4685 assert(ret == 0, "check");
4687 void* oldhand2 = oldAct.sa_sigaction
4688 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4689 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4690 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4691 }
4693 // install signal handlers for signals that HotSpot needs to
4694 // handle in order to support Java-level exception handling.
4696 void os::Linux::install_signal_handlers() {
4697 if (!signal_handlers_are_installed) {
4698 signal_handlers_are_installed = true;
4700 // signal-chaining
4701 typedef void (*signal_setting_t)();
4702 signal_setting_t begin_signal_setting = NULL;
4703 signal_setting_t end_signal_setting = NULL;
4704 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4705 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4706 if (begin_signal_setting != NULL) {
4707 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4708 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4709 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4710 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4711 libjsig_is_loaded = true;
4712 assert(UseSignalChaining, "should enable signal-chaining");
4713 }
4714 if (libjsig_is_loaded) {
4715 // Tell libjsig jvm is setting signal handlers
4716 (*begin_signal_setting)();
4717 }
4719 set_signal_handler(SIGSEGV, true);
4720 set_signal_handler(SIGPIPE, true);
4721 set_signal_handler(SIGBUS, true);
4722 set_signal_handler(SIGILL, true);
4723 set_signal_handler(SIGFPE, true);
4724 #if defined(PPC64)
4725 set_signal_handler(SIGTRAP, true);
4726 #endif
4727 set_signal_handler(SIGXFSZ, true);
4729 if (libjsig_is_loaded) {
4730 // Tell libjsig jvm finishes setting signal handlers
4731 (*end_signal_setting)();
4732 }
4734 // We don't activate signal checker if libjsig is in place, we trust ourselves
4735 // and if UserSignalHandler is installed all bets are off.
4736 // Log that signal checking is off only if -verbose:jni is specified.
4737 if (CheckJNICalls) {
4738 if (libjsig_is_loaded) {
4739 if (PrintJNIResolving) {
4740 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4741 }
4742 check_signals = false;
4743 }
4744 if (AllowUserSignalHandlers) {
4745 if (PrintJNIResolving) {
4746 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4747 }
4748 check_signals = false;
4749 }
4750 }
4751 }
4752 }
4754 // This is the fastest way to get thread cpu time on Linux.
4755 // Returns cpu time (user+sys) for any thread, not only for current.
4756 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4757 // It might work on 2.6.10+ with a special kernel/glibc patch.
4758 // For reference, please, see IEEE Std 1003.1-2004:
4759 // http://www.unix.org/single_unix_specification
4761 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4762 struct timespec tp;
4763 int rc = os::Linux::clock_gettime(clockid, &tp);
4764 assert(rc == 0, "clock_gettime is expected to return 0 code");
4766 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4767 }
4769 /////
4770 // glibc on Linux platform uses non-documented flag
4771 // to indicate, that some special sort of signal
4772 // trampoline is used.
4773 // We will never set this flag, and we should
4774 // ignore this flag in our diagnostic
4775 #ifdef SIGNIFICANT_SIGNAL_MASK
4776 #undef SIGNIFICANT_SIGNAL_MASK
4777 #endif
4778 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4780 static const char* get_signal_handler_name(address handler,
4781 char* buf, int buflen) {
4782 int offset = 0;
4783 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4784 if (found) {
4785 // skip directory names
4786 const char *p1, *p2;
4787 p1 = buf;
4788 size_t len = strlen(os::file_separator());
4789 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4790 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4791 } else {
4792 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4793 }
4794 return buf;
4795 }
4797 static void print_signal_handler(outputStream* st, int sig,
4798 char* buf, size_t buflen) {
4799 struct sigaction sa;
4801 sigaction(sig, NULL, &sa);
4803 // See comment for SIGNIFICANT_SIGNAL_MASK define
4804 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4806 st->print("%s: ", os::exception_name(sig, buf, buflen));
4808 address handler = (sa.sa_flags & SA_SIGINFO)
4809 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4810 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4812 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4813 st->print("SIG_DFL");
4814 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4815 st->print("SIG_IGN");
4816 } else {
4817 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4818 }
4820 st->print(", sa_mask[0]=");
4821 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4823 address rh = VMError::get_resetted_sighandler(sig);
4824 // May be, handler was resetted by VMError?
4825 if(rh != NULL) {
4826 handler = rh;
4827 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4828 }
4830 st->print(", sa_flags=");
4831 os::Posix::print_sa_flags(st, sa.sa_flags);
4833 // Check: is it our handler?
4834 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4835 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4836 // It is our signal handler
4837 // check for flags, reset system-used one!
4838 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4839 st->print(
4840 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4841 os::Linux::get_our_sigflags(sig));
4842 }
4843 }
4844 st->cr();
4845 }
4848 #define DO_SIGNAL_CHECK(sig) \
4849 if (!sigismember(&check_signal_done, sig)) \
4850 os::Linux::check_signal_handler(sig)
4852 // This method is a periodic task to check for misbehaving JNI applications
4853 // under CheckJNI, we can add any periodic checks here
4855 void os::run_periodic_checks() {
4857 if (check_signals == false) return;
4859 // SEGV and BUS if overridden could potentially prevent
4860 // generation of hs*.log in the event of a crash, debugging
4861 // such a case can be very challenging, so we absolutely
4862 // check the following for a good measure:
4863 DO_SIGNAL_CHECK(SIGSEGV);
4864 DO_SIGNAL_CHECK(SIGILL);
4865 DO_SIGNAL_CHECK(SIGFPE);
4866 DO_SIGNAL_CHECK(SIGBUS);
4867 DO_SIGNAL_CHECK(SIGPIPE);
4868 DO_SIGNAL_CHECK(SIGXFSZ);
4869 #if defined(PPC64)
4870 DO_SIGNAL_CHECK(SIGTRAP);
4871 #endif
4873 // ReduceSignalUsage allows the user to override these handlers
4874 // see comments at the very top and jvm_solaris.h
4875 if (!ReduceSignalUsage) {
4876 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4877 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4878 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4879 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4880 }
4882 DO_SIGNAL_CHECK(SR_signum);
4883 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4884 }
4886 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4888 static os_sigaction_t os_sigaction = NULL;
4890 void os::Linux::check_signal_handler(int sig) {
4891 char buf[O_BUFLEN];
4892 address jvmHandler = NULL;
4895 struct sigaction act;
4896 if (os_sigaction == NULL) {
4897 // only trust the default sigaction, in case it has been interposed
4898 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4899 if (os_sigaction == NULL) return;
4900 }
4902 os_sigaction(sig, (struct sigaction*)NULL, &act);
4905 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4907 address thisHandler = (act.sa_flags & SA_SIGINFO)
4908 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4909 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4912 switch(sig) {
4913 case SIGSEGV:
4914 case SIGBUS:
4915 case SIGFPE:
4916 case SIGPIPE:
4917 case SIGILL:
4918 case SIGXFSZ:
4919 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4920 break;
4922 case SHUTDOWN1_SIGNAL:
4923 case SHUTDOWN2_SIGNAL:
4924 case SHUTDOWN3_SIGNAL:
4925 case BREAK_SIGNAL:
4926 jvmHandler = (address)user_handler();
4927 break;
4929 case INTERRUPT_SIGNAL:
4930 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4931 break;
4933 default:
4934 if (sig == SR_signum) {
4935 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4936 } else {
4937 return;
4938 }
4939 break;
4940 }
4942 if (thisHandler != jvmHandler) {
4943 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4944 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4945 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4946 // No need to check this sig any longer
4947 sigaddset(&check_signal_done, sig);
4948 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4949 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4950 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4951 exception_name(sig, buf, O_BUFLEN));
4952 }
4953 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4954 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4955 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4956 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4957 // No need to check this sig any longer
4958 sigaddset(&check_signal_done, sig);
4959 }
4961 // Dump all the signal
4962 if (sigismember(&check_signal_done, sig)) {
4963 print_signal_handlers(tty, buf, O_BUFLEN);
4964 }
4965 }
4967 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4969 extern bool signal_name(int signo, char* buf, size_t len);
4971 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4972 if (0 < exception_code && exception_code <= SIGRTMAX) {
4973 // signal
4974 if (!signal_name(exception_code, buf, size)) {
4975 jio_snprintf(buf, size, "SIG%d", exception_code);
4976 }
4977 return buf;
4978 } else {
4979 return NULL;
4980 }
4981 }
4983 // this is called _before_ the most of global arguments have been parsed
4984 void os::init(void) {
4985 char dummy; /* used to get a guess on initial stack address */
4986 // first_hrtime = gethrtime();
4988 // With LinuxThreads the JavaMain thread pid (primordial thread)
4989 // is different than the pid of the java launcher thread.
