Thu, 13 Mar 2014 14:57:01 -0700
Merge
1 /*
2 * Copyright (c) 1999, 2013, 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 "prims/jniFastGetField.hpp"
41 #include "prims/jvm.h"
42 #include "prims/jvm_misc.hpp"
43 #include "runtime/arguments.hpp"
44 #include "runtime/extendedPC.hpp"
45 #include "runtime/globals.hpp"
46 #include "runtime/interfaceSupport.hpp"
47 #include "runtime/init.hpp"
48 #include "runtime/java.hpp"
49 #include "runtime/javaCalls.hpp"
50 #include "runtime/mutexLocker.hpp"
51 #include "runtime/objectMonitor.hpp"
52 #include "runtime/osThread.hpp"
53 #include "runtime/perfMemory.hpp"
54 #include "runtime/sharedRuntime.hpp"
55 #include "runtime/statSampler.hpp"
56 #include "runtime/stubRoutines.hpp"
57 #include "runtime/thread.inline.hpp"
58 #include "runtime/threadCritical.hpp"
59 #include "runtime/timer.hpp"
60 #include "services/attachListener.hpp"
61 #include "services/memTracker.hpp"
62 #include "services/runtimeService.hpp"
63 #include "utilities/decoder.hpp"
64 #include "utilities/defaultStream.hpp"
65 #include "utilities/events.hpp"
66 #include "utilities/elfFile.hpp"
67 #include "utilities/growableArray.hpp"
68 #include "utilities/vmError.hpp"
70 // put OS-includes here
71 # include <sys/types.h>
72 # include <sys/mman.h>
73 # include <sys/stat.h>
74 # include <sys/select.h>
75 # include <pthread.h>
76 # include <signal.h>
77 # include <errno.h>
78 # include <dlfcn.h>
79 # include <stdio.h>
80 # include <unistd.h>
81 # include <sys/resource.h>
82 # include <pthread.h>
83 # include <sys/stat.h>
84 # include <sys/time.h>
85 # include <sys/times.h>
86 # include <sys/utsname.h>
87 # include <sys/socket.h>
88 # include <sys/wait.h>
89 # include <pwd.h>
90 # include <poll.h>
91 # include <semaphore.h>
92 # include <fcntl.h>
93 # include <string.h>
94 # include <syscall.h>
95 # include <sys/sysinfo.h>
96 # include <gnu/libc-version.h>
97 # include <sys/ipc.h>
98 # include <sys/shm.h>
99 # include <link.h>
100 # include <stdint.h>
101 # include <inttypes.h>
102 # include <sys/ioctl.h>
104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
105 // getrusage() is prepared to handle the associated failure.
106 #ifndef RUSAGE_THREAD
107 #define RUSAGE_THREAD (1) /* only the calling thread */
108 #endif
110 #define MAX_PATH (2 * K)
112 // for timer info max values which include all bits
113 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
115 #define LARGEPAGES_BIT (1 << 6)
116 ////////////////////////////////////////////////////////////////////////////////
117 // global variables
118 julong os::Linux::_physical_memory = 0;
120 address os::Linux::_initial_thread_stack_bottom = NULL;
121 uintptr_t os::Linux::_initial_thread_stack_size = 0;
123 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
124 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
125 Mutex* os::Linux::_createThread_lock = NULL;
126 pthread_t os::Linux::_main_thread;
127 int os::Linux::_page_size = -1;
128 const int os::Linux::_vm_default_page_size = (8 * K);
129 bool os::Linux::_is_floating_stack = false;
130 bool os::Linux::_is_NPTL = false;
131 bool os::Linux::_supports_fast_thread_cpu_time = false;
132 const char * os::Linux::_glibc_version = NULL;
133 const char * os::Linux::_libpthread_version = NULL;
134 pthread_condattr_t os::Linux::_condattr[1];
136 static jlong initial_time_count=0;
138 static int clock_tics_per_sec = 100;
140 // For diagnostics to print a message once. see run_periodic_checks
141 static sigset_t check_signal_done;
142 static bool check_signals = true;
144 static pid_t _initial_pid = 0;
146 /* Signal number used to suspend/resume a thread */
148 /* do not use any signal number less than SIGSEGV, see 4355769 */
149 static int SR_signum = SIGUSR2;
150 sigset_t SR_sigset;
152 /* Used to protect dlsym() calls */
153 static pthread_mutex_t dl_mutex;
155 // Declarations
156 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
158 #ifdef JAVASE_EMBEDDED
159 class MemNotifyThread: public Thread {
160 friend class VMStructs;
161 public:
162 virtual void run();
164 private:
165 static MemNotifyThread* _memnotify_thread;
166 int _fd;
168 public:
170 // Constructor
171 MemNotifyThread(int fd);
173 // Tester
174 bool is_memnotify_thread() const { return true; }
176 // Printing
177 char* name() const { return (char*)"Linux MemNotify Thread"; }
179 // Returns the single instance of the MemNotifyThread
180 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
182 // Create and start the single instance of MemNotifyThread
183 static void start();
184 };
185 #endif // JAVASE_EMBEDDED
187 // utility functions
189 static int SR_initialize();
191 julong os::available_memory() {
192 return Linux::available_memory();
193 }
195 julong os::Linux::available_memory() {
196 // values in struct sysinfo are "unsigned long"
197 struct sysinfo si;
198 sysinfo(&si);
200 return (julong)si.freeram * si.mem_unit;
201 }
203 julong os::physical_memory() {
204 return Linux::physical_memory();
205 }
207 ////////////////////////////////////////////////////////////////////////////////
208 // environment support
210 bool os::getenv(const char* name, char* buf, int len) {
211 const char* val = ::getenv(name);
212 if (val != NULL && strlen(val) < (size_t)len) {
213 strcpy(buf, val);
214 return true;
215 }
216 if (len > 0) buf[0] = 0; // return a null string
217 return false;
218 }
221 // Return true if user is running as root.
223 bool os::have_special_privileges() {
224 static bool init = false;
225 static bool privileges = false;
226 if (!init) {
227 privileges = (getuid() != geteuid()) || (getgid() != getegid());
228 init = true;
229 }
230 return privileges;
231 }
234 #ifndef SYS_gettid
235 // i386: 224, ia64: 1105, amd64: 186, sparc 143
236 #ifdef __ia64__
237 #define SYS_gettid 1105
238 #elif __i386__
239 #define SYS_gettid 224
240 #elif __amd64__
241 #define SYS_gettid 186
242 #elif __sparc__
243 #define SYS_gettid 143
244 #else
245 #error define gettid for the arch
246 #endif
247 #endif
249 // Cpu architecture string
250 #if defined(ZERO)
251 static char cpu_arch[] = ZERO_LIBARCH;
252 #elif defined(IA64)
253 static char cpu_arch[] = "ia64";
254 #elif defined(IA32)
255 static char cpu_arch[] = "i386";
256 #elif defined(AMD64)
257 static char cpu_arch[] = "amd64";
258 #elif defined(ARM)
259 static char cpu_arch[] = "arm";
260 #elif defined(PPC32)
261 static char cpu_arch[] = "ppc";
262 #elif defined(PPC64)
263 static char cpu_arch[] = "ppc64";
264 #elif defined(SPARC)
265 # ifdef _LP64
266 static char cpu_arch[] = "sparcv9";
267 # else
268 static char cpu_arch[] = "sparc";
269 # endif
270 #else
271 #error Add appropriate cpu_arch setting
272 #endif
275 // pid_t gettid()
276 //
277 // Returns the kernel thread id of the currently running thread. Kernel
278 // thread id is used to access /proc.
279 //
280 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
281 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
282 //
283 pid_t os::Linux::gettid() {
284 int rslt = syscall(SYS_gettid);
285 if (rslt == -1) {
286 // old kernel, no NPTL support
287 return getpid();
288 } else {
289 return (pid_t)rslt;
290 }
291 }
293 // Most versions of linux have a bug where the number of processors are
294 // determined by looking at the /proc file system. In a chroot environment,
295 // the system call returns 1. This causes the VM to act as if it is
296 // a single processor and elide locking (see is_MP() call).
297 static bool unsafe_chroot_detected = false;
298 static const char *unstable_chroot_error = "/proc file system not found.\n"
299 "Java may be unstable running multithreaded in a chroot "
300 "environment on Linux when /proc filesystem is not mounted.";
302 void os::Linux::initialize_system_info() {
303 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
304 if (processor_count() == 1) {
305 pid_t pid = os::Linux::gettid();
306 char fname[32];
307 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
308 FILE *fp = fopen(fname, "r");
309 if (fp == NULL) {
310 unsafe_chroot_detected = true;
311 } else {
312 fclose(fp);
313 }
314 }
315 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
316 assert(processor_count() > 0, "linux error");
317 }
319 void os::init_system_properties_values() {
320 // char arch[12];
321 // sysinfo(SI_ARCHITECTURE, arch, sizeof(arch));
323 // The next steps are taken in the product version:
324 //
325 // Obtain the JAVA_HOME value from the location of libjvm.so.
326 // This library should be located at:
327 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
328 //
329 // If "/jre/lib/" appears at the right place in the path, then we
330 // assume libjvm.so is installed in a JDK and we use this path.
331 //
332 // Otherwise exit with message: "Could not create the Java virtual machine."
333 //
334 // The following extra steps are taken in the debugging version:
335 //
336 // If "/jre/lib/" does NOT appear at the right place in the path
337 // instead of exit check for $JAVA_HOME environment variable.
338 //
339 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
340 // then we append a fake suffix "hotspot/libjvm.so" to this path so
341 // it looks like libjvm.so is installed there
342 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
343 //
344 // Otherwise exit.
345 //
346 // Important note: if the location of libjvm.so changes this
347 // code needs to be changed accordingly.
349 // The next few definitions allow the code to be verbatim:
350 #define malloc(n) (char*)NEW_C_HEAP_ARRAY(char, (n), mtInternal)
351 #define getenv(n) ::getenv(n)
353 /*
354 * See ld(1):
355 * The linker uses the following search paths to locate required
356 * shared libraries:
357 * 1: ...
358 * ...
359 * 7: The default directories, normally /lib and /usr/lib.
360 */
361 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
362 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
363 #else
364 #define DEFAULT_LIBPATH "/lib:/usr/lib"
365 #endif
367 #define EXTENSIONS_DIR "/lib/ext"
368 #define ENDORSED_DIR "/lib/endorsed"
369 #define REG_DIR "/usr/java/packages"
371 {
372 /* sysclasspath, java_home, dll_dir */
373 {
374 char *home_path;
375 char *dll_path;
376 char *pslash;
377 char buf[MAXPATHLEN];
378 os::jvm_path(buf, sizeof(buf));
380 // Found the full path to libjvm.so.
381 // Now cut the path to <java_home>/jre if we can.
382 *(strrchr(buf, '/')) = '\0'; /* get rid of /libjvm.so */
383 pslash = strrchr(buf, '/');
384 if (pslash != NULL)
385 *pslash = '\0'; /* get rid of /{client|server|hotspot} */
386 dll_path = malloc(strlen(buf) + 1);
387 if (dll_path == NULL)
388 return;
389 strcpy(dll_path, buf);
390 Arguments::set_dll_dir(dll_path);
392 if (pslash != NULL) {
393 pslash = strrchr(buf, '/');
394 if (pslash != NULL) {
395 *pslash = '\0'; /* get rid of /<arch> */
396 pslash = strrchr(buf, '/');
397 if (pslash != NULL)
398 *pslash = '\0'; /* get rid of /lib */
399 }
400 }
402 home_path = malloc(strlen(buf) + 1);
403 if (home_path == NULL)
404 return;
405 strcpy(home_path, buf);
406 Arguments::set_java_home(home_path);
408 if (!set_boot_path('/', ':'))
409 return;
410 }
412 /*
413 * Where to look for native libraries
414 *
415 * Note: Due to a legacy implementation, most of the library path
416 * is set in the launcher. This was to accomodate linking restrictions
417 * on legacy Linux implementations (which are no longer supported).
418 * Eventually, all the library path setting will be done here.
419 *
420 * However, to prevent the proliferation of improperly built native
421 * libraries, the new path component /usr/java/packages is added here.
422 * Eventually, all the library path setting will be done here.
423 */
424 {
425 char *ld_library_path;
427 /*
428 * Construct the invariant part of ld_library_path. Note that the
429 * space for the colon and the trailing null are provided by the
430 * nulls included by the sizeof operator (so actually we allocate
431 * a byte more than necessary).
432 */
433 ld_library_path = (char *) malloc(sizeof(REG_DIR) + sizeof("/lib/") +
434 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH));
435 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
437 /*
438 * Get the user setting of LD_LIBRARY_PATH, and prepended it. It
439 * should always exist (until the legacy problem cited above is
440 * addressed).
441 */
442 char *v = getenv("LD_LIBRARY_PATH");
443 if (v != NULL) {
444 char *t = ld_library_path;
445 /* That's +1 for the colon and +1 for the trailing '\0' */
446 ld_library_path = (char *) malloc(strlen(v) + 1 + strlen(t) + 1);
447 sprintf(ld_library_path, "%s:%s", v, t);
448 }
449 Arguments::set_library_path(ld_library_path);
450 }
452 /*
453 * Extensions directories.
454 *
455 * Note that the space for the colon and the trailing null are provided
456 * by the nulls included by the sizeof operator (so actually one byte more
457 * than necessary is allocated).
458 */
459 {
460 char *buf = malloc(strlen(Arguments::get_java_home()) +
461 sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR));
462 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR,
463 Arguments::get_java_home());
464 Arguments::set_ext_dirs(buf);
465 }
467 /* Endorsed standards default directory. */
468 {
469 char * buf;
470 buf = malloc(strlen(Arguments::get_java_home()) + sizeof(ENDORSED_DIR));
471 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
472 Arguments::set_endorsed_dirs(buf);
473 }
474 }
476 #undef malloc
477 #undef getenv
478 #undef EXTENSIONS_DIR
479 #undef ENDORSED_DIR
481 // Done
482 return;
483 }
485 ////////////////////////////////////////////////////////////////////////////////
486 // breakpoint support
488 void os::breakpoint() {
489 BREAKPOINT;
490 }
492 extern "C" void breakpoint() {
493 // use debugger to set breakpoint here
494 }
496 ////////////////////////////////////////////////////////////////////////////////
497 // signal support
499 debug_only(static bool signal_sets_initialized = false);
500 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
502 bool os::Linux::is_sig_ignored(int sig) {
503 struct sigaction oact;
504 sigaction(sig, (struct sigaction*)NULL, &oact);
505 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
506 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
507 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
508 return true;
509 else
510 return false;
511 }
513 void os::Linux::signal_sets_init() {
514 // Should also have an assertion stating we are still single-threaded.
515 assert(!signal_sets_initialized, "Already initialized");
516 // Fill in signals that are necessarily unblocked for all threads in
517 // the VM. Currently, we unblock the following signals:
518 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
519 // by -Xrs (=ReduceSignalUsage));
520 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
521 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
522 // the dispositions or masks wrt these signals.
523 // Programs embedding the VM that want to use the above signals for their
524 // own purposes must, at this time, use the "-Xrs" option to prevent
525 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
526 // (See bug 4345157, and other related bugs).
527 // In reality, though, unblocking these signals is really a nop, since
528 // these signals are not blocked by default.
529 sigemptyset(&unblocked_sigs);
530 sigemptyset(&allowdebug_blocked_sigs);
531 sigaddset(&unblocked_sigs, SIGILL);
532 sigaddset(&unblocked_sigs, SIGSEGV);
533 sigaddset(&unblocked_sigs, SIGBUS);
534 sigaddset(&unblocked_sigs, SIGFPE);
535 #if defined(PPC64)
536 sigaddset(&unblocked_sigs, SIGTRAP);
537 #endif
538 sigaddset(&unblocked_sigs, SR_signum);
540 if (!ReduceSignalUsage) {
541 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
542 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
543 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
544 }
545 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
546 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
547 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
548 }
549 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
550 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
551 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
552 }
553 }
554 // Fill in signals that are blocked by all but the VM thread.
555 sigemptyset(&vm_sigs);
556 if (!ReduceSignalUsage)
557 sigaddset(&vm_sigs, BREAK_SIGNAL);
558 debug_only(signal_sets_initialized = true);
560 }
562 // These are signals that are unblocked while a thread is running Java.
563 // (For some reason, they get blocked by default.)
564 sigset_t* os::Linux::unblocked_signals() {
565 assert(signal_sets_initialized, "Not initialized");
566 return &unblocked_sigs;
567 }
569 // These are the signals that are blocked while a (non-VM) thread is
570 // running Java. Only the VM thread handles these signals.
571 sigset_t* os::Linux::vm_signals() {
572 assert(signal_sets_initialized, "Not initialized");
573 return &vm_sigs;
574 }
576 // These are signals that are blocked during cond_wait to allow debugger in
577 sigset_t* os::Linux::allowdebug_blocked_signals() {
578 assert(signal_sets_initialized, "Not initialized");
579 return &allowdebug_blocked_sigs;
580 }
582 void os::Linux::hotspot_sigmask(Thread* thread) {
584 //Save caller's signal mask before setting VM signal mask
585 sigset_t caller_sigmask;
586 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
588 OSThread* osthread = thread->osthread();
589 osthread->set_caller_sigmask(caller_sigmask);
591 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
593 if (!ReduceSignalUsage) {
594 if (thread->is_VM_thread()) {
595 // Only the VM thread handles BREAK_SIGNAL ...
596 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
597 } else {
598 // ... all other threads block BREAK_SIGNAL
599 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
600 }
601 }
602 }
604 //////////////////////////////////////////////////////////////////////////////
605 // detecting pthread library
607 void os::Linux::libpthread_init() {
608 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
609 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
610 // generic name for earlier versions.
611 // Define macros here so we can build HotSpot on old systems.
612 # ifndef _CS_GNU_LIBC_VERSION
613 # define _CS_GNU_LIBC_VERSION 2
614 # endif
615 # ifndef _CS_GNU_LIBPTHREAD_VERSION
616 # define _CS_GNU_LIBPTHREAD_VERSION 3
617 # endif
619 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
620 if (n > 0) {
621 char *str = (char *)malloc(n, mtInternal);
622 confstr(_CS_GNU_LIBC_VERSION, str, n);
623 os::Linux::set_glibc_version(str);
624 } else {
625 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
626 static char _gnu_libc_version[32];
627 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
628 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
629 os::Linux::set_glibc_version(_gnu_libc_version);
630 }
632 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
633 if (n > 0) {
634 char *str = (char *)malloc(n, mtInternal);
635 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
636 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
637 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
638 // is the case. LinuxThreads has a hard limit on max number of threads.
639 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
640 // On the other hand, NPTL does not have such a limit, sysconf()
641 // will return -1 and errno is not changed. Check if it is really NPTL.
642 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
643 strstr(str, "NPTL") &&
644 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
645 free(str);
646 os::Linux::set_libpthread_version("linuxthreads");
647 } else {
648 os::Linux::set_libpthread_version(str);
649 }
650 } else {
651 // glibc before 2.3.2 only has LinuxThreads.
652 os::Linux::set_libpthread_version("linuxthreads");
653 }
655 if (strstr(libpthread_version(), "NPTL")) {
656 os::Linux::set_is_NPTL();
657 } else {
658 os::Linux::set_is_LinuxThreads();
659 }
661 // LinuxThreads have two flavors: floating-stack mode, which allows variable
662 // stack size; and fixed-stack mode. NPTL is always floating-stack.
663 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
664 os::Linux::set_is_floating_stack();
665 }
666 }
668 /////////////////////////////////////////////////////////////////////////////
669 // thread stack
671 // Force Linux kernel to expand current thread stack. If "bottom" is close
672 // to the stack guard, caller should block all signals.
673 //
674 // MAP_GROWSDOWN:
675 // A special mmap() flag that is used to implement thread stacks. It tells
676 // kernel that the memory region should extend downwards when needed. This
677 // allows early versions of LinuxThreads to only mmap the first few pages
678 // when creating a new thread. Linux kernel will automatically expand thread
679 // stack as needed (on page faults).
680 //
681 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
682 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
683 // region, it's hard to tell if the fault is due to a legitimate stack
684 // access or because of reading/writing non-exist memory (e.g. buffer
685 // overrun). As a rule, if the fault happens below current stack pointer,
686 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
687 // application (see Linux kernel fault.c).
688 //
689 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
690 // stack overflow detection.
691 //
692 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
693 // not use this flag. However, the stack of initial thread is not created
694 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
695 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
696 // and then attach the thread to JVM.
697 //
698 // To get around the problem and allow stack banging on Linux, we need to
699 // manually expand thread stack after receiving the SIGSEGV.
700 //
701 // There are two ways to expand thread stack to address "bottom", we used
702 // both of them in JVM before 1.5:
703 // 1. adjust stack pointer first so that it is below "bottom", and then
704 // touch "bottom"
705 // 2. mmap() the page in question
706 //
707 // Now alternate signal stack is gone, it's harder to use 2. For instance,
708 // if current sp is already near the lower end of page 101, and we need to
709 // call mmap() to map page 100, it is possible that part of the mmap() frame
710 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
711 // That will destroy the mmap() frame and cause VM to crash.
712 //
713 // The following code works by adjusting sp first, then accessing the "bottom"
714 // page to force a page fault. Linux kernel will then automatically expand the
715 // stack mapping.
716 //
717 // _expand_stack_to() assumes its frame size is less than page size, which
718 // should always be true if the function is not inlined.
720 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
721 #define NOINLINE
722 #else
723 #define NOINLINE __attribute__ ((noinline))
724 #endif
726 static void _expand_stack_to(address bottom) NOINLINE;
728 static void _expand_stack_to(address bottom) {
729 address sp;
730 size_t size;
731 volatile char *p;
733 // Adjust bottom to point to the largest address within the same page, it
734 // gives us a one-page buffer if alloca() allocates slightly more memory.
735 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
736 bottom += os::Linux::page_size() - 1;
738 // sp might be slightly above current stack pointer; if that's the case, we
739 // will alloca() a little more space than necessary, which is OK. Don't use
740 // os::current_stack_pointer(), as its result can be slightly below current
741 // stack pointer, causing us to not alloca enough to reach "bottom".
742 sp = (address)&sp;
744 if (sp > bottom) {
745 size = sp - bottom;
746 p = (volatile char *)alloca(size);
747 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
748 p[0] = '\0';
749 }
750 }
752 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
753 assert(t!=NULL, "just checking");
754 assert(t->osthread()->expanding_stack(), "expand should be set");
755 assert(t->stack_base() != NULL, "stack_base was not initialized");
757 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
758 sigset_t mask_all, old_sigset;
759 sigfillset(&mask_all);
760 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
761 _expand_stack_to(addr);
762 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
763 return true;
764 }
765 return false;
766 }
768 //////////////////////////////////////////////////////////////////////////////
769 // create new thread
771 static address highest_vm_reserved_address();
773 // check if it's safe to start a new thread
774 static bool _thread_safety_check(Thread* thread) {
775 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
776 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
777 // Heap is mmap'ed at lower end of memory space. Thread stacks are
778 // allocated (MAP_FIXED) from high address space. Every thread stack
779 // occupies a fixed size slot (usually 2Mbytes, but user can change
780 // it to other values if they rebuild LinuxThreads).
