Wed, 27 Apr 2016 01:25:04 +0800
Initial load
http://hg.openjdk.java.net/jdk8u/jdk8u/hotspot/
changeset: 6782:28b50d07f6f8
tag: jdk8u25-b17
1 /*
2 * Copyright (c) 1999, 2014, 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 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
106 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
107 // getrusage() is prepared to handle the associated failure.
108 #ifndef RUSAGE_THREAD
109 #define RUSAGE_THREAD (1) /* only the calling thread */
110 #endif
112 #define MAX_PATH (2 * K)
114 #define MAX_SECS 100000000
116 // for timer info max values which include all bits
117 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
119 #define LARGEPAGES_BIT (1 << 6)
120 ////////////////////////////////////////////////////////////////////////////////
121 // global variables
122 julong os::Linux::_physical_memory = 0;
124 address os::Linux::_initial_thread_stack_bottom = NULL;
125 uintptr_t os::Linux::_initial_thread_stack_size = 0;
127 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
128 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
129 Mutex* os::Linux::_createThread_lock = NULL;
130 pthread_t os::Linux::_main_thread;
131 int os::Linux::_page_size = -1;
132 const int os::Linux::_vm_default_page_size = (8 * K);
133 bool os::Linux::_is_floating_stack = false;
134 bool os::Linux::_is_NPTL = false;
135 bool os::Linux::_supports_fast_thread_cpu_time = false;
136 const char * os::Linux::_glibc_version = NULL;
137 const char * os::Linux::_libpthread_version = NULL;
138 pthread_condattr_t os::Linux::_condattr[1];
140 static jlong initial_time_count=0;
142 static int clock_tics_per_sec = 100;
144 // For diagnostics to print a message once. see run_periodic_checks
145 static sigset_t check_signal_done;
146 static bool check_signals = true;
148 static pid_t _initial_pid = 0;
150 /* Signal number used to suspend/resume a thread */
152 /* do not use any signal number less than SIGSEGV, see 4355769 */
153 static int SR_signum = SIGUSR2;
154 sigset_t SR_sigset;
156 /* Used to protect dlsym() calls */
157 static pthread_mutex_t dl_mutex;
159 // Declarations
160 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
162 #ifdef JAVASE_EMBEDDED
163 class MemNotifyThread: public Thread {
164 friend class VMStructs;
165 public:
166 virtual void run();
168 private:
169 static MemNotifyThread* _memnotify_thread;
170 int _fd;
172 public:
174 // Constructor
175 MemNotifyThread(int fd);
177 // Tester
178 bool is_memnotify_thread() const { return true; }
180 // Printing
181 char* name() const { return (char*)"Linux MemNotify Thread"; }
183 // Returns the single instance of the MemNotifyThread
184 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
186 // Create and start the single instance of MemNotifyThread
187 static void start();
188 };
189 #endif // JAVASE_EMBEDDED
191 // utility functions
193 static int SR_initialize();
195 julong os::available_memory() {
196 return Linux::available_memory();
197 }
199 julong os::Linux::available_memory() {
200 // values in struct sysinfo are "unsigned long"
201 struct sysinfo si;
202 sysinfo(&si);
204 return (julong)si.freeram * si.mem_unit;
205 }
207 julong os::physical_memory() {
208 return Linux::physical_memory();
209 }
211 ////////////////////////////////////////////////////////////////////////////////
212 // environment support
214 bool os::getenv(const char* name, char* buf, int len) {
215 const char* val = ::getenv(name);
216 if (val != NULL && strlen(val) < (size_t)len) {
217 strcpy(buf, val);
218 return true;
219 }
220 if (len > 0) buf[0] = 0; // return a null string
221 return false;
222 }
225 // Return true if user is running as root.