4990 // So, on Linux, the launcher thread pid is passed to the VM
4991 // via the sun.java.launcher.pid property.
4992 // Use this property instead of getpid() if it was correctly passed.
4993 // See bug 6351349.
4994 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4996 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4998 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
5000 init_random(1234567);
5002 ThreadCritical::initialize();
5004 Linux::set_page_size(sysconf(_SC_PAGESIZE));
5005 if (Linux::page_size() == -1) {
5006 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
5007 strerror(errno)));
5008 }
5009 init_page_sizes((size_t) Linux::page_size());
5011 Linux::initialize_system_info();
5013 // main_thread points to the aboriginal thread
5014 Linux::_main_thread = pthread_self();
5016 Linux::clock_init();
5017 initial_time_count = javaTimeNanos();
5019 // pthread_condattr initialization for monotonic clock
5020 int status;
5021 pthread_condattr_t* _condattr = os::Linux::condAttr();
5022 if ((status = pthread_condattr_init(_condattr)) != 0) {
5023 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
5024 }
5025 // Only set the clock if CLOCK_MONOTONIC is available
5026 if (Linux::supports_monotonic_clock()) {
5027 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
5028 if (status == EINVAL) {
5029 warning("Unable to use monotonic clock with relative timed-waits" \
5030 " - changes to the time-of-day clock may have adverse affects");
5031 } else {
5032 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
5033 }
5034 }
5035 }
5036 // else it defaults to CLOCK_REALTIME
5038 pthread_mutex_init(&dl_mutex, NULL);
5040 // If the pagesize of the VM is greater than 8K determine the appropriate
5041 // number of initial guard pages. The user can change this with the
5042 // command line arguments, if needed.
5043 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
5044 StackYellowPages = 1;
5045 StackRedPages = 1;
5046 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
5047 }
5048 }
5050 // To install functions for atexit system call
5051 extern "C" {
5052 static void perfMemory_exit_helper() {
5053 perfMemory_exit();
5054 }
5055 }
5057 void os::pd_init_container_support() {
5058 OSContainer::init();
5059 }
5061 // this is called _after_ the global arguments have been parsed
5062 jint os::init_2(void)
5063 {
5064 Linux::fast_thread_clock_init();
5066 // Allocate a single page and mark it as readable for safepoint polling
5067 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5068 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
5070 os::set_polling_page( polling_page );
5072 #ifndef PRODUCT
5073 if(Verbose && PrintMiscellaneous)
5074 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
5075 #endif
5077 if (!UseMembar) {
5078 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
5079 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
5080 os::set_memory_serialize_page( mem_serialize_page );
5082 #ifndef PRODUCT
5083 if(Verbose && PrintMiscellaneous)
5084 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
5085 #endif
5086 }
5088 // initialize suspend/resume support - must do this before signal_sets_init()
5089 if (SR_initialize() != 0) {
5090 perror("SR_initialize failed");
5091 return JNI_ERR;
5092 }
5094 Linux::signal_sets_init();
5095 Linux::install_signal_handlers();
5097 // Check minimum allowable stack size for thread creation and to initialize
5098 // the java system classes, including StackOverflowError - depends on page
5099 // size. Add a page for compiler2 recursion in main thread.
5100 // Add in 2*BytesPerWord times page size to account for VM stack during
5101 // class initialization depending on 32 or 64 bit VM.
5102 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
5103 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
5104 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
5106 size_t threadStackSizeInBytes = ThreadStackSize * K;
5107 if (threadStackSizeInBytes != 0 &&
5108 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
5109 tty->print_cr("\nThe stack size specified is too small, "
5110 "Specify at least %dk",
5111 os::Linux::min_stack_allowed/ K);
5112 return JNI_ERR;
5113 }
5115 // Make the stack size a multiple of the page size so that
5116 // the yellow/red zones can be guarded.
5117 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
5118 vm_page_size()));
5120 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
5122 #if defined(IA32)
5123 workaround_expand_exec_shield_cs_limit();
5124 #endif
5126 Linux::libpthread_init();
5127 if (PrintMiscellaneous && (Verbose || WizardMode)) {
5128 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
5129 Linux::glibc_version(), Linux::libpthread_version(),
5130 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
5131 }
5133 if (UseNUMA) {
5134 if (!Linux::libnuma_init()) {
5135 UseNUMA = false;
5136 } else {
5137 if ((Linux::numa_max_node() < 1)) {
5138 // There's only one node(they start from 0), disable NUMA.
5139 UseNUMA = false;
5140 }
5141 }
5142 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
5143 // we can make the adaptive lgrp chunk resizing work. If the user specified
5144 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
5145 // disable adaptive resizing.
5146 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
5147 if (FLAG_IS_DEFAULT(UseNUMA)) {
5148 UseNUMA = false;
5149 } else {
5150 if (FLAG_IS_DEFAULT(UseLargePages) &&
5151 FLAG_IS_DEFAULT(UseSHM) &&
5152 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
5153 UseLargePages = false;
5154 } else {
5155 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
5156 UseAdaptiveSizePolicy = false;
5157 UseAdaptiveNUMAChunkSizing = false;
5158 }
5159 }
5160 }
5161 if (!UseNUMA && ForceNUMA) {
5162 UseNUMA = true;
5163 }
5164 }
5166 if (MaxFDLimit) {
5167 // set the number of file descriptors to max. print out error
5168 // if getrlimit/setrlimit fails but continue regardless.
5169 struct rlimit nbr_files;
5170 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
5171 if (status != 0) {
5172 if (PrintMiscellaneous && (Verbose || WizardMode))
5173 perror("os::init_2 getrlimit failed");
5174 } else {
5175 nbr_files.rlim_cur = nbr_files.rlim_max;
5176 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
5177 if (status != 0) {
5178 if (PrintMiscellaneous && (Verbose || WizardMode))
5179 perror("os::init_2 setrlimit failed");
5180 }
5181 }
5182 }
5184 // Initialize lock used to serialize thread creation (see os::create_thread)
5185 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
5187 // at-exit methods are called in the reverse order of their registration.
5188 // atexit functions are called on return from main or as a result of a
5189 // call to exit(3C). There can be only 32 of these functions registered
5190 // and atexit() does not set errno.