781 //
782 // Problem with MAP_FIXED is that mmap() can still succeed even part of
783 // the memory region has already been mmap'ed. That means if we have too
784 // many threads and/or very large heap, eventually thread stack will
785 // collide with heap.
786 //
787 // Here we try to prevent heap/stack collision by comparing current
788 // stack bottom with the highest address that has been mmap'ed by JVM
789 // plus a safety margin for memory maps created by native code.
790 //
791 // This feature can be disabled by setting ThreadSafetyMargin to 0
792 //
793 if (ThreadSafetyMargin > 0) {
794 address stack_bottom = os::current_stack_base() - os::current_stack_size();
796 // not safe if our stack extends below the safety margin
797 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
798 } else {
799 return true;
800 }
801 } else {
802 // Floating stack LinuxThreads or NPTL:
803 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
804 // there's not enough space left, pthread_create() will fail. If we come
805 // here, that means enough space has been reserved for stack.
806 return true;
807 }
808 }
810 // Thread start routine for all newly created threads
811 static void *java_start(Thread *thread) {
812 // Try to randomize the cache line index of hot stack frames.
813 // This helps when threads of the same stack traces evict each other's
814 // cache lines. The threads can be either from the same JVM instance, or
815 // from different JVM instances. The benefit is especially true for
816 // processors with hyperthreading technology.
817 static int counter = 0;
818 int pid = os::current_process_id();
819 alloca(((pid ^ counter++) & 7) * 128);
821 ThreadLocalStorage::set_thread(thread);
823 OSThread* osthread = thread->osthread();
824 Monitor* sync = osthread->startThread_lock();
826 // non floating stack LinuxThreads needs extra check, see above
827 if (!_thread_safety_check(thread)) {
828 // notify parent thread
829 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
830 osthread->set_state(ZOMBIE);
831 sync->notify_all();
832 return NULL;
833 }
835 // thread_id is kernel thread id (similar to Solaris LWP id)
836 osthread->set_thread_id(os::Linux::gettid());
838 if (UseNUMA) {
839 int lgrp_id = os::numa_get_group_id();
840 if (lgrp_id != -1) {
841 thread->set_lgrp_id(lgrp_id);
842 }
843 }
844 // initialize signal mask for this thread
845 os::Linux::hotspot_sigmask(thread);
847 // initialize floating point control register
848 os::Linux::init_thread_fpu_state();
850 // handshaking with parent thread
851 {
852 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
854 // notify parent thread
855 osthread->set_state(INITIALIZED);
856 sync->notify_all();
858 // wait until os::start_thread()
859 while (osthread->get_state() == INITIALIZED) {
860 sync->wait(Mutex::_no_safepoint_check_flag);
861 }
862 }
864 // call one more level start routine
865 thread->run();
867 return 0;
868 }
870 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
871 assert(thread->osthread() == NULL, "caller responsible");
873 // Allocate the OSThread object
874 OSThread* osthread = new OSThread(NULL, NULL);
875 if (osthread == NULL) {
876 return false;
877 }
879 // set the correct thread state
880 osthread->set_thread_type(thr_type);
882 // Initial state is ALLOCATED but not INITIALIZED
883 osthread->set_state(ALLOCATED);
885 thread->set_osthread(osthread);
887 // init thread attributes
888 pthread_attr_t attr;
889 pthread_attr_init(&attr);
890 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
892 // stack size
893 if (os::Linux::supports_variable_stack_size()) {
894 // calculate stack size if it's not specified by caller
895 if (stack_size == 0) {
896 stack_size = os::Linux::default_stack_size(thr_type);
898 switch (thr_type) {
899 case os::java_thread:
900 // Java threads use ThreadStackSize which default value can be
901 // changed with the flag -Xss
902 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
903 stack_size = JavaThread::stack_size_at_create();
904 break;
905 case os::compiler_thread:
906 if (CompilerThreadStackSize > 0) {
907 stack_size = (size_t)(CompilerThreadStackSize * K);
908 break;
909 } // else fall through:
910 // use VMThreadStackSize if CompilerThreadStackSize is not defined
911 case os::vm_thread:
912 case os::pgc_thread:
913 case os::cgc_thread:
914 case os::watcher_thread:
915 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
916 break;
917 }
918 }
920 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
921 pthread_attr_setstacksize(&attr, stack_size);
922 } else {
923 // let pthread_create() pick the default value.
924 }
926 // glibc guard page
927 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
929 ThreadState state;
931 {
932 // Serialize thread creation if we are running with fixed stack LinuxThreads
933 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
934 if (lock) {
935 os::Linux::createThread_lock()->lock_without_safepoint_check();
936 }
938 pthread_t tid;
939 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
941 pthread_attr_destroy(&attr);
943 if (ret != 0) {
944 if (PrintMiscellaneous && (Verbose || WizardMode)) {
945 perror("pthread_create()");
946 }
947 // Need to clean up stuff we've allocated so far
948 thread->set_osthread(NULL);
949 delete osthread;
950 if (lock) os::Linux::createThread_lock()->unlock();
951 return false;
952 }
954 // Store pthread info into the OSThread
955 osthread->set_pthread_id(tid);
957 // Wait until child thread is either initialized or aborted
958 {
959 Monitor* sync_with_child = osthread->startThread_lock();
960 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
961 while ((state = osthread->get_state()) == ALLOCATED) {
962 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
963 }
964 }
966 if (lock) {
967 os::Linux::createThread_lock()->unlock();
968 }
969 }
971 // Aborted due to thread limit being reached
972 if (state == ZOMBIE) {
973 thread->set_osthread(NULL);
974 delete osthread;
975 return false;
976 }
978 // The thread is returned suspended (in state INITIALIZED),
979 // and is started higher up in the call chain
980 assert(state == INITIALIZED, "race condition");
981 return true;
982 }
984 /////////////////////////////////////////////////////////////////////////////
985 // attach existing thread
987 // bootstrap the main thread
988 bool os::create_main_thread(JavaThread* thread) {
989 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
990 return create_attached_thread(thread);
991 }
993 bool os::create_attached_thread(JavaThread* thread) {
994 #ifdef ASSERT
995 thread->verify_not_published();
996 #endif
998 // Allocate the OSThread object
999 OSThread* osthread = new OSThread(NULL, NULL);
1001 if (osthread == NULL) {
1002 return false;
1003 }
1005 // Store pthread info into the OSThread
1006 osthread->set_thread_id(os::Linux::gettid());
1007 osthread->set_pthread_id(::pthread_self());
1009 // initialize floating point control register
1010 os::Linux::init_thread_fpu_state();
1012 // Initial thread state is RUNNABLE
1013 osthread->set_state(RUNNABLE);
1015 thread->set_osthread(osthread);
1017 if (UseNUMA) {
1018 int lgrp_id = os::numa_get_group_id();
1019 if (lgrp_id != -1) {
1020 thread->set_lgrp_id(lgrp_id);
1021 }
1022 }
1024 if (os::Linux::is_initial_thread()) {
1025 // If current thread is initial thread, its stack is mapped on demand,
1026 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
1027 // the entire stack region to avoid SEGV in stack banging.
1028 // It is also useful to get around the heap-stack-gap problem on SuSE
1029 // kernel (see 4821821 for details). We first expand stack to the top
1030 // of yellow zone, then enable stack yellow zone (order is significant,
1031 // enabling yellow zone first will crash JVM on SuSE Linux), so there
1032 // is no gap between the last two virtual memory regions.
1034 JavaThread *jt = (JavaThread *)thread;
1035 address addr = jt->stack_yellow_zone_base();
1036 assert(addr != NULL, "initialization problem?");
1037 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1039 osthread->set_expanding_stack();
1040 os::Linux::manually_expand_stack(jt, addr);
1041 osthread->clear_expanding_stack();
1042 }
1044 // initialize signal mask for this thread
1045 // and save the caller's signal mask
1046 os::Linux::hotspot_sigmask(thread);
1048 return true;
1049 }
1051 void os::pd_start_thread(Thread* thread) {
1052 OSThread * osthread = thread->osthread();
1053 assert(osthread->get_state() != INITIALIZED, "just checking");
1054 Monitor* sync_with_child = osthread->startThread_lock();
1055 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1056 sync_with_child->notify();
1057 }
1059 // Free Linux resources related to the OSThread
1060 void os::free_thread(OSThread* osthread) {
1061 assert(osthread != NULL, "osthread not set");
1063 if (Thread::current()->osthread() == osthread) {
1064 // Restore caller's signal mask
1065 sigset_t sigmask = osthread->caller_sigmask();
1066 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1067 }
1069 delete osthread;
1070 }
1072 //////////////////////////////////////////////////////////////////////////////
1073 // thread local storage
1075 int os::allocate_thread_local_storage() {
1076 pthread_key_t key;
1077 int rslt = pthread_key_create(&key, NULL);
1078 assert(rslt == 0, "cannot allocate thread local storage");
1079 return (int)key;
1080 }
1082 // Note: This is currently not used by VM, as we don't destroy TLS key
1083 // on VM exit.
1084 void os::free_thread_local_storage(int index) {
1085 int rslt = pthread_key_delete((pthread_key_t)index);
1086 assert(rslt == 0, "invalid index");
1087 }
1089 void os::thread_local_storage_at_put(int index, void* value) {
1090 int rslt = pthread_setspecific((pthread_key_t)index, value);
1091 assert(rslt == 0, "pthread_setspecific failed");
1092 }
1094 extern "C" Thread* get_thread() {
1095 return ThreadLocalStorage::thread();
1096 }
1098 //////////////////////////////////////////////////////////////////////////////
1099 // initial thread
1101 // Check if current thread is the initial thread, similar to Solaris thr_main.
1102 bool os::Linux::is_initial_thread(void) {
1103 char dummy;
1104 // If called before init complete, thread stack bottom will be null.
1105 // Can be called if fatal error occurs before initialization.
1106 if (initial_thread_stack_bottom() == NULL) return false;
1107 assert(initial_thread_stack_bottom() != NULL &&
1108 initial_thread_stack_size() != 0,
1109 "os::init did not locate initial thread's stack region");
1110 if ((address)&dummy >= initial_thread_stack_bottom() &&
1111 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1112 return true;
1113 else return false;
1114 }
1116 // Find the virtual memory area that contains addr
1117 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1118 FILE *fp = fopen("/proc/self/maps", "r");
1119 if (fp) {
1120 address low, high;
1121 while (!feof(fp)) {
1122 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1123 if (low <= addr && addr < high) {
1124 if (vma_low) *vma_low = low;
1125 if (vma_high) *vma_high = high;
1126 fclose (fp);
1127 return true;
1128 }
1129 }
1130 for (;;) {
1131 int ch = fgetc(fp);
1132 if (ch == EOF || ch == (int)'\n') break;
1133 }
1134 }
1135 fclose(fp);
1136 }
1137 return false;
1138 }
1140 // Locate initial thread stack. This special handling of initial thread stack
1141 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1142 // bogus value for initial thread.
1143 void os::Linux::capture_initial_stack(size_t max_size) {
1144 // stack size is the easy part, get it from RLIMIT_STACK
1145 size_t stack_size;
1146 struct rlimit rlim;
1147 getrlimit(RLIMIT_STACK, &rlim);
1148 stack_size = rlim.rlim_cur;
1150 // 6308388: a bug in ld.so will relocate its own .data section to the
1151 // lower end of primordial stack; reduce ulimit -s value a little bit
1152 // so we won't install guard page on ld.so's data section.
1153 stack_size -= 2 * page_size();
1155 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1156 // 7.1, in both cases we will get 2G in return value.
1157 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1158 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1159 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1160 // in case other parts in glibc still assumes 2M max stack size.
1161 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1162 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1163 if (stack_size > 2 * K * K IA64_ONLY(*2))
1164 stack_size = 2 * K * K IA64_ONLY(*2);
1165 // Try to figure out where the stack base (top) is. This is harder.
1166 //
1167 // When an application is started, glibc saves the initial stack pointer in
1168 // a global variable "__libc_stack_end", which is then used by system
1169 // libraries. __libc_stack_end should be pretty close to stack top. The
1170 // variable is available since the very early days. However, because it is
1171 // a private interface, it could disappear in the future.
1172 //
1173 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1174 // to __libc_stack_end, it is very close to stack top, but isn't the real
1175 // stack top. Note that /proc may not exist if VM is running as a chroot
1176 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1177 // /proc/<pid>/stat could change in the future (though unlikely).
1178 //
1179 // We try __libc_stack_end first. If that doesn't work, look for
1180 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1181 // as a hint, which should work well in most cases.
1183 uintptr_t stack_start;
1185 // try __libc_stack_end first
1186 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1187 if (p && *p) {
1188 stack_start = *p;
1189 } else {
1190 // see if we can get the start_stack field from /proc/self/stat
1191 FILE *fp;
1192 int pid;
1193 char state;
1194 int ppid;
1195 int pgrp;
1196 int session;
1197 int nr;
1198 int tpgrp;
1199 unsigned long flags;
1200 unsigned long minflt;
1201 unsigned long cminflt;
1202 unsigned long majflt;
1203 unsigned long cmajflt;
1204 unsigned long utime;
1205 unsigned long stime;
1206 long cutime;
1207 long cstime;
1208 long prio;
1209 long nice;
1210 long junk;
1211 long it_real;
1212 uintptr_t start;
1213 uintptr_t vsize;
1214 intptr_t rss;
1215 uintptr_t rsslim;
1216 uintptr_t scodes;
1217 uintptr_t ecode;
1218 int i;
1220 // Figure what the primordial thread stack base is. Code is inspired
1221 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1222 // followed by command name surrounded by parentheses, state, etc.
1223 char stat[2048];
1224 int statlen;
1226 fp = fopen("/proc/self/stat", "r");
1227 if (fp) {
1228 statlen = fread(stat, 1, 2047, fp);
1229 stat[statlen] = '\0';
1230 fclose(fp);
1232 // Skip pid and the command string. Note that we could be dealing with
1233 // weird command names, e.g. user could decide to rename java launcher
1234 // to "java 1.4.2 :)", then the stat file would look like
1235 // 1234 (java 1.4.2 :)) R ... ...
1236 // We don't really need to know the command string, just find the last
1237 // occurrence of ")" and then start parsing from there. See bug 4726580.
1238 char * s = strrchr(stat, ')');
1240 i = 0;
1241 if (s) {
1242 // Skip blank chars
1243 do s++; while (isspace(*s));
1245 #define _UFM UINTX_FORMAT
1246 #define _DFM INTX_FORMAT
1248 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1249 /* 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 */
1250 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,
1251 &state, /* 3 %c */
1252 &ppid, /* 4 %d */
1253 &pgrp, /* 5 %d */
1254 &session, /* 6 %d */
1255 &nr, /* 7 %d */
1256 &tpgrp, /* 8 %d */
1257 &flags, /* 9 %lu */
1258 &minflt, /* 10 %lu */
1259 &cminflt, /* 11 %lu */
1260 &majflt, /* 12 %lu */
1261 &cmajflt, /* 13 %lu */
1262 &utime, /* 14 %lu */
1263 &stime, /* 15 %lu */
1264 &cutime, /* 16 %ld */
1265 &cstime, /* 17 %ld */
1266 &prio, /* 18 %ld */
1267 &nice, /* 19 %ld */
1268 &junk, /* 20 %ld */
1269 &it_real, /* 21 %ld */
1270 &start, /* 22 UINTX_FORMAT */
1271 &vsize, /* 23 UINTX_FORMAT */
1272 &rss, /* 24 INTX_FORMAT */
1273 &rsslim, /* 25 UINTX_FORMAT */
1274 &scodes, /* 26 UINTX_FORMAT */
1275 &ecode, /* 27 UINTX_FORMAT */
1276 &stack_start); /* 28 UINTX_FORMAT */
1277 }
1279 #undef _UFM
1280 #undef _DFM
1282 if (i != 28 - 2) {
1283 assert(false, "Bad conversion from /proc/self/stat");
1284 // product mode - assume we are the initial thread, good luck in the
1285 // embedded case.
1286 warning("Can't detect initial thread stack location - bad conversion");
1287 stack_start = (uintptr_t) &rlim;
1288 }
1289 } else {
1290 // For some reason we can't open /proc/self/stat (for example, running on
1291 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1292 // most cases, so don't abort:
1293 warning("Can't detect initial thread stack location - no /proc/self/stat");
1294 stack_start = (uintptr_t) &rlim;
1295 }
1296 }
1298 // Now we have a pointer (stack_start) very close to the stack top, the
1299 // next thing to do is to figure out the exact location of stack top. We
1300 // can find out the virtual memory area that contains stack_start by
1301 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1302 // and its upper limit is the real stack top. (again, this would fail if
1303 // running inside chroot, because /proc may not exist.)
1305 uintptr_t stack_top;
1306 address low, high;
1307 if (find_vma((address)stack_start, &low, &high)) {
1308 // success, "high" is the true stack top. (ignore "low", because initial
1309 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1310 stack_top = (uintptr_t)high;
1311 } else {
1312 // failed, likely because /proc/self/maps does not exist
1313 warning("Can't detect initial thread stack location - find_vma failed");
1314 // best effort: stack_start is normally within a few pages below the real
1315 // stack top, use it as stack top, and reduce stack size so we won't put
1316 // guard page outside stack.
1317 stack_top = stack_start;
1318 stack_size -= 16 * page_size();
1319 }
1321 // stack_top could be partially down the page so align it
1322 stack_top = align_size_up(stack_top, page_size());
1324 if (max_size && stack_size > max_size) {
1325 _initial_thread_stack_size = max_size;
1326 } else {
1327 _initial_thread_stack_size = stack_size;
1328 }
1330 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1331 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1332 }
1334 ////////////////////////////////////////////////////////////////////////////////
1335 // time support
1337 // Time since start-up in seconds to a fine granularity.
1338 // Used by VMSelfDestructTimer and the MemProfiler.
1339 double os::elapsedTime() {
1341 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1342 }
1344 jlong os::elapsed_counter() {
1345 return javaTimeNanos() - initial_time_count;
1346 }
1348 jlong os::elapsed_frequency() {
1349 return NANOSECS_PER_SEC; // nanosecond resolution
1350 }
1352 bool os::supports_vtime() { return true; }
1353 bool os::enable_vtime() { return false; }
1354 bool os::vtime_enabled() { return false; }
1356 double os::elapsedVTime() {
1357 struct rusage usage;
1358 int retval = getrusage(RUSAGE_THREAD, &usage);
1359 if (retval == 0) {
1360 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);
1361 } else {
1362 // better than nothing, but not much
1363 return elapsedTime();
1364 }
1365 }
1367 jlong os::javaTimeMillis() {
1368 timeval time;
1369 int status = gettimeofday(&time, NULL);
1370 assert(status != -1, "linux error");
1371 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1372 }
1374 #ifndef CLOCK_MONOTONIC
1375 #define CLOCK_MONOTONIC (1)
1376 #endif
1378 void os::Linux::clock_init() {
1379 // we do dlopen's in this particular order due to bug in linux
1380 // dynamical loader (see 6348968) leading to crash on exit
1381 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1382 if (handle == NULL) {
1383 handle = dlopen("librt.so", RTLD_LAZY);
1384 }
1386 if (handle) {
1387 int (*clock_getres_func)(clockid_t, struct timespec*) =
1388 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1389 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1390 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1391 if (clock_getres_func && clock_gettime_func) {
1392 // See if monotonic clock is supported by the kernel. Note that some
1393 // early implementations simply return kernel jiffies (updated every
1394 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1395 // for nano time (though the monotonic property is still nice to have).
1396 // It's fixed in newer kernels, however clock_getres() still returns
1397 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1398 // resolution for now. Hopefully as people move to new kernels, this
1399 // won't be a problem.
1400 struct timespec res;
1401 struct timespec tp;
1402 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1403 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1404 // yes, monotonic clock is supported
1405 _clock_gettime = clock_gettime_func;
1406 return;
1407 } else {
1408 // close librt if there is no monotonic clock
1409 dlclose(handle);
1410 }
1411 }
1412 }
1413 warning("No monotonic clock was available - timed services may " \
1414 "be adversely affected if the time-of-day clock changes");
1415 }
1417 #ifndef SYS_clock_getres
1419 #if defined(IA32) || defined(AMD64)
1420 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1421 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1422 #else
1423 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1424 #define sys_clock_getres(x,y) -1
1425 #endif
1427 #else
1428 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1429 #endif
1431 void os::Linux::fast_thread_clock_init() {
1432 if (!UseLinuxPosixThreadCPUClocks) {
1433 return;
1434 }
1435 clockid_t clockid;
1436 struct timespec tp;
1437 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1438 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1440 // Switch to using fast clocks for thread cpu time if
1441 // the sys_clock_getres() returns 0 error code.
1442 // Note, that some kernels may support the current thread
1443 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1444 // returned by the pthread_getcpuclockid().
1445 // If the fast Posix clocks are supported then the sys_clock_getres()
1446 // must return at least tp.tv_sec == 0 which means a resolution
1447 // better than 1 sec. This is extra check for reliability.
1449 if(pthread_getcpuclockid_func &&
1450 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1451 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1453 _supports_fast_thread_cpu_time = true;
1454 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1455 }
1456 }
1458 jlong os::javaTimeNanos() {
1459 if (Linux::supports_monotonic_clock()) {
1460 struct timespec tp;
1461 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1462 assert(status == 0, "gettime error");
1463 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1464 return result;
1465 } else {
1466 timeval time;
1467 int status = gettimeofday(&time, NULL);
1468 assert(status != -1, "linux error");
1469 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1470 return 1000 * usecs;
1471 }
1472 }
1474 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1475 if (Linux::supports_monotonic_clock()) {
1476 info_ptr->max_value = ALL_64_BITS;
1478 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1479 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1480 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1481 } else {
1482 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1483 info_ptr->max_value = ALL_64_BITS;
1485 // gettimeofday is a real time clock so it skips
1486 info_ptr->may_skip_backward = true;
1487 info_ptr->may_skip_forward = true;
1488 }
1490 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1491 }
1493 // Return the real, user, and system times in seconds from an
1494 // arbitrary fixed point in the past.