227 bool os::have_special_privileges() {
228 static bool init = false;
229 static bool privileges = false;
230 if (!init) {
231 privileges = (getuid() != geteuid()) || (getgid() != getegid());
232 init = true;
233 }
234 return privileges;
235 }
238 #ifndef SYS_gettid
239 // i386: 224, ia64: 1105, amd64: 186, sparc 143
240 #ifdef __ia64__
241 #define SYS_gettid 1105
242 #elif __i386__
243 #define SYS_gettid 224
244 #elif __amd64__
245 #define SYS_gettid 186
246 #elif __sparc__
247 #define SYS_gettid 143
248 #else
249 #error define gettid for the arch
250 #endif
251 #endif
253 // Cpu architecture string
254 #if defined(ZERO)
255 static char cpu_arch[] = ZERO_LIBARCH;
256 #elif defined(IA64)
257 static char cpu_arch[] = "ia64";
258 #elif defined(IA32)
259 static char cpu_arch[] = "i386";
260 #elif defined(AMD64)
261 static char cpu_arch[] = "amd64";
262 #elif defined(ARM)
263 static char cpu_arch[] = "arm";
264 #elif defined(PPC32)
265 static char cpu_arch[] = "ppc";
266 #elif defined(PPC64)
267 static char cpu_arch[] = "ppc64";
268 #elif defined(SPARC)
269 # ifdef _LP64
270 static char cpu_arch[] = "sparcv9";
271 # else
272 static char cpu_arch[] = "sparc";
273 # endif
274 #else
275 #error Add appropriate cpu_arch setting
276 #endif
279 // pid_t gettid()
280 //
281 // Returns the kernel thread id of the currently running thread. Kernel
282 // thread id is used to access /proc.
283 //
284 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
285 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
286 //
287 pid_t os::Linux::gettid() {
288 int rslt = syscall(SYS_gettid);
289 if (rslt == -1) {
290 // old kernel, no NPTL support
291 return getpid();
292 } else {
293 return (pid_t)rslt;
294 }
295 }
297 // Most versions of linux have a bug where the number of processors are
298 // determined by looking at the /proc file system. In a chroot environment,
299 // the system call returns 1. This causes the VM to act as if it is
300 // a single processor and elide locking (see is_MP() call).
301 static bool unsafe_chroot_detected = false;
302 static const char *unstable_chroot_error = "/proc file system not found.\n"
303 "Java may be unstable running multithreaded in a chroot "
304 "environment on Linux when /proc filesystem is not mounted.";
306 void os::Linux::initialize_system_info() {
307 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
308 if (processor_count() == 1) {
309 pid_t pid = os::Linux::gettid();
310 char fname[32];
311 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
312 FILE *fp = fopen(fname, "r");
313 if (fp == NULL) {
314 unsafe_chroot_detected = true;
315 } else {
316 fclose(fp);
317 }
318 }
319 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
320 assert(processor_count() > 0, "linux error");
321 }
323 void os::init_system_properties_values() {
324 // The next steps are taken in the product version:
325 //
326 // Obtain the JAVA_HOME value from the location of libjvm.so.
327 // This library should be located at:
328 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
329 //
330 // If "/jre/lib/" appears at the right place in the path, then we
331 // assume libjvm.so is installed in a JDK and we use this path.
332 //
333 // Otherwise exit with message: "Could not create the Java virtual machine."
334 //
335 // The following extra steps are taken in the debugging version:
336 //
337 // If "/jre/lib/" does NOT appear at the right place in the path
338 // instead of exit check for $JAVA_HOME environment variable.
339 //
340 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
341 // then we append a fake suffix "hotspot/libjvm.so" to this path so
342 // it looks like libjvm.so is installed there
343 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
344 //
345 // Otherwise exit.
346 //
347 // Important note: if the location of libjvm.so changes this
348 // code needs to be changed accordingly.
350 // See ld(1):
351 // The linker uses the following search paths to locate required
352 // shared libraries:
353 // 1: ...
354 // ...
355 // 7: The default directories, normally /lib and /usr/lib.
356 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
357 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
358 #else
359 #define DEFAULT_LIBPATH "/lib:/usr/lib"
360 #endif
362 // Base path of extensions installed on the system.
363 #define SYS_EXT_DIR "/usr/java/packages"
364 #define EXTENSIONS_DIR "/lib/ext"
365 #define ENDORSED_DIR "/lib/endorsed"
367 // Buffer that fits several sprintfs.
368 // Note that the space for the colon and the trailing null are provided
369 // by the nulls included by the sizeof operator.
370 const size_t bufsize =
371 MAX3((size_t)MAXPATHLEN, // For dll_dir & friends.
372 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
373 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
374 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
376 // sysclasspath, java_home, dll_dir
377 {
378 char *pslash;
379 os::jvm_path(buf, bufsize);
381 // Found the full path to libjvm.so.
382 // Now cut the path to <java_home>/jre if we can.
383 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
384 pslash = strrchr(buf, '/');
385 if (pslash != NULL) {
386 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
387 }
388 Arguments::set_dll_dir(buf);
390 if (pslash != NULL) {
391 pslash = strrchr(buf, '/');
392 if (pslash != NULL) {
393 *pslash = '\0'; // Get rid of /<arch>.
394 pslash = strrchr(buf, '/');
395 if (pslash != NULL) {
396 *pslash = '\0'; // Get rid of /lib.