5192 if (PerfAllowAtExitRegistration) {
5193 // only register atexit functions if PerfAllowAtExitRegistration is set.
5194 // atexit functions can be delayed until process exit time, which
5195 // can be problematic for embedded VM situations. Embedded VMs should
5196 // call DestroyJavaVM() to assure that VM resources are released.
5198 // note: perfMemory_exit_helper atexit function may be removed in
5199 // the future if the appropriate cleanup code can be added to the
5200 // VM_Exit VMOperation's doit method.
5201 if (atexit(perfMemory_exit_helper) != 0) {
5202 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
5203 }
5204 }
5206 // initialize thread priority policy
5207 prio_init();
5209 return JNI_OK;
5210 }
5212 // Mark the polling page as unreadable
5213 void os::make_polling_page_unreadable(void) {
5214 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
5215 fatal("Could not disable polling page");
5216 };
5218 // Mark the polling page as readable
5219 void os::make_polling_page_readable(void) {
5220 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
5221 fatal("Could not enable polling page");
5222 }
5223 };
5225 static int os_cpu_count(const cpu_set_t* cpus) {
5226 int count = 0;
5227 // only look up to the number of configured processors
5228 for (int i = 0; i < os::processor_count(); i++) {
5229 if (CPU_ISSET(i, cpus)) {
5230 count++;
5231 }
5232 }
5233 return count;
5234 }
5236 // Get the current number of available processors for this process.
5237 // This value can change at any time during a process's lifetime.
5238 // sched_getaffinity gives an accurate answer as it accounts for cpusets.
5239 // If anything goes wrong we fallback to returning the number of online
5240 // processors - which can be greater than the number available to the process.
5241 int os::Linux::active_processor_count() {
5242 cpu_set_t cpus; // can represent at most 1024 (CPU_SETSIZE) processors
5243 int cpus_size = sizeof(cpu_set_t);
5244 int cpu_count = 0;
5246 // pid 0 means the current thread - which we have to assume represents the process
5247 if (sched_getaffinity(0, cpus_size, &cpus) == 0) {
5248 cpu_count = os_cpu_count(&cpus);
5249 if (PrintActiveCpus) {
5250 tty->print_cr("active_processor_count: sched_getaffinity processor count: %d", cpu_count);
5251 }
5252 }
5253 else {
5254 cpu_count = ::sysconf(_SC_NPROCESSORS_ONLN);
5255 warning("sched_getaffinity failed (%s)- using online processor count (%d) "
5256 "which may exceed available processors", strerror(errno), cpu_count);
5257 }
5259 assert(cpu_count > 0 && cpu_count <= os::processor_count(), "sanity check");
5260 return cpu_count;
5261 }
5263 // Determine the active processor count from one of
5264 // three different sources:
5265 //
5266 // 1. User option -XX:ActiveProcessorCount
5267 // 2. kernel os calls (sched_getaffinity or sysconf(_SC_NPROCESSORS_ONLN)
5268 // 3. extracted from cgroup cpu subsystem (shares and quotas)
5269 //
5270 // Option 1, if specified, will always override.
5271 // If the cgroup subsystem is active and configured, we
5272 // will return the min of the cgroup and option 2 results.
5273 // This is required since tools, such as numactl, that
5274 // alter cpu affinity do not update cgroup subsystem
5275 // cpuset configuration files.
5276 int os::active_processor_count() {
5277 // User has overridden the number of active processors
5278 if (ActiveProcessorCount > 0) {
5279 if (PrintActiveCpus) {
5280 tty->print_cr("active_processor_count: "
5281 "active processor count set by user : %d",
5282 ActiveProcessorCount);
5283 }
5284 return ActiveProcessorCount;
5285 }
5287 int active_cpus;
5288 if (OSContainer::is_containerized()) {
5289 active_cpus = OSContainer::active_processor_count();
5290 if (PrintActiveCpus) {
5291 tty->print_cr("active_processor_count: determined by OSContainer: %d",
5292 active_cpus);
5293 }
5294 } else {
5295 active_cpus = os::Linux::active_processor_count();
5296 }
5298 return active_cpus;
5299 }
5301 void os::set_native_thread_name(const char *name) {
5302 // Not yet implemented.
5303 return;
5304 }
5306 bool os::distribute_processes(uint length, uint* distribution) {
5307 // Not yet implemented.
5308 return false;
5309 }
5311 bool os::bind_to_processor(uint processor_id) {
5312 // Not yet implemented.
5313 return false;
5314 }
5316 ///
5318 void os::SuspendedThreadTask::internal_do_task() {
5319 if (do_suspend(_thread->osthread())) {
5320 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
5321 do_task(context);
5322 do_resume(_thread->osthread());
5323 }
5324 }
5326 class PcFetcher : public os::SuspendedThreadTask {
5327 public:
5328 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
5329 ExtendedPC result();
5330 protected:
5331 void do_task(const os::SuspendedThreadTaskContext& context);
5332 private:
5333 ExtendedPC _epc;
5334 };
5336 ExtendedPC PcFetcher::result() {
5337 guarantee(is_done(), "task is not done yet.");
5338 return _epc;
5339 }
5341 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5342 Thread* thread = context.thread();
5343 OSThread* osthread = thread->osthread();
5344 if (osthread->ucontext() != NULL) {
5345 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5346 } else {
5347 // NULL context is unexpected, double-check this is the VMThread
5348 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5349 }
5350 }
5352 // Suspends the target using the signal mechanism and then grabs the PC before
5353 // resuming the target. Used by the flat-profiler only
5354 ExtendedPC os::get_thread_pc(Thread* thread) {
5355 // Make sure that it is called by the watcher for the VMThread
5356 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5357 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5359 PcFetcher fetcher(thread);
5360 fetcher.run();
5361 return fetcher.result();
5362 }
5364 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5365 {
5366 if (is_NPTL()) {
5367 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5368 } else {
5369 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5370 // word back to default 64bit precision if condvar is signaled. Java
5371 // wants 53bit precision. Save and restore current value.
5372 int fpu = get_fpu_control_word();
5373 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5374 set_fpu_control_word(fpu);
5375 return status;
5376 }
5377 }
5379 ////////////////////////////////////////////////////////////////////////////////
5380 // debug support
5382 bool os::find(address addr, outputStream* st) {
5383 Dl_info dlinfo;
5384 memset(&dlinfo, 0, sizeof(dlinfo));
5385 if (dladdr(addr, &dlinfo) != 0) {
5386 st->print(PTR_FORMAT ": ", addr);
5387 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5388 st->print("%s+%#x", dlinfo.dli_sname,
5389 addr - (intptr_t)dlinfo.dli_saddr);
5390 } else if (dlinfo.dli_fbase != NULL) {
5391 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5392 } else {
5393 st->print("<absolute address>");
5394 }
5395 if (dlinfo.dli_fname != NULL) {
5396 st->print(" in %s", dlinfo.dli_fname);
5397 }
5398 if (dlinfo.dli_fbase != NULL) {
5399 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5400 }
5401 st->cr();
5403 if (Verbose) {
5404 // decode some bytes around the PC
5405 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5406 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5407 address lowest = (address) dlinfo.dli_sname;
5408 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5409 if (begin < lowest) begin = lowest;
5410 Dl_info dlinfo2;
5411 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5412 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5413 end = (address) dlinfo2.dli_saddr;
5414 Disassembler::decode(begin, end, st);
5415 }
5416 return true;
5417 }
5418 return false;
5419 }
5421 ////////////////////////////////////////////////////////////////////////////////
5422 // misc
5424 // This does not do anything on Linux. This is basically a hook for being
5425 // able to use structured exception handling (thread-local exception filters)
5426 // on, e.g., Win32.