1495 bool os::getTimesSecs(double* process_real_time,
1496 double* process_user_time,
1497 double* process_system_time) {
1498 struct tms ticks;
1499 clock_t real_ticks = times(&ticks);
1501 if (real_ticks == (clock_t) (-1)) {
1502 return false;
1503 } else {
1504 double ticks_per_second = (double) clock_tics_per_sec;
1505 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1506 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1507 *process_real_time = ((double) real_ticks) / ticks_per_second;
1509 return true;
1510 }
1511 }
1514 char * os::local_time_string(char *buf, size_t buflen) {
1515 struct tm t;
1516 time_t long_time;
1517 time(&long_time);
1518 localtime_r(&long_time, &t);
1519 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1520 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1521 t.tm_hour, t.tm_min, t.tm_sec);
1522 return buf;
1523 }
1525 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1526 return localtime_r(clock, res);
1527 }
1529 ////////////////////////////////////////////////////////////////////////////////
1530 // runtime exit support
1532 // Note: os::shutdown() might be called very early during initialization, or
1533 // called from signal handler. Before adding something to os::shutdown(), make
1534 // sure it is async-safe and can handle partially initialized VM.
1535 void os::shutdown() {
1537 // allow PerfMemory to attempt cleanup of any persistent resources
1538 perfMemory_exit();
1540 // needs to remove object in file system
1541 AttachListener::abort();
1543 // flush buffered output, finish log files
1544 ostream_abort();
1546 // Check for abort hook
1547 abort_hook_t abort_hook = Arguments::abort_hook();
1548 if (abort_hook != NULL) {
1549 abort_hook();
1550 }
1552 }
1554 // Note: os::abort() might be called very early during initialization, or
1555 // called from signal handler. Before adding something to os::abort(), make
1556 // sure it is async-safe and can handle partially initialized VM.
1557 void os::abort(bool dump_core) {
1558 os::shutdown();
1559 if (dump_core) {
1560 #ifndef PRODUCT
1561 fdStream out(defaultStream::output_fd());
1562 out.print_raw("Current thread is ");
1563 char buf[16];
1564 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1565 out.print_raw_cr(buf);
1566 out.print_raw_cr("Dumping core ...");
1567 #endif
1568 ::abort(); // dump core
1569 }
1571 ::exit(1);
1572 }
1574 // Die immediately, no exit hook, no abort hook, no cleanup.
1575 void os::die() {
1576 // _exit() on LinuxThreads only kills current thread
1577 ::abort();
1578 }
1580 // unused on linux for now.
1581 void os::set_error_file(const char *logfile) {}
1584 // This method is a copy of JDK's sysGetLastErrorString
1585 // from src/solaris/hpi/src/system_md.c
1587 size_t os::lasterror(char *buf, size_t len) {
1589 if (errno == 0) return 0;
1591 const char *s = ::strerror(errno);
1592 size_t n = ::strlen(s);
1593 if (n >= len) {
1594 n = len - 1;
1595 }
1596 ::strncpy(buf, s, n);
1597 buf[n] = '\0';
1598 return n;
1599 }
1601 intx os::current_thread_id() { return (intx)pthread_self(); }
1602 int os::current_process_id() {
1604 // Under the old linux thread library, linux gives each thread
1605 // its own process id. Because of this each thread will return
1606 // a different pid if this method were to return the result
1607 // of getpid(2). Linux provides no api that returns the pid
1608 // of the launcher thread for the vm. This implementation
1609 // returns a unique pid, the pid of the launcher thread
1610 // that starts the vm 'process'.
1612 // Under the NPTL, getpid() returns the same pid as the
1613 // launcher thread rather than a unique pid per thread.
1614 // Use gettid() if you want the old pre NPTL behaviour.
1616 // if you are looking for the result of a call to getpid() that
1617 // returns a unique pid for the calling thread, then look at the
1618 // OSThread::thread_id() method in osThread_linux.hpp file
1620 return (int)(_initial_pid ? _initial_pid : getpid());
1621 }
1623 // DLL functions
1625 const char* os::dll_file_extension() { return ".so"; }
1627 // This must be hard coded because it's the system's temporary
1628 // directory not the java application's temp directory, ala java.io.tmpdir.
1629 const char* os::get_temp_directory() { return "/tmp"; }
1631 static bool file_exists(const char* filename) {
1632 struct stat statbuf;
1633 if (filename == NULL || strlen(filename) == 0) {
1634 return false;
1635 }
1636 return os::stat(filename, &statbuf) == 0;
1637 }
1639 bool os::dll_build_name(char* buffer, size_t buflen,
1640 const char* pname, const char* fname) {
1641 bool retval = false;
1642 // Copied from libhpi
1643 const size_t pnamelen = pname ? strlen(pname) : 0;
1645 // Return error on buffer overflow.
1646 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1647 return retval;
1648 }
1650 if (pnamelen == 0) {
1651 snprintf(buffer, buflen, "lib%s.so", fname);
1652 retval = true;
1653 } else if (strchr(pname, *os::path_separator()) != NULL) {
1654 int n;
1655 char** pelements = split_path(pname, &n);
1656 if (pelements == NULL) {
1657 return false;
1658 }
1659 for (int i = 0 ; i < n ; i++) {
1660 // Really shouldn't be NULL, but check can't hurt
1661 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1662 continue; // skip the empty path values
1663 }
1664 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1665 if (file_exists(buffer)) {
1666 retval = true;
1667 break;
1668 }
1669 }
1670 // release the storage
1671 for (int i = 0 ; i < n ; i++) {
1672 if (pelements[i] != NULL) {
1673 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1674 }
1675 }
1676 if (pelements != NULL) {
1677 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1678 }
1679 } else {
1680 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1681 retval = true;
1682 }
1683 return retval;
1684 }
1686 // check if addr is inside libjvm.so
1687 bool os::address_is_in_vm(address addr) {
1688 static address libjvm_base_addr;
1689 Dl_info dlinfo;
1691 if (libjvm_base_addr == NULL) {
1692 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1693 libjvm_base_addr = (address)dlinfo.dli_fbase;
1694 }
1695 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1696 }
1698 if (dladdr((void *)addr, &dlinfo) != 0) {
1699 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1700 }
1702 return false;
1703 }
1705 bool os::dll_address_to_function_name(address addr, char *buf,
1706 int buflen, int *offset) {
1707 // buf is not optional, but offset is optional
1708 assert(buf != NULL, "sanity check");
1710 Dl_info dlinfo;
1712 if (dladdr((void*)addr, &dlinfo) != 0) {
1713 // see if we have a matching symbol
1714 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1715 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1716 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1717 }
1718 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1719 return true;
1720 }
1721 // no matching symbol so try for just file info
1722 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1723 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1724 buf, buflen, offset, dlinfo.dli_fname)) {
1725 return true;
1726 }
1727 }
1728 }
1730 buf[0] = '\0';
1731 if (offset != NULL) *offset = -1;
1732 return false;
1733 }
1735 struct _address_to_library_name {
1736 address addr; // input : memory address
1737 size_t buflen; // size of fname
1738 char* fname; // output: library name
1739 address base; // library base addr
1740 };
1742 static int address_to_library_name_callback(struct dl_phdr_info *info,
1743 size_t size, void *data) {
1744 int i;
1745 bool found = false;
1746 address libbase = NULL;
1747 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1749 // iterate through all loadable segments
1750 for (i = 0; i < info->dlpi_phnum; i++) {
1751 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1752 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1753 // base address of a library is the lowest address of its loaded
1754 // segments.
1755 if (libbase == NULL || libbase > segbase) {
1756 libbase = segbase;
1757 }
1758 // see if 'addr' is within current segment
1759 if (segbase <= d->addr &&
1760 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1761 found = true;
1762 }
1763 }
1764 }
1766 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1767 // so dll_address_to_library_name() can fall through to use dladdr() which
1768 // can figure out executable name from argv[0].
1769 if (found && info->dlpi_name && info->dlpi_name[0]) {
1770 d->base = libbase;
1771 if (d->fname) {
1772 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1773 }
1774 return 1;
1775 }
1776 return 0;
1777 }
1779 bool os::dll_address_to_library_name(address addr, char* buf,
1780 int buflen, int* offset) {
1781 // buf is not optional, but offset is optional
1782 assert(buf != NULL, "sanity check");
1784 Dl_info dlinfo;
1785 struct _address_to_library_name data;
1787 // There is a bug in old glibc dladdr() implementation that it could resolve
1788 // to wrong library name if the .so file has a base address != NULL. Here
1789 // we iterate through the program headers of all loaded libraries to find
1790 // out which library 'addr' really belongs to. This workaround can be
1791 // removed once the minimum requirement for glibc is moved to 2.3.x.
1792 data.addr = addr;
1793 data.fname = buf;
1794 data.buflen = buflen;
1795 data.base = NULL;
1796 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1798 if (rslt) {
1799 // buf already contains library name
1800 if (offset) *offset = addr - data.base;
1801 return true;
1802 }
1803 if (dladdr((void*)addr, &dlinfo) != 0) {
1804 if (dlinfo.dli_fname != NULL) {
1805 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1806 }
1807 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1808 *offset = addr - (address)dlinfo.dli_fbase;
1809 }
1810 return true;
1811 }
1813 buf[0] = '\0';
1814 if (offset) *offset = -1;
1815 return false;
1816 }
1818 // Loads .dll/.so and
1819 // in case of error it checks if .dll/.so was built for the
1820 // same architecture as Hotspot is running on
1823 // Remember the stack's state. The Linux dynamic linker will change
1824 // the stack to 'executable' at most once, so we must safepoint only once.
1825 bool os::Linux::_stack_is_executable = false;
1827 // VM operation that loads a library. This is necessary if stack protection
1828 // of the Java stacks can be lost during loading the library. If we
1829 // do not stop the Java threads, they can stack overflow before the stacks
1830 // are protected again.
1831 class VM_LinuxDllLoad: public VM_Operation {
1832 private:
1833 const char *_filename;
1834 char *_ebuf;
1835 int _ebuflen;
1836 void *_lib;
1837 public:
1838 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1839 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1840 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1841 void doit() {
1842 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1843 os::Linux::_stack_is_executable = true;
1844 }
1845 void* loaded_library() { return _lib; }
1846 };
1848 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1849 {
1850 void * result = NULL;
1851 bool load_attempted = false;
1853 // Check whether the library to load might change execution rights
1854 // of the stack. If they are changed, the protection of the stack
1855 // guard pages will be lost. We need a safepoint to fix this.
1856 //
1857 // See Linux man page execstack(8) for more info.
1858 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1859 ElfFile ef(filename);
1860 if (!ef.specifies_noexecstack()) {
1861 if (!is_init_completed()) {
1862 os::Linux::_stack_is_executable = true;
1863 // This is OK - No Java threads have been created yet, and hence no
1864 // stack guard pages to fix.
1865 //
1866 // This should happen only when you are building JDK7 using a very
1867 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1868 //
1869 // Dynamic loader will make all stacks executable after
1870 // this function returns, and will not do that again.
1871 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1872 } else {
1873 warning("You have loaded library %s which might have disabled stack guard. "
1874 "The VM will try to fix the stack guard now.\n"
1875 "It's highly recommended that you fix the library with "
1876 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1877 filename);
1879 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1880 JavaThread *jt = JavaThread::current();
1881 if (jt->thread_state() != _thread_in_native) {
1882 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1883 // that requires ExecStack. Cannot enter safe point. Let's give up.
1884 warning("Unable to fix stack guard. Giving up.");
1885 } else {
1886 if (!LoadExecStackDllInVMThread) {
1887 // This is for the case where the DLL has an static
1888 // constructor function that executes JNI code. We cannot
1889 // load such DLLs in the VMThread.
1890 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1891 }
1893 ThreadInVMfromNative tiv(jt);
1894 debug_only(VMNativeEntryWrapper vew;)
1896 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1897 VMThread::execute(&op);
1898 if (LoadExecStackDllInVMThread) {
1899 result = op.loaded_library();
1900 }
1901 load_attempted = true;
1902 }
1903 }
1904 }
1905 }
1907 if (!load_attempted) {
1908 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1909 }
1911 if (result != NULL) {
1912 // Successful loading
1913 return result;
1914 }
1916 Elf32_Ehdr elf_head;
1917 int diag_msg_max_length=ebuflen-strlen(ebuf);
1918 char* diag_msg_buf=ebuf+strlen(ebuf);
1920 if (diag_msg_max_length==0) {
1921 // No more space in ebuf for additional diagnostics message
1922 return NULL;
1923 }
1926 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1928 if (file_descriptor < 0) {
1929 // Can't open library, report dlerror() message
1930 return NULL;
1931 }
1933 bool failed_to_read_elf_head=
1934 (sizeof(elf_head)!=
1935 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1937 ::close(file_descriptor);
1938 if (failed_to_read_elf_head) {
1939 // file i/o error - report dlerror() msg
1940 return NULL;
1941 }
1943 typedef struct {
1944 Elf32_Half code; // Actual value as defined in elf.h
1945 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1946 char elf_class; // 32 or 64 bit
1947 char endianess; // MSB or LSB
1948 char* name; // String representation
1949 } arch_t;
1951 #ifndef EM_486
1952 #define EM_486 6 /* Intel 80486 */
1953 #endif
1955 static const arch_t arch_array[]={
1956 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1957 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1958 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1959 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1960 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1961 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1962 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1963 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1964 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1965 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1966 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1967 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1968 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1969 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1970 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1971 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1972 };
1974 #if (defined IA32)
1975 static Elf32_Half running_arch_code=EM_386;
1976 #elif (defined AMD64)
1977 static Elf32_Half running_arch_code=EM_X86_64;
1978 #elif (defined IA64)
1979 static Elf32_Half running_arch_code=EM_IA_64;
1980 #elif (defined __sparc) && (defined _LP64)
1981 static Elf32_Half running_arch_code=EM_SPARCV9;
1982 #elif (defined __sparc) && (!defined _LP64)
1983 static Elf32_Half running_arch_code=EM_SPARC;
1984 #elif (defined __powerpc64__)
1985 static Elf32_Half running_arch_code=EM_PPC64;
1986 #elif (defined __powerpc__)
1987 static Elf32_Half running_arch_code=EM_PPC;
1988 #elif (defined ARM)
1989 static Elf32_Half running_arch_code=EM_ARM;
1990 #elif (defined S390)
1991 static Elf32_Half running_arch_code=EM_S390;
1992 #elif (defined ALPHA)
1993 static Elf32_Half running_arch_code=EM_ALPHA;
1994 #elif (defined MIPSEL)
1995 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1996 #elif (defined PARISC)
1997 static Elf32_Half running_arch_code=EM_PARISC;
1998 #elif (defined MIPS)
1999 static Elf32_Half running_arch_code=EM_MIPS;
2000 #elif (defined M68K)
2001 static Elf32_Half running_arch_code=EM_68K;
2002 #else
2003 #error Method os::dll_load requires that one of following is defined:\
2004 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
2005 #endif
2007 // Identify compatability class for VM's architecture and library's architecture
2008 // Obtain string descriptions for architectures
2010 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
2011 int running_arch_index=-1;
2013 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
2014 if (running_arch_code == arch_array[i].code) {
2015 running_arch_index = i;
2016 }
2017 if (lib_arch.code == arch_array[i].code) {
2018 lib_arch.compat_class = arch_array[i].compat_class;
2019 lib_arch.name = arch_array[i].name;
2020 }
2021 }
2023 assert(running_arch_index != -1,
2024 "Didn't find running architecture code (running_arch_code) in arch_array");
2025 if (running_arch_index == -1) {
2026 // Even though running architecture detection failed
2027 // we may still continue with reporting dlerror() message
2028 return NULL;
2029 }
2031 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2032 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2033 return NULL;
2034 }
2036 #ifndef S390
2037 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2038 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2039 return NULL;
2040 }
2041 #endif // !S390
2043 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2044 if ( lib_arch.name!=NULL ) {
2045 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2046 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2047 lib_arch.name, arch_array[running_arch_index].name);
2048 } else {
2049 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2050 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2051 lib_arch.code,
2052 arch_array[running_arch_index].name);
2053 }
2054 }
2056 return NULL;
2057 }
2059 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2060 void * result = ::dlopen(filename, RTLD_LAZY);
2061 if (result == NULL) {
2062 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2063 ebuf[ebuflen-1] = '\0';
2064 }
2065 return result;
2066 }
2068 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2069 void * result = NULL;
2070 if (LoadExecStackDllInVMThread) {
2071 result = dlopen_helper(filename, ebuf, ebuflen);
2072 }
2074 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2075 // library that requires an executable stack, or which does not have this
2076 // stack attribute set, dlopen changes the stack attribute to executable. The
2077 // read protection of the guard pages gets lost.
2078 //
2079 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2080 // may have been queued at the same time.
2082 if (!_stack_is_executable) {
2083 JavaThread *jt = Threads::first();
2085 while (jt) {
2086 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2087 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2088 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2089 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2090 warning("Attempt to reguard stack yellow zone failed.");
2091 }
2092 }
2093 jt = jt->next();
2094 }
2095 }
2097 return result;
2098 }
2100 /*
2101 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2102 * chances are you might want to run the generated bits against glibc-2.0
2103 * libdl.so, so always use locking for any version of glibc.
2104 */
2105 void* os::dll_lookup(void* handle, const char* name) {
2106 pthread_mutex_lock(&dl_mutex);
2107 void* res = dlsym(handle, name);
2108 pthread_mutex_unlock(&dl_mutex);
2109 return res;
2110 }
2112 void* os::get_default_process_handle() {
2113 return (void*)::dlopen(NULL, RTLD_LAZY);
2114 }
2116 static bool _print_ascii_file(const char* filename, outputStream* st) {
2117 int fd = ::open(filename, O_RDONLY);
2118 if (fd == -1) {
2119 return false;
2120 }
2122 char buf[32];
2123 int bytes;
2124 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2125 st->print_raw(buf, bytes);
2126 }
2128 ::close(fd);
2130 return true;
2131 }
2133 void os::print_dll_info(outputStream *st) {
2134 st->print_cr("Dynamic libraries:");
2136 char fname[32];
2137 pid_t pid = os::Linux::gettid();
2139 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2141 if (!_print_ascii_file(fname, st)) {
2142 st->print("Can not get library information for pid = %d\n", pid);
2143 }
2144 }
2146 void os::print_os_info_brief(outputStream* st) {
2147 os::Linux::print_distro_info(st);
2149 os::Posix::print_uname_info(st);
2151 os::Linux::print_libversion_info(st);
2153 }
2155 void os::print_os_info(outputStream* st) {
2156 st->print("OS:");
2158 os::Linux::print_distro_info(st);
2160 os::Posix::print_uname_info(st);
2162 // Print warning if unsafe chroot environment detected
2163 if (unsafe_chroot_detected) {
2164 st->print("WARNING!! ");
2165 st->print_cr(unstable_chroot_error);
2166 }
2168 os::Linux::print_libversion_info(st);
2170 os::Posix::print_rlimit_info(st);
2172 os::Posix::print_load_average(st);
2174 os::Linux::print_full_memory_info(st);
2175 }
2177 // Try to identify popular distros.
2178 // Most Linux distributions have a /etc/XXX-release file, which contains
2179 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2180 // file that also contains the OS version string. Some have more than one
2181 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2182 // /etc/redhat-release.), so the order is important.
2183 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2184 // their own specific XXX-release file as well as a redhat-release file.
2185 // Because of this the XXX-release file needs to be searched for before the
2186 // redhat-release file.
2187 // Since Red Hat has a lsb-release file that is not very descriptive the
2188 // search for redhat-release needs to be before lsb-release.
2189 // Since the lsb-release file is the new standard it needs to be searched
2190 // before the older style release files.
2191 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2192 // next to last resort. The os-release file is a new standard that contains
2193 // distribution information and the system-release file seems to be an old
2194 // standard that has been replaced by the lsb-release and os-release files.
2195 // Searching for the debian_version file is the last resort. It contains
2196 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2197 // "Debian " is printed before the contents of the debian_version file.
2198 void os::Linux::print_distro_info(outputStream* st) {
2199 if (!_print_ascii_file("/etc/oracle-release", st) &&
2200 !_print_ascii_file("/etc/mandriva-release", st) &&
2201 !_print_ascii_file("/etc/mandrake-release", st) &&
2202 !_print_ascii_file("/etc/sun-release", st) &&
2203 !_print_ascii_file("/etc/redhat-release", st) &&
2204 !_print_ascii_file("/etc/lsb-release", st) &&
2205 !_print_ascii_file("/etc/SuSE-release", st) &&
2206 !_print_ascii_file("/etc/turbolinux-release", st) &&
2207 !_print_ascii_file("/etc/gentoo-release", st) &&
2208 !_print_ascii_file("/etc/ltib-release", st) &&
2209 !_print_ascii_file("/etc/angstrom-version", st) &&
2210 !_print_ascii_file("/etc/system-release", st) &&
2211 !_print_ascii_file("/etc/os-release", st)) {
2213 if (file_exists("/etc/debian_version")) {
2214 st->print("Debian ");
2215 _print_ascii_file("/etc/debian_version", st);
2216 } else {
2217 st->print("Linux");
2218 }
2219 }
2220 st->cr();
2221 }
2223 void os::Linux::print_libversion_info(outputStream* st) {
2224 // libc, pthread
2225 st->print("libc:");
2226 st->print(os::Linux::glibc_version()); st->print(" ");
2227 st->print(os::Linux::libpthread_version()); st->print(" ");
2228 if (os::Linux::is_LinuxThreads()) {
2229 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2230 }
2231 st->cr();
2232 }
2234 void os::Linux::print_full_memory_info(outputStream* st) {
2235 st->print("\n/proc/meminfo:\n");
2236 _print_ascii_file("/proc/meminfo", st);
2237 st->cr();
2238 }
2240 void os::print_memory_info(outputStream* st) {
2242 st->print("Memory:");
2243 st->print(" %dk page", os::vm_page_size()>>10);
2245 // values in struct sysinfo are "unsigned long"
2246 struct sysinfo si;
2247 sysinfo(&si);
2249 st->print(", physical " UINT64_FORMAT "k",
2250 os::physical_memory() >> 10);
2251 st->print("(" UINT64_FORMAT "k free)",
2252 os::available_memory() >> 10);
2253 st->print(", swap " UINT64_FORMAT "k",
2254 ((jlong)si.totalswap * si.mem_unit) >> 10);
2255 st->print("(" UINT64_FORMAT "k free)",
2256 ((jlong)si.freeswap * si.mem_unit) >> 10);
2257 st->cr();
2258 }
2260 void os::pd_print_cpu_info(outputStream* st) {
2261 st->print("\n/proc/cpuinfo:\n");
2262 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2263 st->print(" <Not Available>");
2264 }
2265 st->cr();
2266 }
2268 void os::print_siginfo(outputStream* st, void* siginfo) {
2269 const siginfo_t* si = (const siginfo_t*)siginfo;
2271 os::Posix::print_siginfo_brief(st, si);
2273 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2274 UseSharedSpaces) {
2275 FileMapInfo* mapinfo = FileMapInfo::current_info();
2276 if (mapinfo->is_in_shared_space(si->si_addr)) {
2277 st->print("\n\nError accessing class data sharing archive." \
2278 " Mapped file inaccessible during execution, " \
2279 " possible disk/network problem.");
2280 }
2281 }
2282 st->cr();
2283 }
2286 static void print_signal_handler(outputStream* st, int sig,
2287 char* buf, size_t buflen);
2289 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2290 st->print_cr("Signal Handlers:");
2291 print_signal_handler(st, SIGSEGV, buf, buflen);
2292 print_signal_handler(st, SIGBUS , buf, buflen);
2293 print_signal_handler(st, SIGFPE , buf, buflen);
2294 print_signal_handler(st, SIGPIPE, buf, buflen);
2295 print_signal_handler(st, SIGXFSZ, buf, buflen);
2296 print_signal_handler(st, SIGILL , buf, buflen);
2297 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2298 print_signal_handler(st, SR_signum, buf, buflen);
2299 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2300 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2301 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2302 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2303 #if defined(PPC64)
2304 print_signal_handler(st, SIGTRAP, buf, buflen);
2305 #endif
2306 }
2308 static char saved_jvm_path[MAXPATHLEN] = {0};
2310 // Find the full path to the current module, libjvm.so
2311 void os::jvm_path(char *buf, jint buflen) {
2312 // Error checking.