397 }
398 }
399 }
400 Arguments::set_java_home(buf);
401 set_boot_path('/', ':');
402 }
404 // Where to look for native libraries.
405 //
406 // Note: Due to a legacy implementation, most of the library path
407 // is set in the launcher. This was to accomodate linking restrictions
408 // on legacy Linux implementations (which are no longer supported).
409 // Eventually, all the library path setting will be done here.
410 //
411 // However, to prevent the proliferation of improperly built native
412 // libraries, the new path component /usr/java/packages is added here.
413 // Eventually, all the library path setting will be done here.
414 {
415 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
416 // should always exist (until the legacy problem cited above is
417 // addressed).
418 const char *v = ::getenv("LD_LIBRARY_PATH");
419 const char *v_colon = ":";
420 if (v == NULL) { v = ""; v_colon = ""; }
421 // That's +1 for the colon and +1 for the trailing '\0'.
422 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
423 strlen(v) + 1 +
424 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
425 mtInternal);
426 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
427 Arguments::set_library_path(ld_library_path);
428 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
429 }
431 // Extensions directories.
432 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
433 Arguments::set_ext_dirs(buf);
435 // Endorsed standards default directory.
436 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
437 Arguments::set_endorsed_dirs(buf);
439 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
441 #undef DEFAULT_LIBPATH
442 #undef SYS_EXT_DIR
443 #undef EXTENSIONS_DIR
444 #undef ENDORSED_DIR
445 }
447 ////////////////////////////////////////////////////////////////////////////////
448 // breakpoint support
450 void os::breakpoint() {
451 BREAKPOINT;
452 }
454 extern "C" void breakpoint() {
455 // use debugger to set breakpoint here
456 }
458 ////////////////////////////////////////////////////////////////////////////////
459 // signal support
461 debug_only(static bool signal_sets_initialized = false);
462 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
464 bool os::Linux::is_sig_ignored(int sig) {
465 struct sigaction oact;
466 sigaction(sig, (struct sigaction*)NULL, &oact);
467 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
468 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
469 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
470 return true;
471 else
472 return false;
473 }
475 void os::Linux::signal_sets_init() {
476 // Should also have an assertion stating we are still single-threaded.
477 assert(!signal_sets_initialized, "Already initialized");
478 // Fill in signals that are necessarily unblocked for all threads in
479 // the VM. Currently, we unblock the following signals:
480 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
481 // by -Xrs (=ReduceSignalUsage));
482 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
483 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
484 // the dispositions or masks wrt these signals.
485 // Programs embedding the VM that want to use the above signals for their
486 // own purposes must, at this time, use the "-Xrs" option to prevent
487 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
488 // (See bug 4345157, and other related bugs).
489 // In reality, though, unblocking these signals is really a nop, since
490 // these signals are not blocked by default.
491 sigemptyset(&unblocked_sigs);
492 sigemptyset(&allowdebug_blocked_sigs);
493 sigaddset(&unblocked_sigs, SIGILL);
494 sigaddset(&unblocked_sigs, SIGSEGV);
495 sigaddset(&unblocked_sigs, SIGBUS);
496 sigaddset(&unblocked_sigs, SIGFPE);
497 #if defined(PPC64)
498 sigaddset(&unblocked_sigs, SIGTRAP);
499 #endif
500 sigaddset(&unblocked_sigs, SR_signum);
502 if (!ReduceSignalUsage) {
503 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
504 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
505 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
506 }
507 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
508 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
509 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
510 }
511 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
512 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
513 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
514 }
515 }
516 // Fill in signals that are blocked by all but the VM thread.
517 sigemptyset(&vm_sigs);
518 if (!ReduceSignalUsage)
519 sigaddset(&vm_sigs, BREAK_SIGNAL);
520 debug_only(signal_sets_initialized = true);
522 }
524 // These are signals that are unblocked while a thread is running Java.
525 // (For some reason, they get blocked by default.)
526 sigset_t* os::Linux::unblocked_signals() {
527 assert(signal_sets_initialized, "Not initialized");
528 return &unblocked_sigs;
529 }
531 // These are the signals that are blocked while a (non-VM) thread is
532 // running Java. Only the VM thread handles these signals.