5427 void
5428 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5429 JavaCallArguments* args, Thread* thread) {
5430 f(value, method, args, thread);
5431 }
5433 void os::print_statistics() {
5434 }
5436 int os::message_box(const char* title, const char* message) {
5437 int i;
5438 fdStream err(defaultStream::error_fd());
5439 for (i = 0; i < 78; i++) err.print_raw("=");
5440 err.cr();
5441 err.print_raw_cr(title);
5442 for (i = 0; i < 78; i++) err.print_raw("-");
5443 err.cr();
5444 err.print_raw_cr(message);
5445 for (i = 0; i < 78; i++) err.print_raw("=");
5446 err.cr();
5448 char buf[16];
5449 // Prevent process from exiting upon "read error" without consuming all CPU
5450 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5452 return buf[0] == 'y' || buf[0] == 'Y';
5453 }
5455 int os::stat(const char *path, struct stat *sbuf) {
5456 char pathbuf[MAX_PATH];
5457 if (strlen(path) > MAX_PATH - 1) {
5458 errno = ENAMETOOLONG;
5459 return -1;
5460 }
5461 os::native_path(strcpy(pathbuf, path));
5462 return ::stat(pathbuf, sbuf);
5463 }
5465 bool os::check_heap(bool force) {
5466 return true;
5467 }
5469 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5470 return ::vsnprintf(buf, count, format, args);
5471 }
5473 // Is a (classpath) directory empty?
5474 bool os::dir_is_empty(const char* path) {
5475 DIR *dir = NULL;
5476 struct dirent *ptr;
5478 dir = opendir(path);
5479 if (dir == NULL) return true;
5481 /* Scan the directory */
5482 bool result = true;
5483 char buf[sizeof(struct dirent) + MAX_PATH];
5484 while (result && (ptr = ::readdir(dir)) != NULL) {
5485 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5486 result = false;
5487 }
5488 }
5489 closedir(dir);
5490 return result;
5491 }
5493 // This code originates from JDK's sysOpen and open64_w
5494 // from src/solaris/hpi/src/system_md.c
5496 #ifndef O_DELETE
5497 #define O_DELETE 0x10000
5498 #endif
5500 // Open a file. Unlink the file immediately after open returns
5501 // if the specified oflag has the O_DELETE flag set.
5502 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5504 int os::open(const char *path, int oflag, int mode) {
5506 if (strlen(path) > MAX_PATH - 1) {
5507 errno = ENAMETOOLONG;
5508 return -1;
5509 }
5510 int fd;
5511 int o_delete = (oflag & O_DELETE);
5512 oflag = oflag & ~O_DELETE;
5514 fd = ::open64(path, oflag, mode);
5515 if (fd == -1) return -1;
5517 //If the open succeeded, the file might still be a directory
5518 {
5519 struct stat64 buf64;
5520 int ret = ::fstat64(fd, &buf64);
5521 int st_mode = buf64.st_mode;
5523 if (ret != -1) {
5524 if ((st_mode & S_IFMT) == S_IFDIR) {
5525 errno = EISDIR;
5526 ::close(fd);
5527 return -1;
5528 }
5529 } else {
5530 ::close(fd);
5531 return -1;
5532 }
5533 }
5535 /*
5536 * All file descriptors that are opened in the JVM and not
5537 * specifically destined for a subprocess should have the
5538 * close-on-exec flag set. If we don't set it, then careless 3rd
5539 * party native code might fork and exec without closing all
5540 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5541 * UNIXProcess.c), and this in turn might:
5542 *
5543 * - cause end-of-file to fail to be detected on some file
5544 * descriptors, resulting in mysterious hangs, or
5545 *
5546 * - might cause an fopen in the subprocess to fail on a system
5547 * suffering from bug 1085341.
5548 *
5549 * (Yes, the default setting of the close-on-exec flag is a Unix
5550 * design flaw)
5551 *
5552 * See:
5553 * 1085341: 32-bit stdio routines should support file descriptors >255
5554 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5555 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5556 */
5557 #ifdef FD_CLOEXEC
5558 {
5559 int flags = ::fcntl(fd, F_GETFD);
5560 if (flags != -1)
5561 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5562 }
5563 #endif
5565 if (o_delete != 0) {
5566 ::unlink(path);
5567 }
5568 return fd;
5569 }
5572 // create binary file, rewriting existing file if required
5573 int os::create_binary_file(const char* path, bool rewrite_existing) {
5574 int oflags = O_WRONLY | O_CREAT;
5575 if (!rewrite_existing) {
5576 oflags |= O_EXCL;
5577 }
5578 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5579 }
5581 // return current position of file pointer
5582 jlong os::current_file_offset(int fd) {
5583 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5584 }
5586 // move file pointer to the specified offset
5587 jlong os::seek_to_file_offset(int fd, jlong offset) {
5588 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5589 }
5591 // This code originates from JDK's sysAvailable
5592 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5594 int os::available(int fd, jlong *bytes) {
5595 jlong cur, end;
5596 int mode;
5597 struct stat64 buf64;
5599 if (::fstat64(fd, &buf64) >= 0) {
5600 mode = buf64.st_mode;
5601 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5602 /*
5603 * XXX: is the following call interruptible? If so, this might
5604 * need to go through the INTERRUPT_IO() wrapper as for other
5605 * blocking, interruptible calls in this file.
5606 */
5607 int n;
5608 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5609 *bytes = n;
5610 return 1;
5611 }
5612 }
5613 }
5614 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5615 return 0;
5616 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5617 return 0;
5618 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5619 return 0;
5620 }
5621 *bytes = end - cur;
5622 return 1;
5623 }
5625 int os::socket_available(int fd, jint *pbytes) {
5626 // Linux doc says EINTR not returned, unlike Solaris
5627 int ret = ::ioctl(fd, FIONREAD, pbytes);
5629 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5630 // is expected to return 0 on failure and 1 on success to the jdk.
5631 return (ret < 0) ? 0 : 1;
5632 }
5634 // Map a block of memory.
5635 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5636 char *addr, size_t bytes, bool read_only,
5637 bool allow_exec) {
5638 int prot;
5639 int flags = MAP_PRIVATE;
5641 if (read_only) {
5642 prot = PROT_READ;
5643 } else {
5644 prot = PROT_READ | PROT_WRITE;
5645 }
5647 if (allow_exec) {
5648 prot |= PROT_EXEC;
5649 }
5651 if (addr != NULL) {
5652 flags |= MAP_FIXED;
5653 }
5655 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5656 fd, file_offset);
5657 if (mapped_address == MAP_FAILED) {
5658 return NULL;
5659 }
5660 return mapped_address;
5661 }
5664 // Remap a block of memory.
5665 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5666 char *addr, size_t bytes, bool read_only,
5667 bool allow_exec) {
5668 // same as map_memory() on this OS
5669 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5670 allow_exec);
5671 }
5674 // Unmap a block of memory.
5675 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5676 return munmap(addr, bytes) == 0;
5677 }
5679 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5681 static clockid_t thread_cpu_clockid(Thread* thread) {
5682 pthread_t tid = thread->osthread()->pthread_id();
5683 clockid_t clockid;
5685 // Get thread clockid
5686 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5687 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5688 return clockid;
5689 }
5691 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5692 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5693 // of a thread.