2313 if (buflen < MAXPATHLEN) {
2314 assert(false, "must use a large-enough buffer");
2315 buf[0] = '\0';
2316 return;
2317 }
2318 // Lazy resolve the path to current module.
2319 if (saved_jvm_path[0] != 0) {
2320 strcpy(buf, saved_jvm_path);
2321 return;
2322 }
2324 char dli_fname[MAXPATHLEN];
2325 bool ret = dll_address_to_library_name(
2326 CAST_FROM_FN_PTR(address, os::jvm_path),
2327 dli_fname, sizeof(dli_fname), NULL);
2328 assert(ret, "cannot locate libjvm");
2329 char *rp = NULL;
2330 if (ret && dli_fname[0] != '\0') {
2331 rp = realpath(dli_fname, buf);
2332 }
2333 if (rp == NULL)
2334 return;
2336 if (Arguments::created_by_gamma_launcher()) {
2337 // Support for the gamma launcher. Typical value for buf is
2338 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2339 // the right place in the string, then assume we are installed in a JDK and
2340 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2341 // up the path so it looks like libjvm.so is installed there (append a
2342 // fake suffix hotspot/libjvm.so).
2343 const char *p = buf + strlen(buf) - 1;
2344 for (int count = 0; p > buf && count < 5; ++count) {
2345 for (--p; p > buf && *p != '/'; --p)
2346 /* empty */ ;
2347 }
2349 if (strncmp(p, "/jre/lib/", 9) != 0) {
2350 // Look for JAVA_HOME in the environment.
2351 char* java_home_var = ::getenv("JAVA_HOME");
2352 if (java_home_var != NULL && java_home_var[0] != 0) {
2353 char* jrelib_p;
2354 int len;
2356 // Check the current module name "libjvm.so".
2357 p = strrchr(buf, '/');
2358 assert(strstr(p, "/libjvm") == p, "invalid library name");
2360 rp = realpath(java_home_var, buf);
2361 if (rp == NULL)
2362 return;
2364 // determine if this is a legacy image or modules image
2365 // modules image doesn't have "jre" subdirectory
2366 len = strlen(buf);
2367 jrelib_p = buf + len;
2368 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2369 if (0 != access(buf, F_OK)) {
2370 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2371 }
2373 if (0 == access(buf, F_OK)) {
2374 // Use current module name "libjvm.so"
2375 len = strlen(buf);
2376 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2377 } else {
2378 // Go back to path of .so
2379 rp = realpath(dli_fname, buf);
2380 if (rp == NULL)
2381 return;
2382 }
2383 }
2384 }
2385 }
2387 strcpy(saved_jvm_path, buf);
2388 }
2390 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2391 // no prefix required, not even "_"
2392 }
2394 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2395 // no suffix required
2396 }
2398 ////////////////////////////////////////////////////////////////////////////////
2399 // sun.misc.Signal support
2401 static volatile jint sigint_count = 0;
2403 static void
2404 UserHandler(int sig, void *siginfo, void *context) {
2405 // 4511530 - sem_post is serialized and handled by the manager thread. When
2406 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2407 // don't want to flood the manager thread with sem_post requests.
2408 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2409 return;
2411 // Ctrl-C is pressed during error reporting, likely because the error
2412 // handler fails to abort. Let VM die immediately.
2413 if (sig == SIGINT && is_error_reported()) {
2414 os::die();
2415 }
2417 os::signal_notify(sig);
2418 }
2420 void* os::user_handler() {
2421 return CAST_FROM_FN_PTR(void*, UserHandler);
2422 }
2424 class Semaphore : public StackObj {
2425 public:
2426 Semaphore();
2427 ~Semaphore();
2428 void signal();
2429 void wait();
2430 bool trywait();
2431 bool timedwait(unsigned int sec, int nsec);
2432 private:
2433 sem_t _semaphore;
2434 };
2437 Semaphore::Semaphore() {
2438 sem_init(&_semaphore, 0, 0);
2439 }
2441 Semaphore::~Semaphore() {
2442 sem_destroy(&_semaphore);
2443 }
2445 void Semaphore::signal() {
2446 sem_post(&_semaphore);
2447 }
2449 void Semaphore::wait() {
2450 sem_wait(&_semaphore);
2451 }
2453 bool Semaphore::trywait() {
2454 return sem_trywait(&_semaphore) == 0;
2455 }
2457 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2458 struct timespec ts;
2459 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2461 while (1) {
2462 int result = sem_timedwait(&_semaphore, &ts);
2463 if (result == 0) {
2464 return true;
2465 } else if (errno == EINTR) {
2466 continue;
2467 } else if (errno == ETIMEDOUT) {
2468 return false;
2469 } else {
2470 return false;
2471 }
2472 }
2473 }
2475 extern "C" {
2476 typedef void (*sa_handler_t)(int);
2477 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2478 }
2480 void* os::signal(int signal_number, void* handler) {
2481 struct sigaction sigAct, oldSigAct;
2483 sigfillset(&(sigAct.sa_mask));
2484 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2485 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2487 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2488 // -1 means registration failed
2489 return (void *)-1;
2490 }
2492 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2493 }
2495 void os::signal_raise(int signal_number) {
2496 ::raise(signal_number);
2497 }
2499 /*
2500 * The following code is moved from os.cpp for making this
2501 * code platform specific, which it is by its very nature.
2502 */
2504 // Will be modified when max signal is changed to be dynamic
2505 int os::sigexitnum_pd() {
2506 return NSIG;
2507 }
2509 // a counter for each possible signal value
2510 static volatile jint pending_signals[NSIG+1] = { 0 };
2512 // Linux(POSIX) specific hand shaking semaphore.
2513 static sem_t sig_sem;
2514 static Semaphore sr_semaphore;
2516 void os::signal_init_pd() {
2517 // Initialize signal structures
2518 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2520 // Initialize signal semaphore
2521 ::sem_init(&sig_sem, 0, 0);
2522 }
2524 void os::signal_notify(int sig) {
2525 Atomic::inc(&pending_signals[sig]);
2526 ::sem_post(&sig_sem);
2527 }
2529 static int check_pending_signals(bool wait) {
2530 Atomic::store(0, &sigint_count);
2531 for (;;) {
2532 for (int i = 0; i < NSIG + 1; i++) {
2533 jint n = pending_signals[i];
2534 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2535 return i;
2536 }
2537 }
2538 if (!wait) {
2539 return -1;
2540 }
2541 JavaThread *thread = JavaThread::current();
2542 ThreadBlockInVM tbivm(thread);
2544 bool threadIsSuspended;
2545 do {
2546 thread->set_suspend_equivalent();
2547 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2548 ::sem_wait(&sig_sem);
2550 // were we externally suspended while we were waiting?
2551 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2552 if (threadIsSuspended) {
2553 //
2554 // The semaphore has been incremented, but while we were waiting
2555 // another thread suspended us. We don't want to continue running
2556 // while suspended because that would surprise the thread that
2557 // suspended us.
2558 //
2559 ::sem_post(&sig_sem);
2561 thread->java_suspend_self();
2562 }
2563 } while (threadIsSuspended);
2564 }
2565 }
2567 int os::signal_lookup() {
2568 return check_pending_signals(false);
2569 }
2571 int os::signal_wait() {
2572 return check_pending_signals(true);
2573 }
2575 ////////////////////////////////////////////////////////////////////////////////
2576 // Virtual Memory
2578 int os::vm_page_size() {
2579 // Seems redundant as all get out
2580 assert(os::Linux::page_size() != -1, "must call os::init");
2581 return os::Linux::page_size();
2582 }
2584 // Solaris allocates memory by pages.
2585 int os::vm_allocation_granularity() {
2586 assert(os::Linux::page_size() != -1, "must call os::init");
2587 return os::Linux::page_size();
2588 }
2590 // Rationale behind this function:
2591 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2592 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2593 // samples for JITted code. Here we create private executable mapping over the code cache
2594 // and then we can use standard (well, almost, as mapping can change) way to provide
2595 // info for the reporting script by storing timestamp and location of symbol
2596 void linux_wrap_code(char* base, size_t size) {
2597 static volatile jint cnt = 0;
2599 if (!UseOprofile) {
2600 return;
2601 }
2603 char buf[PATH_MAX+1];
2604 int num = Atomic::add(1, &cnt);
2606 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2607 os::get_temp_directory(), os::current_process_id(), num);
2608 unlink(buf);
2610 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2612 if (fd != -1) {
2613 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2614 if (rv != (off_t)-1) {
2615 if (::write(fd, "", 1) == 1) {
2616 mmap(base, size,
2617 PROT_READ|PROT_WRITE|PROT_EXEC,
2618 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2619 }
2620 }
2621 ::close(fd);
2622 unlink(buf);
2623 }
2624 }
2626 static bool recoverable_mmap_error(int err) {
2627 // See if the error is one we can let the caller handle. This
2628 // list of errno values comes from JBS-6843484. I can't find a
2629 // Linux man page that documents this specific set of errno
2630 // values so while this list currently matches Solaris, it may
2631 // change as we gain experience with this failure mode.
2632 switch (err) {
2633 case EBADF:
2634 case EINVAL:
2635 case ENOTSUP:
2636 // let the caller deal with these errors
2637 return true;
2639 default:
2640 // Any remaining errors on this OS can cause our reserved mapping
2641 // to be lost. That can cause confusion where different data
2642 // structures think they have the same memory mapped. The worst
2643 // scenario is if both the VM and a library think they have the
2644 // same memory mapped.
2645 return false;
2646 }
2647 }
2649 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2650 int err) {
2651 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2652 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2653 strerror(err), err);
2654 }
2656 static void warn_fail_commit_memory(char* addr, size_t size,
2657 size_t alignment_hint, bool exec,
2658 int err) {
2659 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2660 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2661 alignment_hint, exec, strerror(err), err);
2662 }
2664 // NOTE: Linux kernel does not really reserve the pages for us.
2665 // All it does is to check if there are enough free pages
2666 // left at the time of mmap(). This could be a potential
2667 // problem.
2668 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2669 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2670 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2671 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2672 if (res != (uintptr_t) MAP_FAILED) {
2673 if (UseNUMAInterleaving) {
2674 numa_make_global(addr, size);
2675 }
2676 return 0;
2677 }
2679 int err = errno; // save errno from mmap() call above
2681 if (!recoverable_mmap_error(err)) {
2682 warn_fail_commit_memory(addr, size, exec, err);
2683 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2684 }
2686 return err;
2687 }
2689 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2690 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2691 }
2693 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2694 const char* mesg) {
2695 assert(mesg != NULL, "mesg must be specified");
2696 int err = os::Linux::commit_memory_impl(addr, size, exec);
2697 if (err != 0) {
2698 // the caller wants all commit errors to exit with the specified mesg:
2699 warn_fail_commit_memory(addr, size, exec, err);
2700 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2701 }
2702 }
2704 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2705 #ifndef MAP_HUGETLB
2706 #define MAP_HUGETLB 0x40000
2707 #endif
2709 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2710 #ifndef MADV_HUGEPAGE
2711 #define MADV_HUGEPAGE 14
2712 #endif
2714 int os::Linux::commit_memory_impl(char* addr, size_t size,
2715 size_t alignment_hint, bool exec) {
2716 int err = os::Linux::commit_memory_impl(addr, size, exec);
2717 if (err == 0) {
2718 realign_memory(addr, size, alignment_hint);
2719 }
2720 return err;
2721 }
2723 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2724 bool exec) {
2725 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2726 }
2728 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2729 size_t alignment_hint, bool exec,
2730 const char* mesg) {
2731 assert(mesg != NULL, "mesg must be specified");
2732 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2733 if (err != 0) {
2734 // the caller wants all commit errors to exit with the specified mesg:
2735 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2736 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2737 }
2738 }
2740 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2741 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2742 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2743 // be supported or the memory may already be backed by huge pages.
2744 ::madvise(addr, bytes, MADV_HUGEPAGE);
2745 }
2746 }
2748 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2749 // This method works by doing an mmap over an existing mmaping and effectively discarding
2750 // the existing pages. However it won't work for SHM-based large pages that cannot be
2751 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2752 // small pages on top of the SHM segment. This method always works for small pages, so we
2753 // allow that in any case.
2754 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2755 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2756 }
2757 }
2759 void os::numa_make_global(char *addr, size_t bytes) {
2760 Linux::numa_interleave_memory(addr, bytes);
2761 }
2763 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2764 // bind policy to MPOL_PREFERRED for the current thread.
2765 #define USE_MPOL_PREFERRED 0
2767 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2768 // To make NUMA and large pages more robust when both enabled, we need to ease
2769 // the requirements on where the memory should be allocated. MPOL_BIND is the
2770 // default policy and it will force memory to be allocated on the specified
2771 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2772 // the specified node, but will not force it. Using this policy will prevent
2773 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2774 // free large pages.
2775 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2776 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2777 }
2779 bool os::numa_topology_changed() { return false; }
2781 size_t os::numa_get_groups_num() {
2782 int max_node = Linux::numa_max_node();
2783 return max_node > 0 ? max_node + 1 : 1;
2784 }
2786 int os::numa_get_group_id() {
2787 int cpu_id = Linux::sched_getcpu();
2788 if (cpu_id != -1) {
2789 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2790 if (lgrp_id != -1) {
2791 return lgrp_id;
2792 }
2793 }
2794 return 0;
2795 }
2797 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2798 for (size_t i = 0; i < size; i++) {
2799 ids[i] = i;
2800 }
2801 return size;
2802 }
2804 bool os::get_page_info(char *start, page_info* info) {
2805 return false;
2806 }
2808 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2809 return end;
2810 }
2813 int os::Linux::sched_getcpu_syscall(void) {
2814 unsigned int cpu;
2815 int retval = -1;
2817 #if defined(IA32)
2818 # ifndef SYS_getcpu
2819 # define SYS_getcpu 318
2820 # endif
2821 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2822 #elif defined(AMD64)
2823 // Unfortunately we have to bring all these macros here from vsyscall.h
2824 // to be able to compile on old linuxes.
2825 # define __NR_vgetcpu 2
2826 # define VSYSCALL_START (-10UL << 20)
2827 # define VSYSCALL_SIZE 1024
2828 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2829 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2830 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2831 retval = vgetcpu(&cpu, NULL, NULL);
2832 #endif
2834 return (retval == -1) ? retval : cpu;
2835 }
2837 // Something to do with the numa-aware allocator needs these symbols
2838 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2839 extern "C" JNIEXPORT void numa_error(char *where) { }
2840 extern "C" JNIEXPORT int fork1() { return fork(); }
2843 // If we are running with libnuma version > 2, then we should
2844 // be trying to use symbols with versions 1.1
2845 // If we are running with earlier version, which did not have symbol versions,
2846 // we should use the base version.
2847 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2848 void *f = dlvsym(handle, name, "libnuma_1.1");
2849 if (f == NULL) {
2850 f = dlsym(handle, name);
2851 }
2852 return f;
2853 }
2855 bool os::Linux::libnuma_init() {
2856 // sched_getcpu() should be in libc.
2857 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2858 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2860 // If it's not, try a direct syscall.
2861 if (sched_getcpu() == -1)
2862 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2864 if (sched_getcpu() != -1) { // Does it work?
2865 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2866 if (handle != NULL) {
2867 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2868 libnuma_dlsym(handle, "numa_node_to_cpus")));
2869 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2870 libnuma_dlsym(handle, "numa_max_node")));
2871 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2872 libnuma_dlsym(handle, "numa_available")));
2873 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2874 libnuma_dlsym(handle, "numa_tonode_memory")));
2875 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2876 libnuma_dlsym(handle, "numa_interleave_memory")));
2877 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2878 libnuma_dlsym(handle, "numa_set_bind_policy")));
2881 if (numa_available() != -1) {
2882 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2883 // Create a cpu -> node mapping
2884 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2885 rebuild_cpu_to_node_map();
2886 return true;
2887 }
2888 }
2889 }
2890 return false;
2891 }
2893 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2894 // The table is later used in get_node_by_cpu().
2895 void os::Linux::rebuild_cpu_to_node_map() {
2896 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2897 // in libnuma (possible values are starting from 16,
2898 // and continuing up with every other power of 2, but less
2899 // than the maximum number of CPUs supported by kernel), and
2900 // is a subject to change (in libnuma version 2 the requirements
2901 // are more reasonable) we'll just hardcode the number they use
2902 // in the library.
2903 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2905 size_t cpu_num = os::active_processor_count();
2906 size_t cpu_map_size = NCPUS / BitsPerCLong;
2907 size_t cpu_map_valid_size =
2908 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2910 cpu_to_node()->clear();
2911 cpu_to_node()->at_grow(cpu_num - 1);
2912 size_t node_num = numa_get_groups_num();
2914 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2915 for (size_t i = 0; i < node_num; i++) {
2916 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2917 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2918 if (cpu_map[j] != 0) {
2919 for (size_t k = 0; k < BitsPerCLong; k++) {
2920 if (cpu_map[j] & (1UL << k)) {
2921 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2922 }
2923 }
2924 }
2925 }
2926 }
2927 }
2928 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2929 }
2931 int os::Linux::get_node_by_cpu(int cpu_id) {
2932 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2933 return cpu_to_node()->at(cpu_id);
2934 }
2935 return -1;
2936 }
2938 GrowableArray<int>* os::Linux::_cpu_to_node;
2939 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2940 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2941 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2942 os::Linux::numa_available_func_t os::Linux::_numa_available;
2943 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2944 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2945 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2946 unsigned long* os::Linux::_numa_all_nodes;
2948 bool os::pd_uncommit_memory(char* addr, size_t size) {
2949 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2950 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2951 return res != (uintptr_t) MAP_FAILED;
2952 }
2954 static
2955 address get_stack_commited_bottom(address bottom, size_t size) {
2956 address nbot = bottom;
2957 address ntop = bottom + size;
2959 size_t page_sz = os::vm_page_size();
2960 unsigned pages = size / page_sz;
2962 unsigned char vec[1];
2963 unsigned imin = 1, imax = pages + 1, imid;
2964 int mincore_return_value;
2966 while (imin < imax) {
2967 imid = (imax + imin) / 2;
2968 nbot = ntop - (imid * page_sz);
2970 // Use a trick with mincore to check whether the page is mapped or not.
2971 // mincore sets vec to 1 if page resides in memory and to 0 if page
2972 // is swapped output but if page we are asking for is unmapped
2973 // it returns -1,ENOMEM
2974 mincore_return_value = mincore(nbot, page_sz, vec);
2976 if (mincore_return_value == -1) {
2977 // Page is not mapped go up
2978 // to find first mapped page
2979 if (errno != EAGAIN) {
2980 assert(errno == ENOMEM, "Unexpected mincore errno");
2981 imax = imid;
2982 }
2983 } else {
2984 // Page is mapped go down
2985 // to find first not mapped page
2986 imin = imid + 1;
2987 }
2988 }
2990 nbot = nbot + page_sz;
2992 // Adjust stack bottom one page up if last checked page is not mapped
2993 if (mincore_return_value == -1) {
2994 nbot = nbot + page_sz;
2995 }
2997 return nbot;
2998 }
3001 // Linux uses a growable mapping for the stack, and if the mapping for
3002 // the stack guard pages is not removed when we detach a thread the
3003 // stack cannot grow beyond the pages where the stack guard was
3004 // mapped. If at some point later in the process the stack expands to
3005 // that point, the Linux kernel cannot expand the stack any further
3006 // because the guard pages are in the way, and a segfault occurs.
3007 //
3008 // However, it's essential not to split the stack region by unmapping
3009 // a region (leaving a hole) that's already part of the stack mapping,
3010 // so if the stack mapping has already grown beyond the guard pages at
3011 // the time we create them, we have to truncate the stack mapping.
3012 // So, we need to know the extent of the stack mapping when
3013 // create_stack_guard_pages() is called.
3015 // We only need this for stacks that are growable: at the time of
3016 // writing thread stacks don't use growable mappings (i.e. those
3017 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3018 // only applies to the main thread.
3020 // If the (growable) stack mapping already extends beyond the point
3021 // where we're going to put our guard pages, truncate the mapping at
3022 // that point by munmap()ping it. This ensures that when we later
3023 // munmap() the guard pages we don't leave a hole in the stack
3024 // mapping. This only affects the main/initial thread
3026 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3028 if (os::Linux::is_initial_thread()) {
3029 // As we manually grow stack up to bottom inside create_attached_thread(),
3030 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3031 // we don't need to do anything special.
3032 // Check it first, before calling heavy function.
3033 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3034 unsigned char vec[1];
3036 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3037 // Fallback to slow path on all errors, including EAGAIN
3038 stack_extent = (uintptr_t) get_stack_commited_bottom(
3039 os::Linux::initial_thread_stack_bottom(),
3040 (size_t)addr - stack_extent);
3041 }
3043 if (stack_extent < (uintptr_t)addr) {
3044 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3045 }
3046 }
3048 return os::commit_memory(addr, size, !ExecMem);
3049 }
3051 // If this is a growable mapping, remove the guard pages entirely by
3052 // munmap()ping them. If not, just call uncommit_memory(). This only
3053 // affects the main/initial thread, but guard against future OS changes
3054 // It's safe to always unmap guard pages for initial thread because we
3055 // always place it right after end of the mapped region
3057 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3058 uintptr_t stack_extent, stack_base;
3060 if (os::Linux::is_initial_thread()) {
3061 return ::munmap(addr, size) == 0;
3062 }
3064 return os::uncommit_memory(addr, size);
3065 }
3067 static address _highest_vm_reserved_address = NULL;
3069 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3070 // at 'requested_addr'. If there are existing memory mappings at the same
3071 // location, however, they will be overwritten. If 'fixed' is false,
3072 // 'requested_addr' is only treated as a hint, the return value may or
3073 // may not start from the requested address. Unlike Linux mmap(), this
3074 // function returns NULL to indicate failure.