533 sigset_t* os::Linux::vm_signals() {
534 assert(signal_sets_initialized, "Not initialized");
535 return &vm_sigs;
536 }
538 // These are signals that are blocked during cond_wait to allow debugger in
539 sigset_t* os::Linux::allowdebug_blocked_signals() {
540 assert(signal_sets_initialized, "Not initialized");
541 return &allowdebug_blocked_sigs;
542 }
544 void os::Linux::hotspot_sigmask(Thread* thread) {
546 //Save caller's signal mask before setting VM signal mask
547 sigset_t caller_sigmask;
548 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
550 OSThread* osthread = thread->osthread();
551 osthread->set_caller_sigmask(caller_sigmask);
553 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
555 if (!ReduceSignalUsage) {
556 if (thread->is_VM_thread()) {
557 // Only the VM thread handles BREAK_SIGNAL ...
558 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
559 } else {
560 // ... all other threads block BREAK_SIGNAL
561 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
562 }
563 }
564 }
566 //////////////////////////////////////////////////////////////////////////////
567 // detecting pthread library
569 void os::Linux::libpthread_init() {
570 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
571 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
572 // generic name for earlier versions.
573 // Define macros here so we can build HotSpot on old systems.
574 # ifndef _CS_GNU_LIBC_VERSION
575 # define _CS_GNU_LIBC_VERSION 2
576 # endif
577 # ifndef _CS_GNU_LIBPTHREAD_VERSION
578 # define _CS_GNU_LIBPTHREAD_VERSION 3
579 # endif
581 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
582 if (n > 0) {
583 char *str = (char *)malloc(n, mtInternal);
584 confstr(_CS_GNU_LIBC_VERSION, str, n);
585 os::Linux::set_glibc_version(str);
586 } else {
587 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
588 static char _gnu_libc_version[32];
589 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
590 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
591 os::Linux::set_glibc_version(_gnu_libc_version);
592 }
594 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
595 if (n > 0) {
596 char *str = (char *)malloc(n, mtInternal);
597 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
598 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
599 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
600 // is the case. LinuxThreads has a hard limit on max number of threads.
601 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
602 // On the other hand, NPTL does not have such a limit, sysconf()
603 // will return -1 and errno is not changed. Check if it is really NPTL.
604 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
605 strstr(str, "NPTL") &&
606 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
607 free(str);
608 os::Linux::set_libpthread_version("linuxthreads");
609 } else {
610 os::Linux::set_libpthread_version(str);
611 }
612 } else {
613 // glibc before 2.3.2 only has LinuxThreads.
614 os::Linux::set_libpthread_version("linuxthreads");
615 }
617 if (strstr(libpthread_version(), "NPTL")) {
618 os::Linux::set_is_NPTL();
619 } else {
620 os::Linux::set_is_LinuxThreads();
621 }
623 // LinuxThreads have two flavors: floating-stack mode, which allows variable
624 // stack size; and fixed-stack mode. NPTL is always floating-stack.
625 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
626 os::Linux::set_is_floating_stack();
627 }
628 }
630 /////////////////////////////////////////////////////////////////////////////
631 // thread stack
633 // Force Linux kernel to expand current thread stack. If "bottom" is close
634 // to the stack guard, caller should block all signals.
635 //
636 // MAP_GROWSDOWN:
637 // A special mmap() flag that is used to implement thread stacks. It tells
638 // kernel that the memory region should extend downwards when needed. This
639 // allows early versions of LinuxThreads to only mmap the first few pages
640 // when creating a new thread. Linux kernel will automatically expand thread
641 // stack as needed (on page faults).
642 //
643 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
644 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
645 // region, it's hard to tell if the fault is due to a legitimate stack
646 // access or because of reading/writing non-exist memory (e.g. buffer
647 // overrun). As a rule, if the fault happens below current stack pointer,
648 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
649 // application (see Linux kernel fault.c).
650 //
651 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
652 // stack overflow detection.
653 //
654 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
655 // not use this flag. However, the stack of initial thread is not created
656 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
657 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
658 // and then attach the thread to JVM.
659 //
660 // To get around the problem and allow stack banging on Linux, we need to
661 // manually expand thread stack after receiving the SIGSEGV.
662 //
663 // There are two ways to expand thread stack to address "bottom", we used
664 // both of them in JVM before 1.5:
665 // 1. adjust stack pointer first so that it is below "bottom", and then
666 // touch "bottom"
667 // 2. mmap() the page in question
668 //
669 // Now alternate signal stack is gone, it's harder to use 2. For instance,
670 // if current sp is already near the lower end of page 101, and we need to
671 // call mmap() to map page 100, it is possible that part of the mmap() frame
672 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
673 // That will destroy the mmap() frame and cause VM to crash.