5694 //
5695 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5696 // the fast estimate available on the platform.
5698 jlong os::current_thread_cpu_time() {
5699 if (os::Linux::supports_fast_thread_cpu_time()) {
5700 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5701 } else {
5702 // return user + sys since the cost is the same
5703 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5704 }
5705 }
5707 jlong os::thread_cpu_time(Thread* thread) {
5708 // consistent with what current_thread_cpu_time() returns
5709 if (os::Linux::supports_fast_thread_cpu_time()) {
5710 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5711 } else {
5712 return slow_thread_cpu_time(thread, true /* user + sys */);
5713 }
5714 }
5716 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5717 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5718 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5719 } else {
5720 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5721 }
5722 }
5724 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5725 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5726 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5727 } else {
5728 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5729 }
5730 }
5732 //
5733 // -1 on error.
5734 //
5736 PRAGMA_DIAG_PUSH
5737 PRAGMA_FORMAT_NONLITERAL_IGNORED
5738 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5739 static bool proc_task_unchecked = true;
5740 static const char *proc_stat_path = "/proc/%d/stat";
5741 pid_t tid = thread->osthread()->thread_id();
5742 char *s;
5743 char stat[2048];
5744 int statlen;
5745 char proc_name[64];
5746 int count;
5747 long sys_time, user_time;
5748 char cdummy;
5749 int idummy;
5750 long ldummy;
5751 FILE *fp;
5753 // The /proc/<tid>/stat aggregates per-process usage on
5754 // new Linux kernels 2.6+ where NPTL is supported.
5755 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5756 // See bug 6328462.
5757 // There possibly can be cases where there is no directory
5758 // /proc/self/task, so we check its availability.
5759 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5760 // This is executed only once
5761 proc_task_unchecked = false;
5762 fp = fopen("/proc/self/task", "r");
5763 if (fp != NULL) {
5764 proc_stat_path = "/proc/self/task/%d/stat";
5765 fclose(fp);
5766 }
5767 }
5769 sprintf(proc_name, proc_stat_path, tid);
5770 fp = fopen(proc_name, "r");
5771 if ( fp == NULL ) return -1;
5772 statlen = fread(stat, 1, 2047, fp);
5773 stat[statlen] = '\0';
5774 fclose(fp);
5776 // Skip pid and the command string. Note that we could be dealing with
5777 // weird command names, e.g. user could decide to rename java launcher
5778 // to "java 1.4.2 :)", then the stat file would look like
5779 // 1234 (java 1.4.2 :)) R ... ...
5780 // We don't really need to know the command string, just find the last
5781 // occurrence of ")" and then start parsing from there. See bug 4726580.
5782 s = strrchr(stat, ')');
5783 if (s == NULL ) return -1;
5785 // Skip blank chars
5786 do s++; while (isspace(*s));
5788 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5789 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5790 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5791 &user_time, &sys_time);
5792 if ( count != 13 ) return -1;
5793 if (user_sys_cpu_time) {
5794 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5795 } else {
5796 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5797 }
5798 }
5799 PRAGMA_DIAG_POP
5801 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5802 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5803 info_ptr->may_skip_backward = false; // elapsed time not wall time
5804 info_ptr->may_skip_forward = false; // elapsed time not wall time
5805 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5806 }
5808 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5809 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5810 info_ptr->may_skip_backward = false; // elapsed time not wall time
5811 info_ptr->may_skip_forward = false; // elapsed time not wall time
5812 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5813 }
5815 bool os::is_thread_cpu_time_supported() {
5816 return true;
5817 }
5819 // System loadavg support. Returns -1 if load average cannot be obtained.
5820 // Linux doesn't yet have a (official) notion of processor sets,
5821 // so just return the system wide load average.
5822 int os::loadavg(double loadavg[], int nelem) {
5823 return ::getloadavg(loadavg, nelem);
5824 }
5826 void os::pause() {
5827 char filename[MAX_PATH];
5828 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5829 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5830 } else {
5831 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5832 }
5834 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5835 if (fd != -1) {
5836 struct stat buf;
5837 ::close(fd);
5838 while (::stat(filename, &buf) == 0) {
5839 (void)::poll(NULL, 0, 100);
5840 }
5841 } else {
5842 jio_fprintf(stderr,
5843 "Could not open pause file '%s', continuing immediately.\n", filename);
5844 }
5845 }
5848 // Refer to the comments in os_solaris.cpp park-unpark.
5849 //
5850 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5851 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5852 // For specifics regarding the bug see GLIBC BUGID 261237 :
5853 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5854 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5855 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5856 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5857 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5858 // and monitorenter when we're using 1-0 locking. All those operations may result in
5859 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5860 // of libpthread avoids the problem, but isn't practical.
5861 //
5862 // Possible remedies:
5863 //
5864 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5865 // This is palliative and probabilistic, however. If the thread is preempted
5866 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5867 // than the minimum period may have passed, and the abstime may be stale (in the
5868 // past) resultin in a hang. Using this technique reduces the odds of a hang
5869 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5870 //
5871 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5872 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5873 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5874 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5875 // thread.
5876 //
5877 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5878 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5879 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5880 // This also works well. In fact it avoids kernel-level scalability impediments
5881 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5882 // timers in a graceful fashion.
5883 //
5884 // 4. When the abstime value is in the past it appears that control returns
5885 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5886 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5887 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5888 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5889 // It may be possible to avoid reinitialization by checking the return
5890 // value from pthread_cond_timedwait(). In addition to reinitializing the
5891 // condvar we must establish the invariant that cond_signal() is only called
5892 // within critical sections protected by the adjunct mutex. This prevents
5893 // cond_signal() from "seeing" a condvar that's in the midst of being
5894 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5895 // desirable signal-after-unlock optimization that avoids futile context switching.
5896 //
5897 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5898 // structure when a condvar is used or initialized. cond_destroy() would
5899 // release the helper structure. Our reinitialize-after-timedwait fix
5900 // put excessive stress on malloc/free and locks protecting the c-heap.
5901 //
5902 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5903 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5904 // and only enabling the work-around for vulnerable environments.
5906 // utility to compute the abstime argument to timedwait:
5907 // millis is the relative timeout time
5908 // abstime will be the absolute timeout time
5909 // TODO: replace compute_abstime() with unpackTime()
5911 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5912 if (millis < 0) millis = 0;
5914 jlong seconds = millis / 1000;
5915 millis %= 1000;
5916 if (seconds > 50000000) { // see man cond_timedwait(3T)
5917 seconds = 50000000;
5918 }
5920 if (os::Linux::supports_monotonic_clock()) {
5921 struct timespec now;
5922 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5923 assert_status(status == 0, status, "clock_gettime");
5924 abstime->tv_sec = now.tv_sec + seconds;
5925 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5926 if (nanos >= NANOSECS_PER_SEC) {
5927 abstime->tv_sec += 1;
5928 nanos -= NANOSECS_PER_SEC;
5929 }
5930 abstime->tv_nsec = nanos;
5931 } else {
5932 struct timeval now;
5933 int status = gettimeofday(&now, NULL);
5934 assert(status == 0, "gettimeofday");
5935 abstime->tv_sec = now.tv_sec + seconds;
5936 long usec = now.tv_usec + millis * 1000;
5937 if (usec >= 1000000) {
5938 abstime->tv_sec += 1;
5939 usec -= 1000000;
5940 }
5941 abstime->tv_nsec = usec * 1000;
5942 }
5943 return abstime;
5944 }
5947 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5948 // Conceptually TryPark() should be equivalent to park(0).