3075 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3076 char * addr;
3077 int flags;
3079 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3080 if (fixed) {
3081 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3082 flags |= MAP_FIXED;
3083 }
3085 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3086 // touch an uncommitted page. Otherwise, the read/write might
3087 // succeed if we have enough swap space to back the physical page.
3088 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3089 flags, -1, 0);
3091 if (addr != MAP_FAILED) {
3092 // anon_mmap() should only get called during VM initialization,
3093 // don't need lock (actually we can skip locking even it can be called
3094 // from multiple threads, because _highest_vm_reserved_address is just a
3095 // hint about the upper limit of non-stack memory regions.)
3096 if ((address)addr + bytes > _highest_vm_reserved_address) {
3097 _highest_vm_reserved_address = (address)addr + bytes;
3098 }
3099 }
3101 return addr == MAP_FAILED ? NULL : addr;
3102 }
3104 // Don't update _highest_vm_reserved_address, because there might be memory
3105 // regions above addr + size. If so, releasing a memory region only creates
3106 // a hole in the address space, it doesn't help prevent heap-stack collision.
3107 //
3108 static int anon_munmap(char * addr, size_t size) {
3109 return ::munmap(addr, size) == 0;
3110 }
3112 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3113 size_t alignment_hint) {
3114 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3115 }
3117 bool os::pd_release_memory(char* addr, size_t size) {
3118 return anon_munmap(addr, size);
3119 }
3121 static address highest_vm_reserved_address() {
3122 return _highest_vm_reserved_address;
3123 }
3125 static bool linux_mprotect(char* addr, size_t size, int prot) {
3126 // Linux wants the mprotect address argument to be page aligned.
3127 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3129 // According to SUSv3, mprotect() should only be used with mappings
3130 // established by mmap(), and mmap() always maps whole pages. Unaligned
3131 // 'addr' likely indicates problem in the VM (e.g. trying to change
3132 // protection of malloc'ed or statically allocated memory). Check the
3133 // caller if you hit this assert.
3134 assert(addr == bottom, "sanity check");
3136 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3137 return ::mprotect(bottom, size, prot) == 0;
3138 }
3140 // Set protections specified
3141 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3142 bool is_committed) {
3143 unsigned int p = 0;
3144 switch (prot) {
3145 case MEM_PROT_NONE: p = PROT_NONE; break;
3146 case MEM_PROT_READ: p = PROT_READ; break;
3147 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3148 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3149 default:
3150 ShouldNotReachHere();
3151 }
3152 // is_committed is unused.
3153 return linux_mprotect(addr, bytes, p);
3154 }
3156 bool os::guard_memory(char* addr, size_t size) {
3157 return linux_mprotect(addr, size, PROT_NONE);
3158 }
3160 bool os::unguard_memory(char* addr, size_t size) {
3161 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3162 }
3164 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3165 bool result = false;
3166 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3167 MAP_ANONYMOUS|MAP_PRIVATE,
3168 -1, 0);
3169 if (p != MAP_FAILED) {
3170 void *aligned_p = align_ptr_up(p, page_size);
3172 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3174 munmap(p, page_size * 2);
3175 }
3177 if (warn && !result) {
3178 warning("TransparentHugePages is not supported by the operating system.");
3179 }
3181 return result;
3182 }
3184 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3185 bool result = false;
3186 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3187 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3188 -1, 0);
3190 if (p != MAP_FAILED) {
3191 // We don't know if this really is a huge page or not.
3192 FILE *fp = fopen("/proc/self/maps", "r");
3193 if (fp) {
3194 while (!feof(fp)) {
3195 char chars[257];
3196 long x = 0;
3197 if (fgets(chars, sizeof(chars), fp)) {
3198 if (sscanf(chars, "%lx-%*x", &x) == 1
3199 && x == (long)p) {
3200 if (strstr (chars, "hugepage")) {
3201 result = true;
3202 break;
3203 }
3204 }
3205 }
3206 }
3207 fclose(fp);
3208 }
3209 munmap(p, page_size);
3210 }
3212 if (warn && !result) {
3213 warning("HugeTLBFS is not supported by the operating system.");
3214 }
3216 return result;
3217 }
3219 /*
3220 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3221 *
3222 * From the coredump_filter documentation:
3223 *
3224 * - (bit 0) anonymous private memory
3225 * - (bit 1) anonymous shared memory
3226 * - (bit 2) file-backed private memory
3227 * - (bit 3) file-backed shared memory
3228 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3229 * effective only if the bit 2 is cleared)
3230 * - (bit 5) hugetlb private memory
3231 * - (bit 6) hugetlb shared memory
3232 */
3233 static void set_coredump_filter(void) {
3234 FILE *f;
3235 long cdm;
3237 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3238 return;
3239 }
3241 if (fscanf(f, "%lx", &cdm) != 1) {
3242 fclose(f);
3243 return;
3244 }
3246 rewind(f);
3248 if ((cdm & LARGEPAGES_BIT) == 0) {
3249 cdm |= LARGEPAGES_BIT;
3250 fprintf(f, "%#lx", cdm);
3251 }
3253 fclose(f);
3254 }
3256 // Large page support
3258 static size_t _large_page_size = 0;
3260 size_t os::Linux::find_large_page_size() {
3261 size_t large_page_size = 0;
3263 // large_page_size on Linux is used to round up heap size. x86 uses either
3264 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3265 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3266 // page as large as 256M.
3267 //
3268 // Here we try to figure out page size by parsing /proc/meminfo and looking
3269 // for a line with the following format:
3270 // Hugepagesize: 2048 kB
3271 //
3272 // If we can't determine the value (e.g. /proc is not mounted, or the text
3273 // format has been changed), we'll use the largest page size supported by
3274 // the processor.
3276 #ifndef ZERO
3277 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3278 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3279 #endif // ZERO
3281 FILE *fp = fopen("/proc/meminfo", "r");
3282 if (fp) {
3283 while (!feof(fp)) {
3284 int x = 0;
3285 char buf[16];
3286 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3287 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3288 large_page_size = x * K;
3289 break;
3290 }
3291 } else {
3292 // skip to next line
3293 for (;;) {
3294 int ch = fgetc(fp);
3295 if (ch == EOF || ch == (int)'\n') break;
3296 }
3297 }
3298 }
3299 fclose(fp);
3300 }
3302 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3303 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3304 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3305 proper_unit_for_byte_size(large_page_size));
3306 }
3308 return large_page_size;
3309 }
3311 size_t os::Linux::setup_large_page_size() {
3312 _large_page_size = Linux::find_large_page_size();
3313 const size_t default_page_size = (size_t)Linux::page_size();
3314 if (_large_page_size > default_page_size) {
3315 _page_sizes[0] = _large_page_size;
3316 _page_sizes[1] = default_page_size;
3317 _page_sizes[2] = 0;
3318 }
3320 return _large_page_size;
3321 }
3323 bool os::Linux::setup_large_page_type(size_t page_size) {
3324 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3325 FLAG_IS_DEFAULT(UseSHM) &&
3326 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3328 // The type of large pages has not been specified by the user.
3330 // Try UseHugeTLBFS and then UseSHM.
3331 UseHugeTLBFS = UseSHM = true;
3333 // Don't try UseTransparentHugePages since there are known
3334 // performance issues with it turned on. This might change in the future.
3335 UseTransparentHugePages = false;
3336 }
3338 if (UseTransparentHugePages) {
3339 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3340 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3341 UseHugeTLBFS = false;
3342 UseSHM = false;
3343 return true;
3344 }
3345 UseTransparentHugePages = false;
3346 }
3348 if (UseHugeTLBFS) {
3349 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3350 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3351 UseSHM = false;
3352 return true;
3353 }
3354 UseHugeTLBFS = false;
3355 }
3357 return UseSHM;
3358 }
3360 void os::large_page_init() {
3361 if (!UseLargePages &&
3362 !UseTransparentHugePages &&
3363 !UseHugeTLBFS &&
3364 !UseSHM) {
3365 // Not using large pages.
3366 return;
3367 }
3369 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3370 // The user explicitly turned off large pages.
3371 // Ignore the rest of the large pages flags.
3372 UseTransparentHugePages = false;
3373 UseHugeTLBFS = false;
3374 UseSHM = false;
3375 return;
3376 }
3378 size_t large_page_size = Linux::setup_large_page_size();
3379 UseLargePages = Linux::setup_large_page_type(large_page_size);
3381 set_coredump_filter();
3382 }
3384 #ifndef SHM_HUGETLB
3385 #define SHM_HUGETLB 04000
3386 #endif
3388 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3389 // "exec" is passed in but not used. Creating the shared image for
3390 // the code cache doesn't have an SHM_X executable permission to check.
3391 assert(UseLargePages && UseSHM, "only for SHM large pages");
3392 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3394 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3395 return NULL; // Fallback to small pages.
3396 }
3398 key_t key = IPC_PRIVATE;
3399 char *addr;
3401 bool warn_on_failure = UseLargePages &&
3402 (!FLAG_IS_DEFAULT(UseLargePages) ||
3403 !FLAG_IS_DEFAULT(UseSHM) ||
3404 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3405 );
3406 char msg[128];
3408 // Create a large shared memory region to attach to based on size.
3409 // Currently, size is the total size of the heap
3410 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3411 if (shmid == -1) {
3412 // Possible reasons for shmget failure:
3413 // 1. shmmax is too small for Java heap.
3414 // > check shmmax value: cat /proc/sys/kernel/shmmax
3415 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3416 // 2. not enough large page memory.
3417 // > check available large pages: cat /proc/meminfo
3418 // > increase amount of large pages:
3419 // echo new_value > /proc/sys/vm/nr_hugepages
3420 // Note 1: different Linux may use different name for this property,
3421 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3422 // Note 2: it's possible there's enough physical memory available but
3423 // they are so fragmented after a long run that they can't
3424 // coalesce into large pages. Try to reserve large pages when
3425 // the system is still "fresh".
3426 if (warn_on_failure) {
3427 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3428 warning(msg);
3429 }
3430 return NULL;
3431 }
3433 // attach to the region
3434 addr = (char*)shmat(shmid, req_addr, 0);
3435 int err = errno;
3437 // Remove shmid. If shmat() is successful, the actual shared memory segment
3438 // will be deleted when it's detached by shmdt() or when the process
3439 // terminates. If shmat() is not successful this will remove the shared
3440 // segment immediately.
3441 shmctl(shmid, IPC_RMID, NULL);
3443 if ((intptr_t)addr == -1) {
3444 if (warn_on_failure) {
3445 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3446 warning(msg);
3447 }
3448 return NULL;
3449 }
3451 return addr;
3452 }
3454 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3455 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3457 bool warn_on_failure = UseLargePages &&
3458 (!FLAG_IS_DEFAULT(UseLargePages) ||
3459 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3460 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3462 if (warn_on_failure) {
3463 char msg[128];
3464 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3465 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3466 warning(msg);
3467 }
3468 }
3470 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3471 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3472 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3473 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3475 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3476 char* addr = (char*)::mmap(req_addr, bytes, prot,
3477 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3478 -1, 0);
3480 if (addr == MAP_FAILED) {
3481 warn_on_large_pages_failure(req_addr, bytes, errno);
3482 return NULL;
3483 }
3485 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3487 return addr;
3488 }
3490 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3491 size_t large_page_size = os::large_page_size();
3493 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3495 // Allocate small pages.
3497 char* start;
3498 if (req_addr != NULL) {
3499 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3500 assert(is_size_aligned(bytes, alignment), "Must be");
3501 start = os::reserve_memory(bytes, req_addr);
3502 assert(start == NULL || start == req_addr, "Must be");
3503 } else {
3504 start = os::reserve_memory_aligned(bytes, alignment);
3505 }
3507 if (start == NULL) {
3508 return NULL;
3509 }
3511 assert(is_ptr_aligned(start, alignment), "Must be");
3513 // os::reserve_memory_special will record this memory area.
3514 // Need to release it here to prevent overlapping reservations.
3515 MemTracker::record_virtual_memory_release((address)start, bytes);
3517 char* end = start + bytes;
3519 // Find the regions of the allocated chunk that can be promoted to large pages.
3520 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3521 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3523 size_t lp_bytes = lp_end - lp_start;
3525 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3527 if (lp_bytes == 0) {
3528 // The mapped region doesn't even span the start and the end of a large page.
3529 // Fall back to allocate a non-special area.
3530 ::munmap(start, end - start);
3531 return NULL;
3532 }
3534 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3537 void* result;
3539 if (start != lp_start) {
3540 result = ::mmap(start, lp_start - start, prot,
3541 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3542 -1, 0);
3543 if (result == MAP_FAILED) {
3544 ::munmap(lp_start, end - lp_start);
3545 return NULL;
3546 }
3547 }
3549 result = ::mmap(lp_start, lp_bytes, prot,
3550 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3551 -1, 0);
3552 if (result == MAP_FAILED) {
3553 warn_on_large_pages_failure(req_addr, bytes, errno);
3554 // If the mmap above fails, the large pages region will be unmapped and we
3555 // have regions before and after with small pages. Release these regions.
3556 //
3557 // | mapped | unmapped | mapped |
3558 // ^ ^ ^ ^
3559 // start lp_start lp_end end
3560 //
3561 ::munmap(start, lp_start - start);
3562 ::munmap(lp_end, end - lp_end);
3563 return NULL;
3564 }
3566 if (lp_end != end) {
3567 result = ::mmap(lp_end, end - lp_end, prot,
3568 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3569 -1, 0);
3570 if (result == MAP_FAILED) {
3571 ::munmap(start, lp_end - start);
3572 return NULL;
3573 }
3574 }
3576 return start;
3577 }
3579 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3580 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3581 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3582 assert(is_power_of_2(alignment), "Must be");
3583 assert(is_power_of_2(os::large_page_size()), "Must be");
3584 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3586 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3587 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3588 } else {
3589 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3590 }
3591 }
3593 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3594 assert(UseLargePages, "only for large pages");
3596 char* addr;
3597 if (UseSHM) {
3598 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3599 } else {
3600 assert(UseHugeTLBFS, "must be");
3601 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3602 }
3604 if (addr != NULL) {
3605 if (UseNUMAInterleaving) {
3606 numa_make_global(addr, bytes);
3607 }
3609 // The memory is committed
3610 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3611 }
3613 return addr;
3614 }
3616 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3617 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3618 return shmdt(base) == 0;
3619 }
3621 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3622 return pd_release_memory(base, bytes);
3623 }
3625 bool os::release_memory_special(char* base, size_t bytes) {
3626 assert(UseLargePages, "only for large pages");
3628 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3630 bool res;
3631 if (UseSHM) {
3632 res = os::Linux::release_memory_special_shm(base, bytes);
3633 } else {
3634 assert(UseHugeTLBFS, "must be");
3635 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3636 }
3638 if (res) {
3639 tkr.record((address)base, bytes);
3640 } else {
3641 tkr.discard();
3642 }
3644 return res;
3645 }
3647 size_t os::large_page_size() {
3648 return _large_page_size;
3649 }
3651 // With SysV SHM the entire memory region must be allocated as shared
3652 // memory.
3653 // HugeTLBFS allows application to commit large page memory on demand.
3654 // However, when committing memory with HugeTLBFS fails, the region
3655 // that was supposed to be committed will lose the old reservation
3656 // and allow other threads to steal that memory region. Because of this
3657 // behavior we can't commit HugeTLBFS memory.
3658 bool os::can_commit_large_page_memory() {
3659 return UseTransparentHugePages;
3660 }
3662 bool os::can_execute_large_page_memory() {
3663 return UseTransparentHugePages || UseHugeTLBFS;
3664 }
3666 // Reserve memory at an arbitrary address, only if that area is
3667 // available (and not reserved for something else).
3669 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3670 const int max_tries = 10;
3671 char* base[max_tries];
3672 size_t size[max_tries];
3673 const size_t gap = 0x000000;
3675 // Assert only that the size is a multiple of the page size, since
3676 // that's all that mmap requires, and since that's all we really know
3677 // about at this low abstraction level. If we need higher alignment,
3678 // we can either pass an alignment to this method or verify alignment
3679 // in one of the methods further up the call chain. See bug 5044738.
3680 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3682 // Repeatedly allocate blocks until the block is allocated at the
3683 // right spot. Give up after max_tries. Note that reserve_memory() will
3684 // automatically update _highest_vm_reserved_address if the call is
3685 // successful. The variable tracks the highest memory address every reserved
3686 // by JVM. It is used to detect heap-stack collision if running with
3687 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3688 // space than needed, it could confuse the collision detecting code. To
3689 // solve the problem, save current _highest_vm_reserved_address and
3690 // calculate the correct value before return.
3691 address old_highest = _highest_vm_reserved_address;
3693 // Linux mmap allows caller to pass an address as hint; give it a try first,
3694 // if kernel honors the hint then we can return immediately.
3695 char * addr = anon_mmap(requested_addr, bytes, false);
3696 if (addr == requested_addr) {
3697 return requested_addr;
3698 }
3700 if (addr != NULL) {
3701 // mmap() is successful but it fails to reserve at the requested address
3702 anon_munmap(addr, bytes);
3703 }
3705 int i;
3706 for (i = 0; i < max_tries; ++i) {
3707 base[i] = reserve_memory(bytes);
3709 if (base[i] != NULL) {
3710 // Is this the block we wanted?
3711 if (base[i] == requested_addr) {
3712 size[i] = bytes;
3713 break;
3714 }
3716 // Does this overlap the block we wanted? Give back the overlapped
3717 // parts and try again.
3719 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3720 if (top_overlap >= 0 && top_overlap < bytes) {
3721 unmap_memory(base[i], top_overlap);
3722 base[i] += top_overlap;
3723 size[i] = bytes - top_overlap;
3724 } else {
3725 size_t bottom_overlap = base[i] + bytes - requested_addr;
3726 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3727 unmap_memory(requested_addr, bottom_overlap);
3728 size[i] = bytes - bottom_overlap;
3729 } else {
3730 size[i] = bytes;
3731 }
3732 }
3733 }
3734 }
3736 // Give back the unused reserved pieces.
3738 for (int j = 0; j < i; ++j) {
3739 if (base[j] != NULL) {
3740 unmap_memory(base[j], size[j]);
3741 }
3742 }
3744 if (i < max_tries) {
3745 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3746 return requested_addr;
3747 } else {
3748 _highest_vm_reserved_address = old_highest;
3749 return NULL;
3750 }
3751 }
3753 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3754 return ::read(fd, buf, nBytes);
3755 }
3757 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3758 // Solaris uses poll(), linux uses park().
3759 // Poll() is likely a better choice, assuming that Thread.interrupt()
3760 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3761 // SIGSEGV, see 4355769.
3763 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3764 assert(thread == Thread::current(), "thread consistency check");
3766 ParkEvent * const slp = thread->_SleepEvent ;
3767 slp->reset() ;
3768 OrderAccess::fence() ;
3770 if (interruptible) {
3771 jlong prevtime = javaTimeNanos();
3773 for (;;) {
3774 if (os::is_interrupted(thread, true)) {
3775 return OS_INTRPT;
3776 }
3778 jlong newtime = javaTimeNanos();
3780 if (newtime - prevtime < 0) {
3781 // time moving backwards, should only happen if no monotonic clock
3782 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3783 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3784 } else {
3785 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3786 }
3788 if(millis <= 0) {
3789 return OS_OK;
3790 }
3792 prevtime = newtime;
3794 {
3795 assert(thread->is_Java_thread(), "sanity check");
3796 JavaThread *jt = (JavaThread *) thread;
3797 ThreadBlockInVM tbivm(jt);
3798 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3800 jt->set_suspend_equivalent();
3801 // cleared by handle_special_suspend_equivalent_condition() or
3802 // java_suspend_self() via check_and_wait_while_suspended()
3804 slp->park(millis);
3806 // were we externally suspended while we were waiting?
3807 jt->check_and_wait_while_suspended();
3808 }
3809 }
3810 } else {
3811 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3812 jlong prevtime = javaTimeNanos();
3814 for (;;) {
3815 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3816 // the 1st iteration ...
3817 jlong newtime = javaTimeNanos();
3819 if (newtime - prevtime < 0) {
3820 // time moving backwards, should only happen if no monotonic clock
3821 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3822 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3823 } else {
3824 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3825 }
3827 if(millis <= 0) break ;
3829 prevtime = newtime;
3830 slp->park(millis);
3831 }
3832 return OS_OK ;
3833 }
3834 }
3836 int os::naked_sleep() {
3837 // %% make the sleep time an integer flag. for now use 1 millisec.
3838 return os::sleep(Thread::current(), 1, false);
3839 }
3841 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3842 void os::infinite_sleep() {
3843 while (true) { // sleep forever ...
3844 ::sleep(100); // ... 100 seconds at a time
3845 }
3846 }
3848 // Used to convert frequent JVM_Yield() to nops
3849 bool os::dont_yield() {
3850 return DontYieldALot;
3851 }
3853 void os::yield() {
3854 sched_yield();
3855 }
3857 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3859 void os::yield_all(int attempts) {
3860 // Yields to all threads, including threads with lower priorities
3861 // Threads on Linux are all with same priority. The Solaris style
3862 // os::yield_all() with nanosleep(1ms) is not necessary.
3863 sched_yield();
3864 }
3866 // Called from the tight loops to possibly influence time-sharing heuristics
3867 void os::loop_breaker(int attempts) {
3868 os::yield_all(attempts);
3869 }
3871 ////////////////////////////////////////////////////////////////////////////////
3872 // thread priority support
3874 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3875 // only supports dynamic priority, static priority must be zero. For real-time
3876 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3877 // However, for large multi-threaded applications, SCHED_RR is not only slower
3878 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3879 // of 5 runs - Sep 2005).
3880 //
3881 // The following code actually changes the niceness of kernel-thread/LWP. It
3882 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3883 // not the entire user process, and user level threads are 1:1 mapped to kernel
3884 // threads. It has always been the case, but could change in the future. For
3885 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3886 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3888 int os::java_to_os_priority[CriticalPriority + 1] = {
3889 19, // 0 Entry should never be used
3891 4, // 1 MinPriority
3892 3, // 2
3893 2, // 3
3895 1, // 4
3896 0, // 5 NormPriority
3897 -1, // 6
3899 -2, // 7
3900 -3, // 8
3901 -4, // 9 NearMaxPriority
3903 -5, // 10 MaxPriority
3905 -5 // 11 CriticalPriority
3906 };
3908 static int prio_init() {
3909 if (ThreadPriorityPolicy == 1) {
3910 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3911 // if effective uid is not root. Perhaps, a more elegant way of doing
3912 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3913 if (geteuid() != 0) {
3914 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3915 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3916 }
3917 ThreadPriorityPolicy = 0;
3918 }
3919 }
3920 if (UseCriticalJavaThreadPriority) {
3921 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3922 }
3923 return 0;
3924 }
3926 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3927 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3929 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3930 return (ret == 0) ? OS_OK : OS_ERR;
3931 }
3933 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3934 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3935 *priority_ptr = java_to_os_priority[NormPriority];
3936 return OS_OK;
3937 }
3939 errno = 0;
3940 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3941 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3942 }
3944 // Hint to the underlying OS that a task switch would not be good.