674 //
675 // The following code works by adjusting sp first, then accessing the "bottom"
676 // page to force a page fault. Linux kernel will then automatically expand the
677 // stack mapping.
678 //
679 // _expand_stack_to() assumes its frame size is less than page size, which
680 // should always be true if the function is not inlined.
682 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
683 #define NOINLINE
684 #else
685 #define NOINLINE __attribute__ ((noinline))
686 #endif
688 static void _expand_stack_to(address bottom) NOINLINE;
690 static void _expand_stack_to(address bottom) {
691 address sp;
692 size_t size;
693 volatile char *p;
695 // Adjust bottom to point to the largest address within the same page, it
696 // gives us a one-page buffer if alloca() allocates slightly more memory.
697 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
698 bottom += os::Linux::page_size() - 1;
700 // sp might be slightly above current stack pointer; if that's the case, we
701 // will alloca() a little more space than necessary, which is OK. Don't use
702 // os::current_stack_pointer(), as its result can be slightly below current
703 // stack pointer, causing us to not alloca enough to reach "bottom".
704 sp = (address)&sp;
706 if (sp > bottom) {
707 size = sp - bottom;
708 p = (volatile char *)alloca(size);
709 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
710 p[0] = '\0';
711 }
712 }
714 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
715 assert(t!=NULL, "just checking");
716 assert(t->osthread()->expanding_stack(), "expand should be set");
717 assert(t->stack_base() != NULL, "stack_base was not initialized");
719 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
720 sigset_t mask_all, old_sigset;
721 sigfillset(&mask_all);
722 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
723 _expand_stack_to(addr);
724 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
725 return true;
726 }
727 return false;
728 }
730 //////////////////////////////////////////////////////////////////////////////
731 // create new thread
733 static address highest_vm_reserved_address();
735 // check if it's safe to start a new thread
736 static bool _thread_safety_check(Thread* thread) {
737 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
738 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
739 // Heap is mmap'ed at lower end of memory space. Thread stacks are
740 // allocated (MAP_FIXED) from high address space. Every thread stack
741 // occupies a fixed size slot (usually 2Mbytes, but user can change
742 // it to other values if they rebuild LinuxThreads).
743 //
744 // Problem with MAP_FIXED is that mmap() can still succeed even part of
745 // the memory region has already been mmap'ed. That means if we have too
746 // many threads and/or very large heap, eventually thread stack will
747 // collide with heap.
748 //
749 // Here we try to prevent heap/stack collision by comparing current
750 // stack bottom with the highest address that has been mmap'ed by JVM
751 // plus a safety margin for memory maps created by native code.
752 //
753 // This feature can be disabled by setting ThreadSafetyMargin to 0
754 //
755 if (ThreadSafetyMargin > 0) {
756 address stack_bottom = os::current_stack_base() - os::current_stack_size();
758 // not safe if our stack extends below the safety margin
759 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
760 } else {
761 return true;
762 }
763 } else {
764 // Floating stack LinuxThreads or NPTL:
765 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
766 // there's not enough space left, pthread_create() will fail. If we come
767 // here, that means enough space has been reserved for stack.
768 return true;
769 }
770 }
772 // Thread start routine for all newly created threads
773 static void *java_start(Thread *thread) {
774 // Try to randomize the cache line index of hot stack frames.
775 // This helps when threads of the same stack traces evict each other's
776 // cache lines. The threads can be either from the same JVM instance, or
777 // from different JVM instances. The benefit is especially true for
778 // processors with hyperthreading technology.