5950 int os::PlatformEvent::TryPark() {
5951 for (;;) {
5952 const int v = _Event ;
5953 guarantee ((v == 0) || (v == 1), "invariant") ;
5954 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5955 }
5956 }
5958 void os::PlatformEvent::park() { // AKA "down()"
5959 // Invariant: Only the thread associated with the Event/PlatformEvent
5960 // may call park().
5961 // TODO: assert that _Assoc != NULL or _Assoc == Self
5962 int v ;
5963 for (;;) {
5964 v = _Event ;
5965 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5966 }
5967 guarantee (v >= 0, "invariant") ;
5968 if (v == 0) {
5969 // Do this the hard way by blocking ...
5970 int status = pthread_mutex_lock(_mutex);
5971 assert_status(status == 0, status, "mutex_lock");
5972 guarantee (_nParked == 0, "invariant") ;
5973 ++ _nParked ;
5974 while (_Event < 0) {
5975 status = pthread_cond_wait(_cond, _mutex);
5976 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5977 // Treat this the same as if the wait was interrupted
5978 if (status == ETIME) { status = EINTR; }
5979 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5980 }
5981 -- _nParked ;
5983 _Event = 0 ;
5984 status = pthread_mutex_unlock(_mutex);
5985 assert_status(status == 0, status, "mutex_unlock");
5986 // Paranoia to ensure our locked and lock-free paths interact
5987 // correctly with each other.
5988 OrderAccess::fence();
5989 }
5990 guarantee (_Event >= 0, "invariant") ;
5991 }
5993 int os::PlatformEvent::park(jlong millis) {
5994 guarantee (_nParked == 0, "invariant") ;
5996 int v ;
5997 for (;;) {
5998 v = _Event ;
5999 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
6000 }
6001 guarantee (v >= 0, "invariant") ;
6002 if (v != 0) return OS_OK ;
6004 // We do this the hard way, by blocking the thread.
6005 // Consider enforcing a minimum timeout value.
6006 struct timespec abst;
6007 compute_abstime(&abst, millis);
6009 int ret = OS_TIMEOUT;
6010 int status = pthread_mutex_lock(_mutex);
6011 assert_status(status == 0, status, "mutex_lock");
6012 guarantee (_nParked == 0, "invariant") ;
6013 ++_nParked ;
6015 // Object.wait(timo) will return because of
6016 // (a) notification
6017 // (b) timeout
6018 // (c) thread.interrupt
6019 //
6020 // Thread.interrupt and object.notify{All} both call Event::set.
6021 // That is, we treat thread.interrupt as a special case of notification.
6022 // The underlying Solaris implementation, cond_timedwait, admits
6023 // spurious/premature wakeups, but the JLS/JVM spec prevents the
6024 // JVM from making those visible to Java code. As such, we must
6025 // filter out spurious wakeups. We assume all ETIME returns are valid.
6026 //
6027 // TODO: properly differentiate simultaneous notify+interrupt.
6028 // In that case, we should propagate the notify to another waiter.
6030 while (_Event < 0) {
6031 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
6032 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6033 pthread_cond_destroy (_cond);
6034 pthread_cond_init (_cond, os::Linux::condAttr()) ;
6035 }
6036 assert_status(status == 0 || status == EINTR ||
6037 status == ETIME || status == ETIMEDOUT,
6038 status, "cond_timedwait");
6039 if (!FilterSpuriousWakeups) break ; // previous semantics
6040 if (status == ETIME || status == ETIMEDOUT) break ;
6041 // We consume and ignore EINTR and spurious wakeups.
6042 }
6043 --_nParked ;
6044 if (_Event >= 0) {
6045 ret = OS_OK;
6046 }
6047 _Event = 0 ;
6048 status = pthread_mutex_unlock(_mutex);
6049 assert_status(status == 0, status, "mutex_unlock");
6050 assert (_nParked == 0, "invariant") ;
6051 // Paranoia to ensure our locked and lock-free paths interact
6052 // correctly with each other.
6053 OrderAccess::fence();
6054 return ret;
6055 }
6057 void os::PlatformEvent::unpark() {
6058 // Transitions for _Event:
6059 // 0 :=> 1
6060 // 1 :=> 1
6061 // -1 :=> either 0 or 1; must signal target thread
6062 // That is, we can safely transition _Event from -1 to either
6063 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
6064 // unpark() calls.
6065 // See also: "Semaphores in Plan 9" by Mullender & Cox
6066 //
6067 // Note: Forcing a transition from "-1" to "1" on an unpark() means
6068 // that it will take two back-to-back park() calls for the owning
6069 // thread to block. This has the benefit of forcing a spurious return
6070 // from the first park() call after an unpark() call which will help
6071 // shake out uses of park() and unpark() without condition variables.
6073 if (Atomic::xchg(1, &_Event) >= 0) return;
6075 // Wait for the thread associated with the event to vacate
6076 int status = pthread_mutex_lock(_mutex);
6077 assert_status(status == 0, status, "mutex_lock");
6078 int AnyWaiters = _nParked;
6079 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
6080 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
6081 AnyWaiters = 0;
6082 pthread_cond_signal(_cond);
6083 }
6084 status = pthread_mutex_unlock(_mutex);
6085 assert_status(status == 0, status, "mutex_unlock");
6086 if (AnyWaiters != 0) {
6087 status = pthread_cond_signal(_cond);
6088 assert_status(status == 0, status, "cond_signal");
6089 }
6091 // Note that we signal() _after dropping the lock for "immortal" Events.
6092 // This is safe and avoids a common class of futile wakeups. In rare
6093 // circumstances this can cause a thread to return prematurely from
6094 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
6095 // simply re-test the condition and re-park itself.
6096 }
6099 // JSR166
6100 // -------------------------------------------------------
6102 /*
6103 * The solaris and linux implementations of park/unpark are fairly
6104 * conservative for now, but can be improved. They currently use a
6105 * mutex/condvar pair, plus a a count.
6106 * Park decrements count if > 0, else does a condvar wait. Unpark
6107 * sets count to 1 and signals condvar. Only one thread ever waits
6108 * on the condvar. Contention seen when trying to park implies that someone
6109 * is unparking you, so don't wait. And spurious returns are fine, so there
6110 * is no need to track notifications.
6111 */
6113 /*
6114 * This code is common to linux and solaris and will be moved to a
6115 * common place in dolphin.
6116 *
6117 * The passed in time value is either a relative time in nanoseconds
6118 * or an absolute time in milliseconds. Either way it has to be unpacked
6119 * into suitable seconds and nanoseconds components and stored in the
6120 * given timespec structure.
6121 * Given time is a 64-bit value and the time_t used in the timespec is only
6122 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
6123 * overflow if times way in the future are given. Further on Solaris versions
6124 * prior to 10 there is a restriction (see cond_timedwait) that the specified
6125 * number of seconds, in abstime, is less than current_time + 100,000,000.
6126 * As it will be 28 years before "now + 100000000" will overflow we can
6127 * ignore overflow and just impose a hard-limit on seconds using the value
6128 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
6129 * years from "now".