3945 // Void return because it's a hint and can fail.
3946 void os::hint_no_preempt() {}
3948 ////////////////////////////////////////////////////////////////////////////////
3949 // suspend/resume support
3951 // the low-level signal-based suspend/resume support is a remnant from the
3952 // old VM-suspension that used to be for java-suspension, safepoints etc,
3953 // within hotspot. Now there is a single use-case for this:
3954 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3955 // that runs in the watcher thread.
3956 // The remaining code is greatly simplified from the more general suspension
3957 // code that used to be used.
3958 //
3959 // The protocol is quite simple:
3960 // - suspend:
3961 // - sends a signal to the target thread
3962 // - polls the suspend state of the osthread using a yield loop
3963 // - target thread signal handler (SR_handler) sets suspend state
3964 // and blocks in sigsuspend until continued
3965 // - resume:
3966 // - sets target osthread state to continue
3967 // - sends signal to end the sigsuspend loop in the SR_handler
3968 //
3969 // Note that the SR_lock plays no role in this suspend/resume protocol.
3970 //
3972 static void resume_clear_context(OSThread *osthread) {
3973 osthread->set_ucontext(NULL);
3974 osthread->set_siginfo(NULL);
3975 }
3977 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3978 osthread->set_ucontext(context);
3979 osthread->set_siginfo(siginfo);
3980 }
3982 //
3983 // Handler function invoked when a thread's execution is suspended or
3984 // resumed. We have to be careful that only async-safe functions are
3985 // called here (Note: most pthread functions are not async safe and
3986 // should be avoided.)
3987 //
3988 // Note: sigwait() is a more natural fit than sigsuspend() from an
3989 // interface point of view, but sigwait() prevents the signal hander
3990 // from being run. libpthread would get very confused by not having
3991 // its signal handlers run and prevents sigwait()'s use with the
3992 // mutex granting granting signal.
3993 //
3994 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3995 //
3996 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3997 // Save and restore errno to avoid confusing native code with EINTR
3998 // after sigsuspend.
3999 int old_errno = errno;
4001 Thread* thread = Thread::current();
4002 OSThread* osthread = thread->osthread();
4003 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4005 os::SuspendResume::State current = osthread->sr.state();
4006 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4007 suspend_save_context(osthread, siginfo, context);
4009 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4010 os::SuspendResume::State state = osthread->sr.suspended();
4011 if (state == os::SuspendResume::SR_SUSPENDED) {
4012 sigset_t suspend_set; // signals for sigsuspend()
4014 // get current set of blocked signals and unblock resume signal
4015 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4016 sigdelset(&suspend_set, SR_signum);
4018 sr_semaphore.signal();
4019 // wait here until we are resumed
4020 while (1) {
4021 sigsuspend(&suspend_set);
4023 os::SuspendResume::State result = osthread->sr.running();
4024 if (result == os::SuspendResume::SR_RUNNING) {
4025 sr_semaphore.signal();
4026 break;
4027 }
4028 }
4030 } else if (state == os::SuspendResume::SR_RUNNING) {
4031 // request was cancelled, continue
4032 } else {
4033 ShouldNotReachHere();
4034 }
4036 resume_clear_context(osthread);
4037 } else if (current == os::SuspendResume::SR_RUNNING) {
4038 // request was cancelled, continue
4039 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4040 // ignore
4041 } else {
4042 // ignore
4043 }
4045 errno = old_errno;
4046 }
4049 static int SR_initialize() {
4050 struct sigaction act;
4051 char *s;
4052 /* Get signal number to use for suspend/resume */
4053 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4054 int sig = ::strtol(s, 0, 10);
4055 if (sig > 0 || sig < _NSIG) {
4056 SR_signum = sig;
4057 }
4058 }
4060 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4061 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4063 sigemptyset(&SR_sigset);
4064 sigaddset(&SR_sigset, SR_signum);
4066 /* Set up signal handler for suspend/resume */
4067 act.sa_flags = SA_RESTART|SA_SIGINFO;
4068 act.sa_handler = (void (*)(int)) SR_handler;
4070 // SR_signum is blocked by default.
4071 // 4528190 - We also need to block pthread restart signal (32 on all
4072 // supported Linux platforms). Note that LinuxThreads need to block
4073 // this signal for all threads to work properly. So we don't have
4074 // to use hard-coded signal number when setting up the mask.
4075 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4077 if (sigaction(SR_signum, &act, 0) == -1) {
4078 return -1;
4079 }
4081 // Save signal flag
4082 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4083 return 0;
4084 }
4086 static int sr_notify(OSThread* osthread) {
4087 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4088 assert_status(status == 0, status, "pthread_kill");
4089 return status;
4090 }
4092 // "Randomly" selected value for how long we want to spin
4093 // before bailing out on suspending a thread, also how often
4094 // we send a signal to a thread we want to resume
4095 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4096 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4098 // returns true on success and false on error - really an error is fatal
4099 // but this seems the normal response to library errors
4100 static bool do_suspend(OSThread* osthread) {
4101 assert(osthread->sr.is_running(), "thread should be running");
4102 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4104 // mark as suspended and send signal
4105 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4106 // failed to switch, state wasn't running?
4107 ShouldNotReachHere();
4108 return false;
4109 }
4111 if (sr_notify(osthread) != 0) {
4112 ShouldNotReachHere();
4113 }
4115 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4116 while (true) {
4117 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4118 break;
4119 } else {
4120 // timeout
4121 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4122 if (cancelled == os::SuspendResume::SR_RUNNING) {
4123 return false;
4124 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4125 // make sure that we consume the signal on the semaphore as well
4126 sr_semaphore.wait();
4127 break;
4128 } else {
4129 ShouldNotReachHere();
4130 return false;
4131 }
4132 }
4133 }
4135 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4136 return true;
4137 }
4139 static void do_resume(OSThread* osthread) {
4140 assert(osthread->sr.is_suspended(), "thread should be suspended");
4141 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4143 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4144 // failed to switch to WAKEUP_REQUEST
4145 ShouldNotReachHere();
4146 return;
4147 }
4149 while (true) {
4150 if (sr_notify(osthread) == 0) {
4151 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4152 if (osthread->sr.is_running()) {
4153 return;
4154 }
4155 }
4156 } else {
4157 ShouldNotReachHere();
4158 }
4159 }
4161 guarantee(osthread->sr.is_running(), "Must be running!");
4162 }
4164 ////////////////////////////////////////////////////////////////////////////////
4165 // interrupt support
4167 void os::interrupt(Thread* thread) {
4168 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4169 "possibility of dangling Thread pointer");
4171 OSThread* osthread = thread->osthread();
4173 if (!osthread->interrupted()) {
4174 osthread->set_interrupted(true);
4175 // More than one thread can get here with the same value of osthread,
4176 // resulting in multiple notifications. We do, however, want the store
4177 // to interrupted() to be visible to other threads before we execute unpark().
4178 OrderAccess::fence();
4179 ParkEvent * const slp = thread->_SleepEvent ;
4180 if (slp != NULL) slp->unpark() ;
4181 }
4183 // For JSR166. Unpark even if interrupt status already was set
4184 if (thread->is_Java_thread())
4185 ((JavaThread*)thread)->parker()->unpark();
4187 ParkEvent * ev = thread->_ParkEvent ;
4188 if (ev != NULL) ev->unpark() ;
4190 }
4192 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4193 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4194 "possibility of dangling Thread pointer");
4196 OSThread* osthread = thread->osthread();
4198 bool interrupted = osthread->interrupted();
4200 if (interrupted && clear_interrupted) {
4201 osthread->set_interrupted(false);
4202 // consider thread->_SleepEvent->reset() ... optional optimization
4203 }
4205 return interrupted;
4206 }
4208 ///////////////////////////////////////////////////////////////////////////////////
4209 // signal handling (except suspend/resume)
4211 // This routine may be used by user applications as a "hook" to catch signals.
4212 // The user-defined signal handler must pass unrecognized signals to this
4213 // routine, and if it returns true (non-zero), then the signal handler must
4214 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4215 // routine will never retun false (zero), but instead will execute a VM panic
4216 // routine kill the process.
4217 //
4218 // If this routine returns false, it is OK to call it again. This allows
4219 // the user-defined signal handler to perform checks either before or after
4220 // the VM performs its own checks. Naturally, the user code would be making
4221 // a serious error if it tried to handle an exception (such as a null check
4222 // or breakpoint) that the VM was generating for its own correct operation.
4223 //
4224 // This routine may recognize any of the following kinds of signals:
4225 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4226 // It should be consulted by handlers for any of those signals.
4227 //
4228 // The caller of this routine must pass in the three arguments supplied
4229 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4230 // field of the structure passed to sigaction(). This routine assumes that
4231 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4232 //
4233 // Note that the VM will print warnings if it detects conflicting signal
4234 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4235 //
4236 extern "C" JNIEXPORT int
4237 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4238 void* ucontext, int abort_if_unrecognized);
4240 void signalHandler(int sig, siginfo_t* info, void* uc) {
4241 assert(info != NULL && uc != NULL, "it must be old kernel");
4242 int orig_errno = errno; // Preserve errno value over signal handler.
4243 JVM_handle_linux_signal(sig, info, uc, true);
4244 errno = orig_errno;
4245 }
4248 // This boolean allows users to forward their own non-matching signals
4249 // to JVM_handle_linux_signal, harmlessly.
4250 bool os::Linux::signal_handlers_are_installed = false;
4252 // For signal-chaining
4253 struct sigaction os::Linux::sigact[MAXSIGNUM];
4254 unsigned int os::Linux::sigs = 0;
4255 bool os::Linux::libjsig_is_loaded = false;
4256 typedef struct sigaction *(*get_signal_t)(int);
4257 get_signal_t os::Linux::get_signal_action = NULL;
4259 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4260 struct sigaction *actp = NULL;
4262 if (libjsig_is_loaded) {
4263 // Retrieve the old signal handler from libjsig
4264 actp = (*get_signal_action)(sig);
4265 }
4266 if (actp == NULL) {
4267 // Retrieve the preinstalled signal handler from jvm
4268 actp = get_preinstalled_handler(sig);
4269 }
4271 return actp;
4272 }
4274 static bool call_chained_handler(struct sigaction *actp, int sig,
4275 siginfo_t *siginfo, void *context) {
4276 // Call the old signal handler
4277 if (actp->sa_handler == SIG_DFL) {
4278 // It's more reasonable to let jvm treat it as an unexpected exception
4279 // instead of taking the default action.
4280 return false;
4281 } else if (actp->sa_handler != SIG_IGN) {
4282 if ((actp->sa_flags & SA_NODEFER) == 0) {
4283 // automaticlly block the signal
4284 sigaddset(&(actp->sa_mask), sig);
4285 }
4287 sa_handler_t hand;
4288 sa_sigaction_t sa;
4289 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4290 // retrieve the chained handler
4291 if (siginfo_flag_set) {
4292 sa = actp->sa_sigaction;
4293 } else {
4294 hand = actp->sa_handler;
4295 }
4297 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4298 actp->sa_handler = SIG_DFL;
4299 }
4301 // try to honor the signal mask
4302 sigset_t oset;
4303 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4305 // call into the chained handler
4306 if (siginfo_flag_set) {
4307 (*sa)(sig, siginfo, context);
4308 } else {
4309 (*hand)(sig);
4310 }
4312 // restore the signal mask
4313 pthread_sigmask(SIG_SETMASK, &oset, 0);
4314 }
4315 // Tell jvm's signal handler the signal is taken care of.
4316 return true;
4317 }
4319 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4320 bool chained = false;
4321 // signal-chaining
4322 if (UseSignalChaining) {
4323 struct sigaction *actp = get_chained_signal_action(sig);
4324 if (actp != NULL) {
4325 chained = call_chained_handler(actp, sig, siginfo, context);
4326 }
4327 }
4328 return chained;
4329 }
4331 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4332 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4333 return &sigact[sig];
4334 }
4335 return NULL;
4336 }
4338 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4339 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4340 sigact[sig] = oldAct;
4341 sigs |= (unsigned int)1 << sig;
4342 }
4344 // for diagnostic
4345 int os::Linux::sigflags[MAXSIGNUM];
4347 int os::Linux::get_our_sigflags(int sig) {
4348 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4349 return sigflags[sig];
4350 }
4352 void os::Linux::set_our_sigflags(int sig, int flags) {
4353 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4354 sigflags[sig] = flags;
4355 }
4357 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4358 // Check for overwrite.
4359 struct sigaction oldAct;
4360 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4362 void* oldhand = oldAct.sa_sigaction
4363 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4364 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4365 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4366 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4367 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4368 if (AllowUserSignalHandlers || !set_installed) {
4369 // Do not overwrite; user takes responsibility to forward to us.
4370 return;
4371 } else if (UseSignalChaining) {
4372 // save the old handler in jvm
4373 save_preinstalled_handler(sig, oldAct);
4374 // libjsig also interposes the sigaction() call below and saves the
4375 // old sigaction on it own.
4376 } else {
4377 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4378 "%#lx for signal %d.", (long)oldhand, sig));
4379 }
4380 }
4382 struct sigaction sigAct;
4383 sigfillset(&(sigAct.sa_mask));
4384 sigAct.sa_handler = SIG_DFL;
4385 if (!set_installed) {
4386 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4387 } else {
4388 sigAct.sa_sigaction = signalHandler;
4389 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4390 }
4391 // Save flags, which are set by ours
4392 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4393 sigflags[sig] = sigAct.sa_flags;
4395 int ret = sigaction(sig, &sigAct, &oldAct);
4396 assert(ret == 0, "check");
4398 void* oldhand2 = oldAct.sa_sigaction
4399 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4400 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4401 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4402 }
4404 // install signal handlers for signals that HotSpot needs to
4405 // handle in order to support Java-level exception handling.
4407 void os::Linux::install_signal_handlers() {
4408 if (!signal_handlers_are_installed) {
4409 signal_handlers_are_installed = true;
4411 // signal-chaining
4412 typedef void (*signal_setting_t)();
4413 signal_setting_t begin_signal_setting = NULL;
4414 signal_setting_t end_signal_setting = NULL;
4415 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4416 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4417 if (begin_signal_setting != NULL) {
4418 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4419 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4420 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4421 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4422 libjsig_is_loaded = true;
4423 assert(UseSignalChaining, "should enable signal-chaining");
4424 }
4425 if (libjsig_is_loaded) {
4426 // Tell libjsig jvm is setting signal handlers
4427 (*begin_signal_setting)();
4428 }
4430 set_signal_handler(SIGSEGV, true);
4431 set_signal_handler(SIGPIPE, true);
4432 set_signal_handler(SIGBUS, true);
4433 set_signal_handler(SIGILL, true);
4434 set_signal_handler(SIGFPE, true);
4435 #if defined(PPC64)
4436 set_signal_handler(SIGTRAP, true);
4437 #endif
4438 set_signal_handler(SIGXFSZ, true);
4440 if (libjsig_is_loaded) {
4441 // Tell libjsig jvm finishes setting signal handlers
4442 (*end_signal_setting)();
4443 }
4445 // We don't activate signal checker if libjsig is in place, we trust ourselves
4446 // and if UserSignalHandler is installed all bets are off.
4447 // Log that signal checking is off only if -verbose:jni is specified.
4448 if (CheckJNICalls) {
4449 if (libjsig_is_loaded) {
4450 if (PrintJNIResolving) {
4451 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4452 }
4453 check_signals = false;
4454 }
4455 if (AllowUserSignalHandlers) {
4456 if (PrintJNIResolving) {
4457 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4458 }
4459 check_signals = false;
4460 }
4461 }
4462 }
4463 }
4465 // This is the fastest way to get thread cpu time on Linux.
4466 // Returns cpu time (user+sys) for any thread, not only for current.
4467 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4468 // It might work on 2.6.10+ with a special kernel/glibc patch.
4469 // For reference, please, see IEEE Std 1003.1-2004:
4470 // http://www.unix.org/single_unix_specification
4472 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4473 struct timespec tp;
4474 int rc = os::Linux::clock_gettime(clockid, &tp);
4475 assert(rc == 0, "clock_gettime is expected to return 0 code");
4477 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4478 }
4480 /////
4481 // glibc on Linux platform uses non-documented flag
4482 // to indicate, that some special sort of signal
4483 // trampoline is used.
4484 // We will never set this flag, and we should
4485 // ignore this flag in our diagnostic
4486 #ifdef SIGNIFICANT_SIGNAL_MASK
4487 #undef SIGNIFICANT_SIGNAL_MASK
4488 #endif
4489 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4491 static const char* get_signal_handler_name(address handler,
4492 char* buf, int buflen) {
4493 int offset;
4494 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4495 if (found) {
4496 // skip directory names
4497 const char *p1, *p2;
4498 p1 = buf;
4499 size_t len = strlen(os::file_separator());
4500 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4501 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4502 } else {
4503 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4504 }
4505 return buf;
4506 }
4508 static void print_signal_handler(outputStream* st, int sig,
4509 char* buf, size_t buflen) {
4510 struct sigaction sa;
4512 sigaction(sig, NULL, &sa);
4514 // See comment for SIGNIFICANT_SIGNAL_MASK define
4515 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4517 st->print("%s: ", os::exception_name(sig, buf, buflen));
4519 address handler = (sa.sa_flags & SA_SIGINFO)
4520 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4521 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4523 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4524 st->print("SIG_DFL");
4525 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4526 st->print("SIG_IGN");
4527 } else {
4528 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4529 }
4531 st->print(", sa_mask[0]=");
4532 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4534 address rh = VMError::get_resetted_sighandler(sig);
4535 // May be, handler was resetted by VMError?
4536 if(rh != NULL) {
4537 handler = rh;
4538 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4539 }
4541 st->print(", sa_flags=");
4542 os::Posix::print_sa_flags(st, sa.sa_flags);
4544 // Check: is it our handler?
4545 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4546 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4547 // It is our signal handler
4548 // check for flags, reset system-used one!
4549 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4550 st->print(
4551 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4552 os::Linux::get_our_sigflags(sig));
4553 }
4554 }
4555 st->cr();
4556 }
4559 #define DO_SIGNAL_CHECK(sig) \
4560 if (!sigismember(&check_signal_done, sig)) \
4561 os::Linux::check_signal_handler(sig)
4563 // This method is a periodic task to check for misbehaving JNI applications
4564 // under CheckJNI, we can add any periodic checks here
4566 void os::run_periodic_checks() {
4568 if (check_signals == false) return;
4570 // SEGV and BUS if overridden could potentially prevent
4571 // generation of hs*.log in the event of a crash, debugging
4572 // such a case can be very challenging, so we absolutely
4573 // check the following for a good measure:
4574 DO_SIGNAL_CHECK(SIGSEGV);
4575 DO_SIGNAL_CHECK(SIGILL);
4576 DO_SIGNAL_CHECK(SIGFPE);
4577 DO_SIGNAL_CHECK(SIGBUS);
4578 DO_SIGNAL_CHECK(SIGPIPE);
4579 DO_SIGNAL_CHECK(SIGXFSZ);
4580 #if defined(PPC64)
4581 DO_SIGNAL_CHECK(SIGTRAP);
4582 #endif
4584 // ReduceSignalUsage allows the user to override these handlers
4585 // see comments at the very top and jvm_solaris.h
4586 if (!ReduceSignalUsage) {
4587 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4588 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4589 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4590 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4591 }
4593 DO_SIGNAL_CHECK(SR_signum);
4594 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4595 }
4597 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4599 static os_sigaction_t os_sigaction = NULL;
4601 void os::Linux::check_signal_handler(int sig) {
4602 char buf[O_BUFLEN];
4603 address jvmHandler = NULL;
4606 struct sigaction act;
4607 if (os_sigaction == NULL) {
4608 // only trust the default sigaction, in case it has been interposed
4609 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4610 if (os_sigaction == NULL) return;
4611 }
4613 os_sigaction(sig, (struct sigaction*)NULL, &act);
4616 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4618 address thisHandler = (act.sa_flags & SA_SIGINFO)
4619 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4620 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4623 switch(sig) {
4624 case SIGSEGV:
4625 case SIGBUS:
4626 case SIGFPE:
4627 case SIGPIPE:
4628 case SIGILL:
4629 case SIGXFSZ:
4630 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4631 break;
4633 case SHUTDOWN1_SIGNAL:
4634 case SHUTDOWN2_SIGNAL:
4635 case SHUTDOWN3_SIGNAL:
4636 case BREAK_SIGNAL:
4637 jvmHandler = (address)user_handler();
4638 break;
4640 case INTERRUPT_SIGNAL:
4641 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4642 break;
4644 default:
4645 if (sig == SR_signum) {
4646 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4647 } else {
4648 return;
4649 }
4650 break;
4651 }
4653 if (thisHandler != jvmHandler) {
4654 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4655 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4656 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4657 // No need to check this sig any longer
4658 sigaddset(&check_signal_done, sig);
4659 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4660 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4661 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4662 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4663 // No need to check this sig any longer
4664 sigaddset(&check_signal_done, sig);
4665 }
4667 // Dump all the signal
4668 if (sigismember(&check_signal_done, sig)) {
4669 print_signal_handlers(tty, buf, O_BUFLEN);
4670 }
4671 }
4673 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4675 extern bool signal_name(int signo, char* buf, size_t len);
4677 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4678 if (0 < exception_code && exception_code <= SIGRTMAX) {
4679 // signal
4680 if (!signal_name(exception_code, buf, size)) {
4681 jio_snprintf(buf, size, "SIG%d", exception_code);
4682 }
4683 return buf;
4684 } else {
4685 return NULL;
4686 }
4687 }
4689 // this is called _before_ the most of global arguments have been parsed
4690 void os::init(void) {
4691 char dummy; /* used to get a guess on initial stack address */
4692 // first_hrtime = gethrtime();
4694 // With LinuxThreads the JavaMain thread pid (primordial thread)
4695 // is different than the pid of the java launcher thread.
4696 // So, on Linux, the launcher thread pid is passed to the VM
4697 // via the sun.java.launcher.pid property.
4698 // Use this property instead of getpid() if it was correctly passed.
4699 // See bug 6351349.