779 static int counter = 0;
780 int pid = os::current_process_id();
781 alloca(((pid ^ counter++) & 7) * 128);
783 ThreadLocalStorage::set_thread(thread);
785 OSThread* osthread = thread->osthread();
786 Monitor* sync = osthread->startThread_lock();
788 // non floating stack LinuxThreads needs extra check, see above
789 if (!_thread_safety_check(thread)) {
790 // notify parent thread
791 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
792 osthread->set_state(ZOMBIE);
793 sync->notify_all();
794 return NULL;
795 }
797 // thread_id is kernel thread id (similar to Solaris LWP id)
798 osthread->set_thread_id(os::Linux::gettid());
800 if (UseNUMA) {
801 int lgrp_id = os::numa_get_group_id();
802 if (lgrp_id != -1) {
803 thread->set_lgrp_id(lgrp_id);
804 }
805 }
806 // initialize signal mask for this thread
807 os::Linux::hotspot_sigmask(thread);
809 // initialize floating point control register
810 os::Linux::init_thread_fpu_state();
812 // handshaking with parent thread
813 {
814 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
816 // notify parent thread
817 osthread->set_state(INITIALIZED);
818 sync->notify_all();
820 // wait until os::start_thread()
821 while (osthread->get_state() == INITIALIZED) {
822 sync->wait(Mutex::_no_safepoint_check_flag);
823 }
824 }
826 // call one more level start routine
827 thread->run();
829 return 0;
830 }
832 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
833 assert(thread->osthread() == NULL, "caller responsible");
835 // Allocate the OSThread object
836 OSThread* osthread = new OSThread(NULL, NULL);
837 if (osthread == NULL) {
838 return false;
839 }
841 // set the correct thread state
842 osthread->set_thread_type(thr_type);
844 // Initial state is ALLOCATED but not INITIALIZED
845 osthread->set_state(ALLOCATED);
847 thread->set_osthread(osthread);
849 // init thread attributes
850 pthread_attr_t attr;
851 pthread_attr_init(&attr);
852 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
854 // stack size
855 if (os::Linux::supports_variable_stack_size()) {
856 // calculate stack size if it's not specified by caller
857 if (stack_size == 0) {
858 stack_size = os::Linux::default_stack_size(thr_type);
860 switch (thr_type) {
861 case os::java_thread:
862 // Java threads use ThreadStackSize which default value can be
863 // changed with the flag -Xss
864 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
865 stack_size = JavaThread::stack_size_at_create();
866 break;
867 case os::compiler_thread:
868 if (CompilerThreadStackSize > 0) {
869 stack_size = (size_t)(CompilerThreadStackSize * K);
870 break;
871 } // else fall through:
872 // use VMThreadStackSize if CompilerThreadStackSize is not defined
873 case os::vm_thread:
874 case os::pgc_thread:
875 case os::cgc_thread:
876 case os::watcher_thread:
877 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
878 break;
879 }
880 }
882 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
883 pthread_attr_setstacksize(&attr, stack_size);
884 } else {
885 // let pthread_create() pick the default value.
886 }
888 // glibc guard page
889 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
891 ThreadState state;
893 {
894 // Serialize thread creation if we are running with fixed stack LinuxThreads
895 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
896 if (lock) {
897 os::Linux::createThread_lock()->lock_without_safepoint_check();
898 }
900 pthread_t tid;
901 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
903 pthread_attr_destroy(&attr);
905 if (ret != 0) {
906 if (PrintMiscellaneous && (Verbose || WizardMode)) {
907 perror("pthread_create()");
908 }
909 // Need to clean up stuff we've allocated so far
910 thread->set_osthread(NULL);
911 delete osthread;
912 if (lock) os::Linux::createThread_lock()->unlock();
913 return false;
914 }
916 // Store pthread info into the OSThread
917 osthread->set_pthread_id(tid);
919 // Wait until child thread is either initialized or aborted
920 {
921 Monitor* sync_with_child = osthread->startThread_lock();
922 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
923 while ((state = osthread->get_state()) == ALLOCATED) {
924 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
925 }
926 }
928 if (lock) {
929 os::Linux::createThread_lock()->unlock();
930 }
931 }
933 // Aborted due to thread limit being reached
934 if (state == ZOMBIE) {
935 thread->set_osthread(NULL);
936 delete osthread;
937 return false;
938 }
940 // The thread is returned suspended (in state INITIALIZED),
941 // and is started higher up in the call chain
942 assert(state == INITIALIZED, "race condition");
943 return true;
944 }
946 /////////////////////////////////////////////////////////////////////////////
947 // attach existing thread
949 // bootstrap the main thread
950 bool os::create_main_thread(JavaThread* thread) {
951 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
952 return create_attached_thread(thread);
953 }
955 bool os::create_attached_thread(JavaThread* thread) {
956 #ifdef ASSERT
957 thread->verify_not_published();
958 #endif
960 // Allocate the OSThread object
961 OSThread* osthread = new OSThread(NULL, NULL);
963 if (osthread == NULL) {
964 return false;
965 }
967 // Store pthread info into the OSThread
968 osthread->set_thread_id(os::Linux::gettid());
969 osthread->set_pthread_id(::pthread_self());
971 // initialize floating point control register
972 os::Linux::init_thread_fpu_state();
974 // Initial thread state is RUNNABLE
975 osthread->set_state(RUNNABLE);
977 thread->set_osthread(osthread);
979 if (UseNUMA) {
980 int lgrp_id = os::numa_get_group_id();
981 if (lgrp_id != -1) {
982 thread->set_lgrp_id(lgrp_id);
983 }
984 }
986 if (os::Linux::is_initial_thread()) {
987 // If current thread is initial thread, its stack is mapped on demand,
988 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
989 // the entire stack region to avoid SEGV in stack banging.