6130 */
6132 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
6133 assert (time > 0, "convertTime");
6134 time_t max_secs = 0;
6136 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
6137 struct timeval now;
6138 int status = gettimeofday(&now, NULL);
6139 assert(status == 0, "gettimeofday");
6141 max_secs = now.tv_sec + MAX_SECS;
6143 if (isAbsolute) {
6144 jlong secs = time / 1000;
6145 if (secs > max_secs) {
6146 absTime->tv_sec = max_secs;
6147 } else {
6148 absTime->tv_sec = secs;
6149 }
6150 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
6151 } else {
6152 jlong secs = time / NANOSECS_PER_SEC;
6153 if (secs >= MAX_SECS) {
6154 absTime->tv_sec = max_secs;
6155 absTime->tv_nsec = 0;
6156 } else {
6157 absTime->tv_sec = now.tv_sec + secs;
6158 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
6159 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6160 absTime->tv_nsec -= NANOSECS_PER_SEC;
6161 ++absTime->tv_sec; // note: this must be <= max_secs
6162 }
6163 }
6164 }
6165 } else {
6166 // must be relative using monotonic clock
6167 struct timespec now;
6168 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
6169 assert_status(status == 0, status, "clock_gettime");
6170 max_secs = now.tv_sec + MAX_SECS;
6171 jlong secs = time / NANOSECS_PER_SEC;
6172 if (secs >= MAX_SECS) {
6173 absTime->tv_sec = max_secs;
6174 absTime->tv_nsec = 0;
6175 } else {
6176 absTime->tv_sec = now.tv_sec + secs;
6177 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
6178 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
6179 absTime->tv_nsec -= NANOSECS_PER_SEC;
6180 ++absTime->tv_sec; // note: this must be <= max_secs
6181 }
6182 }
6183 }
6184 assert(absTime->tv_sec >= 0, "tv_sec < 0");
6185 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
6186 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
6187 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
6188 }
6190 void Parker::park(bool isAbsolute, jlong time) {
6191 // Ideally we'd do something useful while spinning, such
6192 // as calling unpackTime().
6194 // Optional fast-path check:
6195 // Return immediately if a permit is available.
6196 // We depend on Atomic::xchg() having full barrier semantics
6197 // since we are doing a lock-free update to _counter.
6198 if (Atomic::xchg(0, &_counter) > 0) return;
6200 Thread* thread = Thread::current();
6201 assert(thread->is_Java_thread(), "Must be JavaThread");
6202 JavaThread *jt = (JavaThread *)thread;
6204 // Optional optimization -- avoid state transitions if there's an interrupt pending.
6205 // Check interrupt before trying to wait
6206 if (Thread::is_interrupted(thread, false)) {
6207 return;
6208 }
6210 // Next, demultiplex/decode time arguments
6211 timespec absTime;
6212 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
6213 return;
6214 }
6215 if (time > 0) {
6216 unpackTime(&absTime, isAbsolute, time);
6217 }
6220 // Enter safepoint region
6221 // Beware of deadlocks such as 6317397.
6222 // The per-thread Parker:: mutex is a classic leaf-lock.
6223 // In particular a thread must never block on the Threads_lock while
6224 // holding the Parker:: mutex. If safepoints are pending both the
6225 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
6226 ThreadBlockInVM tbivm(jt);
6228 // Don't wait if cannot get lock since interference arises from
6229 // unblocking. Also. check interrupt before trying wait
6230 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
6231 return;
6232 }
6234 int status ;
6235 if (_counter > 0) { // no wait needed
6236 _counter = 0;
6237 status = pthread_mutex_unlock(_mutex);
6238 assert (status == 0, "invariant") ;
6239 // Paranoia to ensure our locked and lock-free paths interact
6240 // correctly with each other and Java-level accesses.
6241 OrderAccess::fence();
6242 return;
6243 }
6245 #ifdef ASSERT
6246 // Don't catch signals while blocked; let the running threads have the signals.
6247 // (This allows a debugger to break into the running thread.)
6248 sigset_t oldsigs;
6249 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
6250 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
6251 #endif
6253 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
6254 jt->set_suspend_equivalent();
6255 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
6257 assert(_cur_index == -1, "invariant");
6258 if (time == 0) {
6259 _cur_index = REL_INDEX; // arbitrary choice when not timed
6260 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
6261 } else {
6262 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
6263 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
6264 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
6265 pthread_cond_destroy (&_cond[_cur_index]) ;
6266 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
6267 }
6268 }
6269 _cur_index = -1;
6270 assert_status(status == 0 || status == EINTR ||
6271 status == ETIME || status == ETIMEDOUT,
6272 status, "cond_timedwait");
6274 #ifdef ASSERT
6275 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
6276 #endif
6278 _counter = 0 ;
6279 status = pthread_mutex_unlock(_mutex) ;
6280 assert_status(status == 0, status, "invariant") ;
6281 // Paranoia to ensure our locked and lock-free paths interact
6282 // correctly with each other and Java-level accesses.
6283 OrderAccess::fence();
6285 // If externally suspended while waiting, re-suspend
6286 if (jt->handle_special_suspend_equivalent_condition()) {
6287 jt->java_suspend_self();
6288 }
6289 }
6291 void Parker::unpark() {
6292 int s, status ;
6293 status = pthread_mutex_lock(_mutex);
6294 assert (status == 0, "invariant") ;
6295 s = _counter;
6296 _counter = 1;
6297 if (s < 1) {
6298 // thread might be parked
6299 if (_cur_index != -1) {
6300 // thread is definitely parked
6301 if (WorkAroundNPTLTimedWaitHang) {
6302 status = pthread_cond_signal (&_cond[_cur_index]);
6303 assert (status == 0, "invariant");
6304 status = pthread_mutex_unlock(_mutex);
6305 assert (status == 0, "invariant");
6306 } else {
6307 // must capture correct index before unlocking
6308 int index = _cur_index;
6309 status = pthread_mutex_unlock(_mutex);
6310 assert (status == 0, "invariant");
6311 status = pthread_cond_signal (&_cond[index]);
6312 assert (status == 0, "invariant");
6313 }
6314 } else {
6315 pthread_mutex_unlock(_mutex);
6316 assert (status == 0, "invariant") ;
6317 }
6318 } else {
6319 pthread_mutex_unlock(_mutex);
6320 assert (status == 0, "invariant") ;
6321 }
6322 }
6325 extern char** environ;
6327 // Run the specified command in a separate process. Return its exit value,
6328 // or -1 on failure (e.g. can't fork a new process).
6329 // Unlike system(), this function can be called from signal handler. It
6330 // doesn't block SIGINT et al.
6331 int os::fork_and_exec(char* cmd) {
6332 const char * argv[4] = {"sh", "-c", cmd, NULL};
6334 pid_t pid = fork();
6336 if (pid < 0) {
6337 // fork failed
6338 return -1;
6340 } else if (pid == 0) {
6341 // child process
6343 execve("/bin/sh", (char* const*)argv, environ);
6345 // execve failed
6346 _exit(-1);
6348 } else {
6349 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6350 // care about the actual exit code, for now.
6352 int status;
6354 // Wait for the child process to exit. This returns immediately if
6355 // the child has already exited. */
6356 while (waitpid(pid, &status, 0) < 0) {
6357 switch (errno) {
6358 case ECHILD: return 0;
6359 case EINTR: break;
6360 default: return -1;
6361 }
6362 }
6364 if (WIFEXITED(status)) {
6365 // The child exited normally; get its exit code.