4700 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4702 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4704 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4706 init_random(1234567);
4708 ThreadCritical::initialize();
4710 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4711 if (Linux::page_size() == -1) {
4712 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4713 strerror(errno)));
4714 }
4715 init_page_sizes((size_t) Linux::page_size());
4717 Linux::initialize_system_info();
4719 // main_thread points to the aboriginal thread
4720 Linux::_main_thread = pthread_self();
4722 Linux::clock_init();
4723 initial_time_count = javaTimeNanos();
4725 // pthread_condattr initialization for monotonic clock
4726 int status;
4727 pthread_condattr_t* _condattr = os::Linux::condAttr();
4728 if ((status = pthread_condattr_init(_condattr)) != 0) {
4729 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4730 }
4731 // Only set the clock if CLOCK_MONOTONIC is available
4732 if (Linux::supports_monotonic_clock()) {
4733 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4734 if (status == EINVAL) {
4735 warning("Unable to use monotonic clock with relative timed-waits" \
4736 " - changes to the time-of-day clock may have adverse affects");
4737 } else {
4738 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4739 }
4740 }
4741 }
4742 // else it defaults to CLOCK_REALTIME
4744 pthread_mutex_init(&dl_mutex, NULL);
4746 // If the pagesize of the VM is greater than 8K determine the appropriate
4747 // number of initial guard pages. The user can change this with the
4748 // command line arguments, if needed.
4749 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4750 StackYellowPages = 1;
4751 StackRedPages = 1;
4752 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4753 }
4754 }
4756 // To install functions for atexit system call
4757 extern "C" {
4758 static void perfMemory_exit_helper() {
4759 perfMemory_exit();
4760 }
4761 }
4763 // this is called _after_ the global arguments have been parsed
4764 jint os::init_2(void)
4765 {
4766 Linux::fast_thread_clock_init();
4768 // Allocate a single page and mark it as readable for safepoint polling
4769 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4770 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4772 os::set_polling_page( polling_page );
4774 #ifndef PRODUCT
4775 if(Verbose && PrintMiscellaneous)
4776 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4777 #endif
4779 if (!UseMembar) {
4780 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4781 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4782 os::set_memory_serialize_page( mem_serialize_page );
4784 #ifndef PRODUCT
4785 if(Verbose && PrintMiscellaneous)
4786 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4787 #endif
4788 }
4790 // initialize suspend/resume support - must do this before signal_sets_init()
4791 if (SR_initialize() != 0) {
4792 perror("SR_initialize failed");
4793 return JNI_ERR;
4794 }
4796 Linux::signal_sets_init();
4797 Linux::install_signal_handlers();
4799 // Check minimum allowable stack size for thread creation and to initialize
4800 // the java system classes, including StackOverflowError - depends on page
4801 // size. Add a page for compiler2 recursion in main thread.
4802 // Add in 2*BytesPerWord times page size to account for VM stack during
4803 // class initialization depending on 32 or 64 bit VM.
4804 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4805 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4806 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4808 size_t threadStackSizeInBytes = ThreadStackSize * K;
4809 if (threadStackSizeInBytes != 0 &&
4810 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4811 tty->print_cr("\nThe stack size specified is too small, "
4812 "Specify at least %dk",
4813 os::Linux::min_stack_allowed/ K);
4814 return JNI_ERR;
4815 }
4817 // Make the stack size a multiple of the page size so that
4818 // the yellow/red zones can be guarded.
4819 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4820 vm_page_size()));
4822 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4824 #if defined(IA32)
4825 workaround_expand_exec_shield_cs_limit();
4826 #endif
4828 Linux::libpthread_init();
4829 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4830 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4831 Linux::glibc_version(), Linux::libpthread_version(),
4832 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4833 }
4835 if (UseNUMA) {
4836 if (!Linux::libnuma_init()) {
4837 UseNUMA = false;
4838 } else {
4839 if ((Linux::numa_max_node() < 1)) {
4840 // There's only one node(they start from 0), disable NUMA.
4841 UseNUMA = false;
4842 }
4843 }
4844 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4845 // we can make the adaptive lgrp chunk resizing work. If the user specified
4846 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4847 // disable adaptive resizing.
4848 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4849 if (FLAG_IS_DEFAULT(UseNUMA)) {
4850 UseNUMA = false;
4851 } else {
4852 if (FLAG_IS_DEFAULT(UseLargePages) &&
4853 FLAG_IS_DEFAULT(UseSHM) &&
4854 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4855 UseLargePages = false;
4856 } else {
4857 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4858 UseAdaptiveSizePolicy = false;
4859 UseAdaptiveNUMAChunkSizing = false;
4860 }
4861 }
4862 }
4863 if (!UseNUMA && ForceNUMA) {
4864 UseNUMA = true;
4865 }
4866 }
4868 if (MaxFDLimit) {
4869 // set the number of file descriptors to max. print out error
4870 // if getrlimit/setrlimit fails but continue regardless.
4871 struct rlimit nbr_files;
4872 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4873 if (status != 0) {
4874 if (PrintMiscellaneous && (Verbose || WizardMode))
4875 perror("os::init_2 getrlimit failed");
4876 } else {
4877 nbr_files.rlim_cur = nbr_files.rlim_max;
4878 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4879 if (status != 0) {
4880 if (PrintMiscellaneous && (Verbose || WizardMode))
4881 perror("os::init_2 setrlimit failed");
4882 }
4883 }
4884 }
4886 // Initialize lock used to serialize thread creation (see os::create_thread)
4887 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4889 // at-exit methods are called in the reverse order of their registration.
4890 // atexit functions are called on return from main or as a result of a
4891 // call to exit(3C). There can be only 32 of these functions registered
4892 // and atexit() does not set errno.
4894 if (PerfAllowAtExitRegistration) {
4895 // only register atexit functions if PerfAllowAtExitRegistration is set.
4896 // atexit functions can be delayed until process exit time, which
4897 // can be problematic for embedded VM situations. Embedded VMs should
4898 // call DestroyJavaVM() to assure that VM resources are released.
4900 // note: perfMemory_exit_helper atexit function may be removed in
4901 // the future if the appropriate cleanup code can be added to the
4902 // VM_Exit VMOperation's doit method.
4903 if (atexit(perfMemory_exit_helper) != 0) {
4904 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4905 }
4906 }
4908 // initialize thread priority policy
4909 prio_init();
4911 return JNI_OK;
4912 }
4914 // this is called at the end of vm_initialization
4915 void os::init_3(void) {
4916 #ifdef JAVASE_EMBEDDED
4917 // Start the MemNotifyThread
4918 if (LowMemoryProtection) {
4919 MemNotifyThread::start();
4920 }
4921 return;
4922 #endif
4923 }
4925 // Mark the polling page as unreadable
4926 void os::make_polling_page_unreadable(void) {
4927 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4928 fatal("Could not disable polling page");
4929 };
4931 // Mark the polling page as readable
4932 void os::make_polling_page_readable(void) {
4933 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4934 fatal("Could not enable polling page");
4935 }
4936 };
4938 int os::active_processor_count() {
4939 // Linux doesn't yet have a (official) notion of processor sets,
4940 // so just return the number of online processors.
4941 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4942 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4943 return online_cpus;
4944 }
4946 void os::set_native_thread_name(const char *name) {
4947 // Not yet implemented.
4948 return;
4949 }
4951 bool os::distribute_processes(uint length, uint* distribution) {
4952 // Not yet implemented.
4953 return false;
4954 }
4956 bool os::bind_to_processor(uint processor_id) {
4957 // Not yet implemented.
4958 return false;
4959 }
4961 ///
4963 void os::SuspendedThreadTask::internal_do_task() {
4964 if (do_suspend(_thread->osthread())) {
4965 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4966 do_task(context);
4967 do_resume(_thread->osthread());
4968 }
4969 }
4971 class PcFetcher : public os::SuspendedThreadTask {
4972 public:
4973 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4974 ExtendedPC result();
4975 protected:
4976 void do_task(const os::SuspendedThreadTaskContext& context);
4977 private:
4978 ExtendedPC _epc;
4979 };
4981 ExtendedPC PcFetcher::result() {
4982 guarantee(is_done(), "task is not done yet.");
4983 return _epc;
4984 }
4986 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4987 Thread* thread = context.thread();
4988 OSThread* osthread = thread->osthread();
4989 if (osthread->ucontext() != NULL) {
4990 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4991 } else {
4992 // NULL context is unexpected, double-check this is the VMThread
4993 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4994 }
4995 }
4997 // Suspends the target using the signal mechanism and then grabs the PC before
4998 // resuming the target. Used by the flat-profiler only
4999 ExtendedPC os::get_thread_pc(Thread* thread) {
5000 // Make sure that it is called by the watcher for the VMThread
5001 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5002 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5004 PcFetcher fetcher(thread);
5005 fetcher.run();
5006 return fetcher.result();
5007 }
5009 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5010 {
5011 if (is_NPTL()) {
5012 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5013 } else {
5014 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5015 // word back to default 64bit precision if condvar is signaled. Java
5016 // wants 53bit precision. Save and restore current value.
5017 int fpu = get_fpu_control_word();
5018 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5019 set_fpu_control_word(fpu);
5020 return status;
5021 }
5022 }
5024 ////////////////////////////////////////////////////////////////////////////////
5025 // debug support
5027 bool os::find(address addr, outputStream* st) {
5028 Dl_info dlinfo;
5029 memset(&dlinfo, 0, sizeof(dlinfo));
5030 if (dladdr(addr, &dlinfo) != 0) {
5031 st->print(PTR_FORMAT ": ", addr);
5032 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5033 st->print("%s+%#x", dlinfo.dli_sname,
5034 addr - (intptr_t)dlinfo.dli_saddr);
5035 } else if (dlinfo.dli_fbase != NULL) {
5036 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5037 } else {
5038 st->print("<absolute address>");
5039 }
5040 if (dlinfo.dli_fname != NULL) {
5041 st->print(" in %s", dlinfo.dli_fname);
5042 }
5043 if (dlinfo.dli_fbase != NULL) {
5044 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5045 }
5046 st->cr();
5048 if (Verbose) {
5049 // decode some bytes around the PC
5050 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5051 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5052 address lowest = (address) dlinfo.dli_sname;
5053 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5054 if (begin < lowest) begin = lowest;
5055 Dl_info dlinfo2;
5056 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5057 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5058 end = (address) dlinfo2.dli_saddr;
5059 Disassembler::decode(begin, end, st);
5060 }
5061 return true;
5062 }
5063 return false;
5064 }
5066 ////////////////////////////////////////////////////////////////////////////////
5067 // misc
5069 // This does not do anything on Linux. This is basically a hook for being
5070 // able to use structured exception handling (thread-local exception filters)
5071 // on, e.g., Win32.
5072 void
5073 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5074 JavaCallArguments* args, Thread* thread) {
5075 f(value, method, args, thread);
5076 }
5078 void os::print_statistics() {
5079 }
5081 int os::message_box(const char* title, const char* message) {
5082 int i;
5083 fdStream err(defaultStream::error_fd());
5084 for (i = 0; i < 78; i++) err.print_raw("=");
5085 err.cr();
5086 err.print_raw_cr(title);
5087 for (i = 0; i < 78; i++) err.print_raw("-");
5088 err.cr();
5089 err.print_raw_cr(message);
5090 for (i = 0; i < 78; i++) err.print_raw("=");
5091 err.cr();
5093 char buf[16];
5094 // Prevent process from exiting upon "read error" without consuming all CPU
5095 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5097 return buf[0] == 'y' || buf[0] == 'Y';
5098 }
5100 int os::stat(const char *path, struct stat *sbuf) {
5101 char pathbuf[MAX_PATH];
5102 if (strlen(path) > MAX_PATH - 1) {
5103 errno = ENAMETOOLONG;
5104 return -1;
5105 }
5106 os::native_path(strcpy(pathbuf, path));
5107 return ::stat(pathbuf, sbuf);
5108 }
5110 bool os::check_heap(bool force) {
5111 return true;
5112 }
5114 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5115 return ::vsnprintf(buf, count, format, args);
5116 }
5118 // Is a (classpath) directory empty?
5119 bool os::dir_is_empty(const char* path) {
5120 DIR *dir = NULL;
5121 struct dirent *ptr;
5123 dir = opendir(path);
5124 if (dir == NULL) return true;
5126 /* Scan the directory */
5127 bool result = true;
5128 char buf[sizeof(struct dirent) + MAX_PATH];
5129 while (result && (ptr = ::readdir(dir)) != NULL) {
5130 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5131 result = false;
5132 }
5133 }
5134 closedir(dir);
5135 return result;
5136 }
5138 // This code originates from JDK's sysOpen and open64_w
5139 // from src/solaris/hpi/src/system_md.c
5141 #ifndef O_DELETE
5142 #define O_DELETE 0x10000
5143 #endif
5145 // Open a file. Unlink the file immediately after open returns
5146 // if the specified oflag has the O_DELETE flag set.
5147 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5149 int os::open(const char *path, int oflag, int mode) {
5151 if (strlen(path) > MAX_PATH - 1) {
5152 errno = ENAMETOOLONG;
5153 return -1;
5154 }
5155 int fd;
5156 int o_delete = (oflag & O_DELETE);
5157 oflag = oflag & ~O_DELETE;
5159 fd = ::open64(path, oflag, mode);
5160 if (fd == -1) return -1;
5162 //If the open succeeded, the file might still be a directory
5163 {
5164 struct stat64 buf64;
5165 int ret = ::fstat64(fd, &buf64);
5166 int st_mode = buf64.st_mode;
5168 if (ret != -1) {
5169 if ((st_mode & S_IFMT) == S_IFDIR) {
5170 errno = EISDIR;
5171 ::close(fd);
5172 return -1;
5173 }
5174 } else {
5175 ::close(fd);
5176 return -1;
5177 }
5178 }
5180 /*
5181 * All file descriptors that are opened in the JVM and not
5182 * specifically destined for a subprocess should have the
5183 * close-on-exec flag set. If we don't set it, then careless 3rd
5184 * party native code might fork and exec without closing all
5185 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5186 * UNIXProcess.c), and this in turn might:
5187 *
5188 * - cause end-of-file to fail to be detected on some file
5189 * descriptors, resulting in mysterious hangs, or
5190 *
5191 * - might cause an fopen in the subprocess to fail on a system
5192 * suffering from bug 1085341.
5193 *
5194 * (Yes, the default setting of the close-on-exec flag is a Unix
5195 * design flaw)
5196 *
5197 * See:
5198 * 1085341: 32-bit stdio routines should support file descriptors >255
5199 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5200 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5201 */
5202 #ifdef FD_CLOEXEC
5203 {
5204 int flags = ::fcntl(fd, F_GETFD);
5205 if (flags != -1)
5206 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5207 }
5208 #endif
5210 if (o_delete != 0) {
5211 ::unlink(path);
5212 }
5213 return fd;
5214 }
5217 // create binary file, rewriting existing file if required
5218 int os::create_binary_file(const char* path, bool rewrite_existing) {
5219 int oflags = O_WRONLY | O_CREAT;
5220 if (!rewrite_existing) {
5221 oflags |= O_EXCL;
5222 }
5223 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5224 }
5226 // return current position of file pointer
5227 jlong os::current_file_offset(int fd) {
5228 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5229 }
5231 // move file pointer to the specified offset
5232 jlong os::seek_to_file_offset(int fd, jlong offset) {
5233 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5234 }
5236 // This code originates from JDK's sysAvailable
5237 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5239 int os::available(int fd, jlong *bytes) {
5240 jlong cur, end;
5241 int mode;
5242 struct stat64 buf64;
5244 if (::fstat64(fd, &buf64) >= 0) {
5245 mode = buf64.st_mode;
5246 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5247 /*
5248 * XXX: is the following call interruptible? If so, this might
5249 * need to go through the INTERRUPT_IO() wrapper as for other
5250 * blocking, interruptible calls in this file.
5251 */
5252 int n;
5253 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5254 *bytes = n;
5255 return 1;
5256 }
5257 }
5258 }
5259 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5260 return 0;
5261 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5262 return 0;
5263 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5264 return 0;
5265 }
5266 *bytes = end - cur;
5267 return 1;
5268 }
5270 int os::socket_available(int fd, jint *pbytes) {
5271 // Linux doc says EINTR not returned, unlike Solaris
5272 int ret = ::ioctl(fd, FIONREAD, pbytes);
5274 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5275 // is expected to return 0 on failure and 1 on success to the jdk.
5276 return (ret < 0) ? 0 : 1;
5277 }
5279 // Map a block of memory.
5280 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5281 char *addr, size_t bytes, bool read_only,
5282 bool allow_exec) {
5283 int prot;
5284 int flags = MAP_PRIVATE;
5286 if (read_only) {
5287 prot = PROT_READ;
5288 } else {
5289 prot = PROT_READ | PROT_WRITE;
5290 }
5292 if (allow_exec) {
5293 prot |= PROT_EXEC;
5294 }
5296 if (addr != NULL) {
5297 flags |= MAP_FIXED;
5298 }
5300 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5301 fd, file_offset);
5302 if (mapped_address == MAP_FAILED) {
5303 return NULL;
5304 }
5305 return mapped_address;
5306 }
5309 // Remap a block of memory.
5310 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5311 char *addr, size_t bytes, bool read_only,
5312 bool allow_exec) {
5313 // same as map_memory() on this OS
5314 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5315 allow_exec);
5316 }
5319 // Unmap a block of memory.
5320 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5321 return munmap(addr, bytes) == 0;
5322 }
5324 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5326 static clockid_t thread_cpu_clockid(Thread* thread) {
5327 pthread_t tid = thread->osthread()->pthread_id();
5328 clockid_t clockid;
5330 // Get thread clockid
5331 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5332 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5333 return clockid;
5334 }
5336 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5337 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5338 // of a thread.
5339 //
5340 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5341 // the fast estimate available on the platform.
5343 jlong os::current_thread_cpu_time() {
5344 if (os::Linux::supports_fast_thread_cpu_time()) {
5345 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5346 } else {
5347 // return user + sys since the cost is the same
5348 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5349 }
5350 }
5352 jlong os::thread_cpu_time(Thread* thread) {
5353 // consistent with what current_thread_cpu_time() returns
5354 if (os::Linux::supports_fast_thread_cpu_time()) {
5355 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5356 } else {
5357 return slow_thread_cpu_time(thread, true /* user + sys */);
5358 }
5359 }
5361 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5362 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5363 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5364 } else {
5365 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5366 }
5367 }
5369 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5370 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5371 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5372 } else {
5373 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5374 }
5375 }
5377 //
5378 // -1 on error.
5379 //
5381 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5382 static bool proc_task_unchecked = true;
5383 static const char *proc_stat_path = "/proc/%d/stat";
5384 pid_t tid = thread->osthread()->thread_id();
5385 char *s;
5386 char stat[2048];
5387 int statlen;
5388 char proc_name[64];
5389 int count;
5390 long sys_time, user_time;
5391 char cdummy;
5392 int idummy;
5393 long ldummy;
5394 FILE *fp;
5396 // The /proc/<tid>/stat aggregates per-process usage on
5397 // new Linux kernels 2.6+ where NPTL is supported.
5398 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5399 // See bug 6328462.
5400 // There possibly can be cases where there is no directory
5401 // /proc/self/task, so we check its availability.
5402 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5403 // This is executed only once
5404 proc_task_unchecked = false;
5405 fp = fopen("/proc/self/task", "r");
5406 if (fp != NULL) {
5407 proc_stat_path = "/proc/self/task/%d/stat";
5408 fclose(fp);
5409 }
5410 }
5412 sprintf(proc_name, proc_stat_path, tid);
5413 fp = fopen(proc_name, "r");
5414 if ( fp == NULL ) return -1;
5415 statlen = fread(stat, 1, 2047, fp);
5416 stat[statlen] = '\0';
5417 fclose(fp);
5419 // Skip pid and the command string. Note that we could be dealing with
5420 // weird command names, e.g. user could decide to rename java launcher
5421 // to "java 1.4.2 :)", then the stat file would look like
5422 // 1234 (java 1.4.2 :)) R ... ...
5423 // We don't really need to know the command string, just find the last
5424 // occurrence of ")" and then start parsing from there. See bug 4726580.
5425 s = strrchr(stat, ')');
5426 if (s == NULL ) return -1;
5428 // Skip blank chars
5429 do s++; while (isspace(*s));
5431 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5432 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5433 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5434 &user_time, &sys_time);
5435 if ( count != 13 ) return -1;
5436 if (user_sys_cpu_time) {
5437 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5438 } else {
5439 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5440 }
5441 }
5443 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5444 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5445 info_ptr->may_skip_backward = false; // elapsed time not wall time
5446 info_ptr->may_skip_forward = false; // elapsed time not wall time
5447 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5448 }
5450 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5451 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5452 info_ptr->may_skip_backward = false; // elapsed time not wall time
5453 info_ptr->may_skip_forward = false; // elapsed time not wall time
5454 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5455 }
5457 bool os::is_thread_cpu_time_supported() {
5458 return true;
5459 }
5461 // System loadavg support. Returns -1 if load average cannot be obtained.
5462 // Linux doesn't yet have a (official) notion of processor sets,
5463 // so just return the system wide load average.
5464 int os::loadavg(double loadavg[], int nelem) {
5465 return ::getloadavg(loadavg, nelem);
5466 }
5468 void os::pause() {
5469 char filename[MAX_PATH];
5470 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5471 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5472 } else {
5473 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5474 }
5476 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5477 if (fd != -1) {
5478 struct stat buf;
5479 ::close(fd);
5480 while (::stat(filename, &buf) == 0) {
5481 (void)::poll(NULL, 0, 100);
5482 }
5483 } else {
5484 jio_fprintf(stderr,
5485 "Could not open pause file '%s', continuing immediately.\n", filename);
5486 }
5487 }
5490 // Refer to the comments in os_solaris.cpp park-unpark.
5491 //
5492 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5493 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5494 // For specifics regarding the bug see GLIBC BUGID 261237 :
5495 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5496 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5497 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5498 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5499 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5500 // and monitorenter when we're using 1-0 locking. All those operations may result in
5501 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5502 // of libpthread avoids the problem, but isn't practical.
5503 //
5504 // Possible remedies:
5505 //
5506 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5507 // This is palliative and probabilistic, however. If the thread is preempted
5508 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5509 // than the minimum period may have passed, and the abstime may be stale (in the
5510 // past) resultin in a hang. Using this technique reduces the odds of a hang
5511 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5512 //
5513 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5514 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5515 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5516 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5517 // thread.
5518 //
5519 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5520 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5521 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5522 // This also works well. In fact it avoids kernel-level scalability impediments
5523 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5524 // timers in a graceful fashion.
5525 //
5526 // 4. When the abstime value is in the past it appears that control returns
5527 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5528 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5529 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5530 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5531 // It may be possible to avoid reinitialization by checking the return
5532 // value from pthread_cond_timedwait(). In addition to reinitializing the
5533 // condvar we must establish the invariant that cond_signal() is only called
5534 // within critical sections protected by the adjunct mutex. This prevents
5535 // cond_signal() from "seeing" a condvar that's in the midst of being
5536 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5537 // desirable signal-after-unlock optimization that avoids futile context switching.
5538 //
5539 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5540 // structure when a condvar is used or initialized. cond_destroy() would
5541 // release the helper structure. Our reinitialize-after-timedwait fix
5542 // put excessive stress on malloc/free and locks protecting the c-heap.
5543 //
5544 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5545 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5546 // and only enabling the work-around for vulnerable environments.