990 // It is also useful to get around the heap-stack-gap problem on SuSE
991 // kernel (see 4821821 for details). We first expand stack to the top
992 // of yellow zone, then enable stack yellow zone (order is significant,
993 // enabling yellow zone first will crash JVM on SuSE Linux), so there
994 // is no gap between the last two virtual memory regions.
996 JavaThread *jt = (JavaThread *)thread;
997 address addr = jt->stack_yellow_zone_base();
998 assert(addr != NULL, "initialization problem?");
999 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1001 osthread->set_expanding_stack();
1002 os::Linux::manually_expand_stack(jt, addr);
1003 osthread->clear_expanding_stack();
1004 }
1006 // initialize signal mask for this thread
1007 // and save the caller's signal mask
1008 os::Linux::hotspot_sigmask(thread);
1010 return true;
1011 }
1013 void os::pd_start_thread(Thread* thread) {
1014 OSThread * osthread = thread->osthread();
1015 assert(osthread->get_state() != INITIALIZED, "just checking");
1016 Monitor* sync_with_child = osthread->startThread_lock();
1017 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1018 sync_with_child->notify();
1019 }
1021 // Free Linux resources related to the OSThread
1022 void os::free_thread(OSThread* osthread) {
1023 assert(osthread != NULL, "osthread not set");
1025 if (Thread::current()->osthread() == osthread) {
1026 // Restore caller's signal mask
1027 sigset_t sigmask = osthread->caller_sigmask();
1028 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1029 }
1031 delete osthread;
1032 }
1034 //////////////////////////////////////////////////////////////////////////////
1035 // thread local storage
1037 // Restore the thread pointer if the destructor is called. This is in case
1038 // someone from JNI code sets up a destructor with pthread_key_create to run
1039 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1040 // will hang or crash. When detachCurrentThread is called the key will be set
1041 // to null and we will not be called again. If detachCurrentThread is never
1042 // called we could loop forever depending on the pthread implementation.
1043 static void restore_thread_pointer(void* p) {
1044 Thread* thread = (Thread*) p;
1045 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1046 }
1048 int os::allocate_thread_local_storage() {
1049 pthread_key_t key;
1050 int rslt = pthread_key_create(&key, restore_thread_pointer);
1051 assert(rslt == 0, "cannot allocate thread local storage");
1052 return (int)key;
1053 }
1055 // Note: This is currently not used by VM, as we don't destroy TLS key
1056 // on VM exit.
1057 void os::free_thread_local_storage(int index) {
1058 int rslt = pthread_key_delete((pthread_key_t)index);
1059 assert(rslt == 0, "invalid index");
1060 }
1062 void os::thread_local_storage_at_put(int index, void* value) {
1063 int rslt = pthread_setspecific((pthread_key_t)index, value);
1064 assert(rslt == 0, "pthread_setspecific failed");
1065 }
1067 extern "C" Thread* get_thread() {
1068 return ThreadLocalStorage::thread();
1069 }
1071 //////////////////////////////////////////////////////////////////////////////
1072 // initial thread
1074 // Check if current thread is the initial thread, similar to Solaris thr_main.
1075 bool os::Linux::is_initial_thread(void) {
1076 char dummy;
1077 // If called before init complete, thread stack bottom will be null.
1078 // Can be called if fatal error occurs before initialization.
1079 if (initial_thread_stack_bottom() == NULL) return false;
1080 assert(initial_thread_stack_bottom() != NULL &&
1081 initial_thread_stack_size() != 0,
1082 "os::init did not locate initial thread's stack region");
1083 if ((address)&dummy >= initial_thread_stack_bottom() &&
1084 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1085 return true;
1086 else return false;
1087 }
1089 // Find the virtual memory area that contains addr
1090 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1091 FILE *fp = fopen("/proc/self/maps", "r");
1092 if (fp) {
1093 address low, high;
1094 while (!feof(fp)) {
1095 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1096 if (low <= addr && addr < high) {
1097 if (vma_low) *vma_low = low;
1098 if (vma_high) *vma_high = high;
1099 fclose (fp);
1100 return true;
1101 }
1102 }
1103 for (;;) {
1104 int ch = fgetc(fp);
1105 if (ch == EOF || ch == (int)'\n') break;
1106 }
1107 }
1108 fclose(fp);
1109 }
1110 return false;
1111 }
1113 // Locate initial thread stack. This special handling of initial thread stack
1114 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1115 // bogus value for initial thread.