6366 return WEXITSTATUS(status);
6367 } else if (WIFSIGNALED(status)) {
6368 // The child exited because of a signal
6369 // The best value to return is 0x80 + signal number,
6370 // because that is what all Unix shells do, and because
6371 // it allows callers to distinguish between process exit and
6372 // process death by signal.
6373 return 0x80 + WTERMSIG(status);
6374 } else {
6375 // Unknown exit code; pass it through
6376 return status;
6377 }
6378 }
6379 }
6381 // is_headless_jre()
6382 //
6383 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6384 // in order to report if we are running in a headless jre
6385 //
6386 // Since JDK8 xawt/libmawt.so was moved into the same directory
6387 // as libawt.so, and renamed libawt_xawt.so
6388 //
6389 bool os::is_headless_jre() {
6390 struct stat statbuf;
6391 char buf[MAXPATHLEN];
6392 char libmawtpath[MAXPATHLEN];
6393 const char *xawtstr = "/xawt/libmawt.so";
6394 const char *new_xawtstr = "/libawt_xawt.so";
6395 char *p;
6397 // Get path to libjvm.so
6398 os::jvm_path(buf, sizeof(buf));
6400 // Get rid of libjvm.so
6401 p = strrchr(buf, '/');
6402 if (p == NULL) return false;
6403 else *p = '\0';
6405 // Get rid of client or server
6406 p = strrchr(buf, '/');
6407 if (p == NULL) return false;
6408 else *p = '\0';
6410 // check xawt/libmawt.so
6411 strcpy(libmawtpath, buf);
6412 strcat(libmawtpath, xawtstr);
6413 if (::stat(libmawtpath, &statbuf) == 0) return false;
6415 // check libawt_xawt.so
6416 strcpy(libmawtpath, buf);
6417 strcat(libmawtpath, new_xawtstr);
6418 if (::stat(libmawtpath, &statbuf) == 0) return false;
6420 return true;
6421 }
6423 // Get the default path to the core file
6424 // Returns the length of the string
6425 int os::get_core_path(char* buffer, size_t bufferSize) {
6426 const char* p = get_current_directory(buffer, bufferSize);
6428 if (p == NULL) {
6429 assert(p != NULL, "failed to get current directory");
6430 return 0;
6431 }
6433 return strlen(buffer);
6434 }
6436 /////////////// Unit tests ///////////////
6438 #ifndef PRODUCT
6440 #define test_log(...) \
6441 do {\
6442 if (VerboseInternalVMTests) { \
6443 tty->print_cr(__VA_ARGS__); \
6444 tty->flush(); \
6445 }\
6446 } while (false)
6448 class TestReserveMemorySpecial : AllStatic {
6449 public:
6450 static void small_page_write(void* addr, size_t size) {
6451 size_t page_size = os::vm_page_size();
6453 char* end = (char*)addr + size;
6454 for (char* p = (char*)addr; p < end; p += page_size) {
6455 *p = 1;
6456 }
6457 }
6459 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6460 if (!UseHugeTLBFS) {
6461 return;
6462 }
6464 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6466 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6468 if (addr != NULL) {
6469 small_page_write(addr, size);
6471 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6472 }
6473 }
6475 static void test_reserve_memory_special_huge_tlbfs_only() {
6476 if (!UseHugeTLBFS) {
6477 return;
6478 }
6480 size_t lp = os::large_page_size();
6482 for (size_t size = lp; size <= lp * 10; size += lp) {
6483 test_reserve_memory_special_huge_tlbfs_only(size);
6484 }
6485 }
6487 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6488 size_t lp = os::large_page_size();
6489 size_t ag = os::vm_allocation_granularity();
6491 // sizes to test
6492 const size_t sizes[] = {
6493 lp, lp + ag, lp + lp / 2, lp * 2,
6494 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6495 lp * 10, lp * 10 + lp / 2
6496 };
6497 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6499 // For each size/alignment combination, we test three scenarios:
6500 // 1) with req_addr == NULL
6501 // 2) with a non-null req_addr at which we expect to successfully allocate
6502 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6503 // expect the allocation to either fail or to ignore req_addr
6505 // Pre-allocate two areas; they shall be as large as the largest allocation
6506 // and aligned to the largest alignment we will be testing.
6507 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6508 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6509 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6510 -1, 0);
6511 assert(mapping1 != MAP_FAILED, "should work");
6513 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6514 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6515 -1, 0);
6516 assert(mapping2 != MAP_FAILED, "should work");
6518 // Unmap the first mapping, but leave the second mapping intact: the first
6519 // mapping will serve as a value for a "good" req_addr (case 2). The second
6520 // mapping, still intact, as "bad" req_addr (case 3).
6521 ::munmap(mapping1, mapping_size);
6523 // Case 1
6524 test_log("%s, req_addr NULL:", __FUNCTION__);
6525 test_log("size align result");
6527 for (int i = 0; i < num_sizes; i++) {
6528 const size_t size = sizes[i];
6529 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6530 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6531 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6532 size, alignment, p, (p != NULL ? "" : "(failed)"));
6533 if (p != NULL) {
6534 assert(is_ptr_aligned(p, alignment), "must be");
6535 small_page_write(p, size);
6536 os::Linux::release_memory_special_huge_tlbfs(p, size);
6537 }
6538 }
6539 }
6541 // Case 2
6542 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6543 test_log("size align req_addr result");
6545 for (int i = 0; i < num_sizes; i++) {
6546 const size_t size = sizes[i];
6547 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6548 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6549 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6550 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6551 size, alignment, req_addr, p,
6552 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6553 if (p != NULL) {
6554 assert(p == req_addr, "must be");
6555 small_page_write(p, size);
6556 os::Linux::release_memory_special_huge_tlbfs(p, size);
6557 }
6558 }
6559 }
6561 // Case 3
6562 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6563 test_log("size align req_addr result");
6565 for (int i = 0; i < num_sizes; i++) {
6566 const size_t size = sizes[i];
6567 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6568 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6569 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6570 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6571 size, alignment, req_addr, p,
6572 ((p != NULL ? "" : "(failed)")));
6573 // as the area around req_addr contains already existing mappings, the API should always
6574 // return NULL (as per contract, it cannot return another address)
6575 assert(p == NULL, "must be");
6576 }
6577 }
6579 ::munmap(mapping2, mapping_size);
6581 }
6583 static void test_reserve_memory_special_huge_tlbfs() {
6584 if (!UseHugeTLBFS) {
6585 return;
6586 }
6588 test_reserve_memory_special_huge_tlbfs_only();
6589 test_reserve_memory_special_huge_tlbfs_mixed();
6590 }
6592 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6593 if (!UseSHM) {
6594 return;
6595 }
6597 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6599 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6601 if (addr != NULL) {
6602 assert(is_ptr_aligned(addr, alignment), "Check");
6603 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6605 small_page_write(addr, size);
6607 os::Linux::release_memory_special_shm(addr, size);
6608 }
6609 }
6611 static void test_reserve_memory_special_shm() {
6612 size_t lp = os::large_page_size();
6613 size_t ag = os::vm_allocation_granularity();
6615 for (size_t size = ag; size < lp * 3; size += ag) {
6616 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6617 test_reserve_memory_special_shm(size, alignment);
6618 }
6619 }
6620 }
6622 static void test() {
6623 test_reserve_memory_special_huge_tlbfs();
6624 test_reserve_memory_special_shm();
6625 }
6626 };
6628 void TestReserveMemorySpecial_test() {
6629 TestReserveMemorySpecial::test();
6630 }
6632 #endif