5548 // utility to compute the abstime argument to timedwait:
5549 // millis is the relative timeout time
5550 // abstime will be the absolute timeout time
5551 // TODO: replace compute_abstime() with unpackTime()
5553 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5554 if (millis < 0) millis = 0;
5556 jlong seconds = millis / 1000;
5557 millis %= 1000;
5558 if (seconds > 50000000) { // see man cond_timedwait(3T)
5559 seconds = 50000000;
5560 }
5562 if (os::Linux::supports_monotonic_clock()) {
5563 struct timespec now;
5564 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5565 assert_status(status == 0, status, "clock_gettime");
5566 abstime->tv_sec = now.tv_sec + seconds;
5567 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5568 if (nanos >= NANOSECS_PER_SEC) {
5569 abstime->tv_sec += 1;
5570 nanos -= NANOSECS_PER_SEC;
5571 }
5572 abstime->tv_nsec = nanos;
5573 } else {
5574 struct timeval now;
5575 int status = gettimeofday(&now, NULL);
5576 assert(status == 0, "gettimeofday");
5577 abstime->tv_sec = now.tv_sec + seconds;
5578 long usec = now.tv_usec + millis * 1000;
5579 if (usec >= 1000000) {
5580 abstime->tv_sec += 1;
5581 usec -= 1000000;
5582 }
5583 abstime->tv_nsec = usec * 1000;
5584 }
5585 return abstime;
5586 }
5589 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5590 // Conceptually TryPark() should be equivalent to park(0).
5592 int os::PlatformEvent::TryPark() {
5593 for (;;) {
5594 const int v = _Event ;
5595 guarantee ((v == 0) || (v == 1), "invariant") ;
5596 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5597 }
5598 }
5600 void os::PlatformEvent::park() { // AKA "down()"
5601 // Invariant: Only the thread associated with the Event/PlatformEvent
5602 // may call park().
5603 // TODO: assert that _Assoc != NULL or _Assoc == Self
5604 int v ;
5605 for (;;) {
5606 v = _Event ;
5607 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5608 }
5609 guarantee (v >= 0, "invariant") ;
5610 if (v == 0) {
5611 // Do this the hard way by blocking ...
5612 int status = pthread_mutex_lock(_mutex);
5613 assert_status(status == 0, status, "mutex_lock");
5614 guarantee (_nParked == 0, "invariant") ;
5615 ++ _nParked ;
5616 while (_Event < 0) {
5617 status = pthread_cond_wait(_cond, _mutex);
5618 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5619 // Treat this the same as if the wait was interrupted
5620 if (status == ETIME) { status = EINTR; }
5621 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5622 }
5623 -- _nParked ;
5625 _Event = 0 ;
5626 status = pthread_mutex_unlock(_mutex);
5627 assert_status(status == 0, status, "mutex_unlock");
5628 // Paranoia to ensure our locked and lock-free paths interact
5629 // correctly with each other.
5630 OrderAccess::fence();
5631 }
5632 guarantee (_Event >= 0, "invariant") ;
5633 }
5635 int os::PlatformEvent::park(jlong millis) {
5636 guarantee (_nParked == 0, "invariant") ;
5638 int v ;
5639 for (;;) {
5640 v = _Event ;
5641 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5642 }
5643 guarantee (v >= 0, "invariant") ;
5644 if (v != 0) return OS_OK ;
5646 // We do this the hard way, by blocking the thread.
5647 // Consider enforcing a minimum timeout value.
5648 struct timespec abst;
5649 compute_abstime(&abst, millis);
5651 int ret = OS_TIMEOUT;
5652 int status = pthread_mutex_lock(_mutex);
5653 assert_status(status == 0, status, "mutex_lock");
5654 guarantee (_nParked == 0, "invariant") ;
5655 ++_nParked ;
5657 // Object.wait(timo) will return because of
5658 // (a) notification
5659 // (b) timeout
5660 // (c) thread.interrupt
5661 //
5662 // Thread.interrupt and object.notify{All} both call Event::set.
5663 // That is, we treat thread.interrupt as a special case of notification.
5664 // The underlying Solaris implementation, cond_timedwait, admits
5665 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5666 // JVM from making those visible to Java code. As such, we must
5667 // filter out spurious wakeups. We assume all ETIME returns are valid.
5668 //
5669 // TODO: properly differentiate simultaneous notify+interrupt.
5670 // In that case, we should propagate the notify to another waiter.
5672 while (_Event < 0) {
5673 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5674 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5675 pthread_cond_destroy (_cond);
5676 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5677 }
5678 assert_status(status == 0 || status == EINTR ||
5679 status == ETIME || status == ETIMEDOUT,
5680 status, "cond_timedwait");
5681 if (!FilterSpuriousWakeups) break ; // previous semantics
5682 if (status == ETIME || status == ETIMEDOUT) break ;
5683 // We consume and ignore EINTR and spurious wakeups.
5684 }
5685 --_nParked ;
5686 if (_Event >= 0) {
5687 ret = OS_OK;
5688 }
5689 _Event = 0 ;
5690 status = pthread_mutex_unlock(_mutex);
5691 assert_status(status == 0, status, "mutex_unlock");
5692 assert (_nParked == 0, "invariant") ;
5693 // Paranoia to ensure our locked and lock-free paths interact
5694 // correctly with each other.
5695 OrderAccess::fence();
5696 return ret;
5697 }
5699 void os::PlatformEvent::unpark() {
5700 // Transitions for _Event:
5701 // 0 :=> 1
5702 // 1 :=> 1
5703 // -1 :=> either 0 or 1; must signal target thread
5704 // That is, we can safely transition _Event from -1 to either
5705 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5706 // unpark() calls.
5707 // See also: "Semaphores in Plan 9" by Mullender & Cox
5708 //
5709 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5710 // that it will take two back-to-back park() calls for the owning
5711 // thread to block. This has the benefit of forcing a spurious return
5712 // from the first park() call after an unpark() call which will help
5713 // shake out uses of park() and unpark() without condition variables.
5715 if (Atomic::xchg(1, &_Event) >= 0) return;
5717 // Wait for the thread associated with the event to vacate
5718 int status = pthread_mutex_lock(_mutex);
5719 assert_status(status == 0, status, "mutex_lock");
5720 int AnyWaiters = _nParked;
5721 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5722 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5723 AnyWaiters = 0;
5724 pthread_cond_signal(_cond);
5725 }
5726 status = pthread_mutex_unlock(_mutex);
5727 assert_status(status == 0, status, "mutex_unlock");
5728 if (AnyWaiters != 0) {
5729 status = pthread_cond_signal(_cond);
5730 assert_status(status == 0, status, "cond_signal");
5731 }
5733 // Note that we signal() _after dropping the lock for "immortal" Events.
5734 // This is safe and avoids a common class of futile wakeups. In rare
5735 // circumstances this can cause a thread to return prematurely from
5736 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5737 // simply re-test the condition and re-park itself.
5738 }
5741 // JSR166
5742 // -------------------------------------------------------
5744 /*
5745 * The solaris and linux implementations of park/unpark are fairly
5746 * conservative for now, but can be improved. They currently use a
5747 * mutex/condvar pair, plus a a count.
5748 * Park decrements count if > 0, else does a condvar wait. Unpark
5749 * sets count to 1 and signals condvar. Only one thread ever waits
5750 * on the condvar. Contention seen when trying to park implies that someone
5751 * is unparking you, so don't wait. And spurious returns are fine, so there
5752 * is no need to track notifications.
5753 */
5755 #define MAX_SECS 100000000
5756 /*
5757 * This code is common to linux and solaris and will be moved to a
5758 * common place in dolphin.
5759 *
5760 * The passed in time value is either a relative time in nanoseconds
5761 * or an absolute time in milliseconds. Either way it has to be unpacked
5762 * into suitable seconds and nanoseconds components and stored in the
5763 * given timespec structure.
5764 * Given time is a 64-bit value and the time_t used in the timespec is only
5765 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5766 * overflow if times way in the future are given. Further on Solaris versions
5767 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5768 * number of seconds, in abstime, is less than current_time + 100,000,000.
5769 * As it will be 28 years before "now + 100000000" will overflow we can
5770 * ignore overflow and just impose a hard-limit on seconds using the value
5771 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5772 * years from "now".
5773 */
5775 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5776 assert (time > 0, "convertTime");
5777 time_t max_secs = 0;
5779 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5780 struct timeval now;
5781 int status = gettimeofday(&now, NULL);
5782 assert(status == 0, "gettimeofday");
5784 max_secs = now.tv_sec + MAX_SECS;
5786 if (isAbsolute) {
5787 jlong secs = time / 1000;
5788 if (secs > max_secs) {
5789 absTime->tv_sec = max_secs;
5790 } else {
5791 absTime->tv_sec = secs;
5792 }
5793 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5794 } else {
5795 jlong secs = time / NANOSECS_PER_SEC;
5796 if (secs >= MAX_SECS) {
5797 absTime->tv_sec = max_secs;
5798 absTime->tv_nsec = 0;
5799 } else {
5800 absTime->tv_sec = now.tv_sec + secs;
5801 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5802 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5803 absTime->tv_nsec -= NANOSECS_PER_SEC;
5804 ++absTime->tv_sec; // note: this must be <= max_secs
5805 }
5806 }
5807 }
5808 } else {
5809 // must be relative using monotonic clock
5810 struct timespec now;
5811 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5812 assert_status(status == 0, status, "clock_gettime");
5813 max_secs = now.tv_sec + MAX_SECS;
5814 jlong secs = time / NANOSECS_PER_SEC;
5815 if (secs >= MAX_SECS) {
5816 absTime->tv_sec = max_secs;
5817 absTime->tv_nsec = 0;
5818 } else {
5819 absTime->tv_sec = now.tv_sec + secs;
5820 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5821 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5822 absTime->tv_nsec -= NANOSECS_PER_SEC;
5823 ++absTime->tv_sec; // note: this must be <= max_secs
5824 }
5825 }
5826 }
5827 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5828 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5829 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5830 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5831 }
5833 void Parker::park(bool isAbsolute, jlong time) {
5834 // Ideally we'd do something useful while spinning, such
5835 // as calling unpackTime().
5837 // Optional fast-path check:
5838 // Return immediately if a permit is available.
5839 // We depend on Atomic::xchg() having full barrier semantics
5840 // since we are doing a lock-free update to _counter.
5841 if (Atomic::xchg(0, &_counter) > 0) return;
5843 Thread* thread = Thread::current();
5844 assert(thread->is_Java_thread(), "Must be JavaThread");
5845 JavaThread *jt = (JavaThread *)thread;
5847 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5848 // Check interrupt before trying to wait
5849 if (Thread::is_interrupted(thread, false)) {
5850 return;
5851 }
5853 // Next, demultiplex/decode time arguments
5854 timespec absTime;
5855 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5856 return;
5857 }
5858 if (time > 0) {
5859 unpackTime(&absTime, isAbsolute, time);
5860 }
5863 // Enter safepoint region
5864 // Beware of deadlocks such as 6317397.
5865 // The per-thread Parker:: mutex is a classic leaf-lock.
5866 // In particular a thread must never block on the Threads_lock while
5867 // holding the Parker:: mutex. If safepoints are pending both the
5868 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5869 ThreadBlockInVM tbivm(jt);
5871 // Don't wait if cannot get lock since interference arises from
5872 // unblocking. Also. check interrupt before trying wait
5873 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5874 return;
5875 }
5877 int status ;
5878 if (_counter > 0) { // no wait needed
5879 _counter = 0;
5880 status = pthread_mutex_unlock(_mutex);
5881 assert (status == 0, "invariant") ;
5882 // Paranoia to ensure our locked and lock-free paths interact
5883 // correctly with each other and Java-level accesses.
5884 OrderAccess::fence();
5885 return;
5886 }
5888 #ifdef ASSERT
5889 // Don't catch signals while blocked; let the running threads have the signals.
5890 // (This allows a debugger to break into the running thread.)
5891 sigset_t oldsigs;
5892 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5893 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5894 #endif
5896 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5897 jt->set_suspend_equivalent();
5898 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5900 assert(_cur_index == -1, "invariant");
5901 if (time == 0) {
5902 _cur_index = REL_INDEX; // arbitrary choice when not timed
5903 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5904 } else {
5905 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5906 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5907 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5908 pthread_cond_destroy (&_cond[_cur_index]) ;
5909 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5910 }
5911 }
5912 _cur_index = -1;
5913 assert_status(status == 0 || status == EINTR ||
5914 status == ETIME || status == ETIMEDOUT,
5915 status, "cond_timedwait");
5917 #ifdef ASSERT
5918 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5919 #endif
5921 _counter = 0 ;
5922 status = pthread_mutex_unlock(_mutex) ;
5923 assert_status(status == 0, status, "invariant") ;
5924 // Paranoia to ensure our locked and lock-free paths interact
5925 // correctly with each other and Java-level accesses.
5926 OrderAccess::fence();
5928 // If externally suspended while waiting, re-suspend
5929 if (jt->handle_special_suspend_equivalent_condition()) {
5930 jt->java_suspend_self();
5931 }
5932 }
5934 void Parker::unpark() {
5935 int s, status ;
5936 status = pthread_mutex_lock(_mutex);
5937 assert (status == 0, "invariant") ;
5938 s = _counter;
5939 _counter = 1;
5940 if (s < 1) {
5941 // thread might be parked
5942 if (_cur_index != -1) {
5943 // thread is definitely parked
5944 if (WorkAroundNPTLTimedWaitHang) {
5945 status = pthread_cond_signal (&_cond[_cur_index]);
5946 assert (status == 0, "invariant");
5947 status = pthread_mutex_unlock(_mutex);
5948 assert (status == 0, "invariant");
5949 } else {
5950 status = pthread_mutex_unlock(_mutex);
5951 assert (status == 0, "invariant");
5952 status = pthread_cond_signal (&_cond[_cur_index]);
5953 assert (status == 0, "invariant");
5954 }
5955 } else {
5956 pthread_mutex_unlock(_mutex);
5957 assert (status == 0, "invariant") ;
5958 }
5959 } else {
5960 pthread_mutex_unlock(_mutex);
5961 assert (status == 0, "invariant") ;
5962 }
5963 }
5966 extern char** environ;
5968 #ifndef __NR_fork
5969 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5970 #endif
5972 #ifndef __NR_execve
5973 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5974 #endif
5976 // Run the specified command in a separate process. Return its exit value,
5977 // or -1 on failure (e.g. can't fork a new process).
5978 // Unlike system(), this function can be called from signal handler. It
5979 // doesn't block SIGINT et al.
5980 int os::fork_and_exec(char* cmd) {
5981 const char * argv[4] = {"sh", "-c", cmd, NULL};
5983 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5984 // pthread_atfork handlers and reset pthread library. All we need is a
5985 // separate process to execve. Make a direct syscall to fork process.
5986 // On IA64 there's no fork syscall, we have to use fork() and hope for
5987 // the best...
5988 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5989 IA64_ONLY(fork();)
5991 if (pid < 0) {
5992 // fork failed
5993 return -1;
5995 } else if (pid == 0) {
5996 // child process
5998 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5999 // first to kill every thread on the thread list. Because this list is
6000 // not reset by fork() (see notes above), execve() will instead kill
6001 // every thread in the parent process. We know this is the only thread
6002 // in the new process, so make a system call directly.
6003 // IA64 should use normal execve() from glibc to match the glibc fork()
6004 // above.
6005 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6006 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6008 // execve failed
6009 _exit(-1);
6011 } else {
6012 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6013 // care about the actual exit code, for now.
6015 int status;
6017 // Wait for the child process to exit. This returns immediately if
6018 // the child has already exited. */
6019 while (waitpid(pid, &status, 0) < 0) {
6020 switch (errno) {
6021 case ECHILD: return 0;
6022 case EINTR: break;
6023 default: return -1;
6024 }
6025 }
6027 if (WIFEXITED(status)) {
6028 // The child exited normally; get its exit code.
6029 return WEXITSTATUS(status);
6030 } else if (WIFSIGNALED(status)) {
6031 // The child exited because of a signal
6032 // The best value to return is 0x80 + signal number,
6033 // because that is what all Unix shells do, and because
6034 // it allows callers to distinguish between process exit and
6035 // process death by signal.
6036 return 0x80 + WTERMSIG(status);
6037 } else {
6038 // Unknown exit code; pass it through
6039 return status;
6040 }
6041 }
6042 }
6044 // is_headless_jre()
6045 //
6046 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6047 // in order to report if we are running in a headless jre
6048 //
6049 // Since JDK8 xawt/libmawt.so was moved into the same directory
6050 // as libawt.so, and renamed libawt_xawt.so
6051 //
6052 bool os::is_headless_jre() {
6053 struct stat statbuf;
6054 char buf[MAXPATHLEN];
6055 char libmawtpath[MAXPATHLEN];
6056 const char *xawtstr = "/xawt/libmawt.so";
6057 const char *new_xawtstr = "/libawt_xawt.so";
6058 char *p;
6060 // Get path to libjvm.so
6061 os::jvm_path(buf, sizeof(buf));
6063 // Get rid of libjvm.so
6064 p = strrchr(buf, '/');
6065 if (p == NULL) return false;
6066 else *p = '\0';
6068 // Get rid of client or server
6069 p = strrchr(buf, '/');
6070 if (p == NULL) return false;
6071 else *p = '\0';
6073 // check xawt/libmawt.so
6074 strcpy(libmawtpath, buf);
6075 strcat(libmawtpath, xawtstr);
6076 if (::stat(libmawtpath, &statbuf) == 0) return false;
6078 // check libawt_xawt.so
6079 strcpy(libmawtpath, buf);
6080 strcat(libmawtpath, new_xawtstr);
6081 if (::stat(libmawtpath, &statbuf) == 0) return false;
6083 return true;
6084 }
6086 // Get the default path to the core file
6087 // Returns the length of the string
6088 int os::get_core_path(char* buffer, size_t bufferSize) {
6089 const char* p = get_current_directory(buffer, bufferSize);
6091 if (p == NULL) {
6092 assert(p != NULL, "failed to get current directory");
6093 return 0;
6094 }
6096 return strlen(buffer);
6097 }
6099 #ifdef JAVASE_EMBEDDED
6100 //
6101 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6102 //
6103 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6105 // ctor
6106 //
6107 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6108 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6109 _fd = fd;
6111 if (os::create_thread(this, os::os_thread)) {
6112 _memnotify_thread = this;
6113 os::set_priority(this, NearMaxPriority);
6114 os::start_thread(this);
6115 }
6116 }
6118 // Where all the work gets done
6119 //
6120 void MemNotifyThread::run() {
6121 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6123 // Set up the select arguments
6124 fd_set rfds;
6125 if (_fd != -1) {
6126 FD_ZERO(&rfds);
6127 FD_SET(_fd, &rfds);
6128 }
6130 // Now wait for the mem_notify device to wake up
6131 while (1) {
6132 // Wait for the mem_notify device to signal us..
6133 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6134 if (rc == -1) {
6135 perror("select!\n");
6136 break;
6137 } else if (rc) {
6138 //ssize_t free_before = os::available_memory();
6139 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6141 // The kernel is telling us there is not much memory left...
6142 // try to do something about that
6144 // If we are not already in a GC, try one.
6145 if (!Universe::heap()->is_gc_active()) {
6146 Universe::heap()->collect(GCCause::_allocation_failure);
6148 //ssize_t free_after = os::available_memory();
6149 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6150 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6151 }
6152 // We might want to do something like the following if we find the GC's are not helping...
6153 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6154 }
6155 }
6156 }
6158 //
6159 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6160 //
6161 void MemNotifyThread::start() {
6162 int fd;
6163 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6164 if (fd < 0) {
6165 return;
6166 }
6168 if (memnotify_thread() == NULL) {
6169 new MemNotifyThread(fd);
6170 }
6171 }
6173 #endif // JAVASE_EMBEDDED
6176 /////////////// Unit tests ///////////////
6178 #ifndef PRODUCT
6180 #define test_log(...) \
6181 do {\
6182 if (VerboseInternalVMTests) { \
6183 tty->print_cr(__VA_ARGS__); \
6184 tty->flush(); \
6185 }\
6186 } while (false)
6188 class TestReserveMemorySpecial : AllStatic {
6189 public:
6190 static void small_page_write(void* addr, size_t size) {
6191 size_t page_size = os::vm_page_size();
6193 char* end = (char*)addr + size;
6194 for (char* p = (char*)addr; p < end; p += page_size) {
6195 *p = 1;
6196 }
6197 }
6199 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6200 if (!UseHugeTLBFS) {
6201 return;
6202 }
6204 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6206 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6208 if (addr != NULL) {
6209 small_page_write(addr, size);
6211 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6212 }
6213 }
6215 static void test_reserve_memory_special_huge_tlbfs_only() {
6216 if (!UseHugeTLBFS) {
6217 return;
6218 }
6220 size_t lp = os::large_page_size();
6222 for (size_t size = lp; size <= lp * 10; size += lp) {
6223 test_reserve_memory_special_huge_tlbfs_only(size);
6224 }
6225 }
6227 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6228 if (!UseHugeTLBFS) {
6229 return;
6230 }
6232 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6233 size, alignment);
6235 assert(size >= os::large_page_size(), "Incorrect input to test");
6237 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6239 if (addr != NULL) {
6240 small_page_write(addr, size);
6242 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6243 }
6244 }
6246 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6247 size_t lp = os::large_page_size();
6248 size_t ag = os::vm_allocation_granularity();
6250 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6251 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6252 }
6253 }
6255 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6256 size_t lp = os::large_page_size();
6257 size_t ag = os::vm_allocation_granularity();
6259 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6260 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6261 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6262 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6263 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6264 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6265 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6266 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6267 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6268 }
6270 static void test_reserve_memory_special_huge_tlbfs() {
6271 if (!UseHugeTLBFS) {
6272 return;
6273 }
6275 test_reserve_memory_special_huge_tlbfs_only();
6276 test_reserve_memory_special_huge_tlbfs_mixed();
6277 }
6279 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6280 if (!UseSHM) {
6281 return;
6282 }
6284 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6286 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6288 if (addr != NULL) {
6289 assert(is_ptr_aligned(addr, alignment), "Check");
6290 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6292 small_page_write(addr, size);
6294 os::Linux::release_memory_special_shm(addr, size);
6295 }
6296 }
6298 static void test_reserve_memory_special_shm() {
6299 size_t lp = os::large_page_size();
6300 size_t ag = os::vm_allocation_granularity();
6302 for (size_t size = ag; size < lp * 3; size += ag) {
6303 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6304 test_reserve_memory_special_shm(size, alignment);
6305 }
6306 }
6307 }
6309 static void test() {
6310 test_reserve_memory_special_huge_tlbfs();
6311 test_reserve_memory_special_shm();
6312 }
6313 };
6315 void TestReserveMemorySpecial_test() {
6316 TestReserveMemorySpecial::test();
6317 }
6319 #endif