1116 void os::Linux::capture_initial_stack(size_t max_size) {
1117 // stack size is the easy part, get it from RLIMIT_STACK
1118 size_t stack_size;
1119 struct rlimit rlim;
1120 getrlimit(RLIMIT_STACK, &rlim);
1121 stack_size = rlim.rlim_cur;
1123 // 6308388: a bug in ld.so will relocate its own .data section to the
1124 // lower end of primordial stack; reduce ulimit -s value a little bit
1125 // so we won't install guard page on ld.so's data section.
1126 stack_size -= 2 * page_size();
1128 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1129 // 7.1, in both cases we will get 2G in return value.
1130 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1131 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1132 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1133 // in case other parts in glibc still assumes 2M max stack size.
1134 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1135 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1136 if (stack_size > 2 * K * K IA64_ONLY(*2))
1137 stack_size = 2 * K * K IA64_ONLY(*2);
1138 // Try to figure out where the stack base (top) is. This is harder.
1139 //
1140 // When an application is started, glibc saves the initial stack pointer in
1141 // a global variable "__libc_stack_end", which is then used by system
1142 // libraries. __libc_stack_end should be pretty close to stack top. The
1143 // variable is available since the very early days. However, because it is
1144 // a private interface, it could disappear in the future.
1145 //
1146 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1147 // to __libc_stack_end, it is very close to stack top, but isn't the real
1148 // stack top. Note that /proc may not exist if VM is running as a chroot
1149 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1150 // /proc/<pid>/stat could change in the future (though unlikely).
1151 //
1152 // We try __libc_stack_end first. If that doesn't work, look for
1153 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1154 // as a hint, which should work well in most cases.
1156 uintptr_t stack_start;
1158 // try __libc_stack_end first
1159 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1160 if (p && *p) {
1161 stack_start = *p;
1162 } else {
1163 // see if we can get the start_stack field from /proc/self/stat
1164 FILE *fp;
1165 int pid;
1166 char state;
1167 int ppid;
1168 int pgrp;
1169 int session;
1170 int nr;
1171 int tpgrp;
1172 unsigned long flags;
1173 unsigned long minflt;
1174 unsigned long cminflt;
1175 unsigned long majflt;
1176 unsigned long cmajflt;
1177 unsigned long utime;
1178 unsigned long stime;
1179 long cutime;
1180 long cstime;
1181 long prio;
1182 long nice;
1183 long junk;
1184 long it_real;
1185 uintptr_t start;
1186 uintptr_t vsize;
1187 intptr_t rss;
1188 uintptr_t rsslim;
1189 uintptr_t scodes;
1190 uintptr_t ecode;
1191 int i;
1193 // Figure what the primordial thread stack base is. Code is inspired
1194 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1195 // followed by command name surrounded by parentheses, state, etc.
1196 char stat[2048];
1197 int statlen;
1199 fp = fopen("/proc/self/stat", "r");
1200 if (fp) {
1201 statlen = fread(stat, 1, 2047, fp);
1202 stat[statlen] = '\0';
1203 fclose(fp);
1205 // Skip pid and the command string. Note that we could be dealing with
1206 // weird command names, e.g. user could decide to rename java launcher
1207 // to "java 1.4.2 :)", then the stat file would look like
1208 // 1234 (java 1.4.2 :)) R ... ...
1209 // We don't really need to know the command string, just find the last
1210 // occurrence of ")" and then start parsing from there. See bug 4726580.
1211 char * s = strrchr(stat, ')');
1213 i = 0;
1214 if (s) {
1215 // Skip blank chars
1216 do s++; while (isspace(*s));
1218 #define _UFM UINTX_FORMAT
1219 #define _DFM INTX_FORMAT
1221 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1222 /* 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 */
1223 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,
1224 &state, /* 3 %c */
1225 &ppid, /* 4 %d */
1226 &pgrp, /* 5 %d */
1227 &session, /* 6 %d */
1228 &nr, /* 7 %d */
1229 &tpgrp, /* 8 %d */
1230 &flags, /* 9 %lu */
1231 &minflt, /* 10 %lu */
1232 &cminflt, /* 11 %lu */
1233 &majflt, /* 12 %lu */
1234 &cmajflt, /* 13 %lu */
1235 &utime, /* 14 %lu */
1236 &stime, /* 15 %lu */
1237 &cutime, /* 16 %ld */
1238 &cstime, /* 17 %ld */
1239 &prio, /* 18 %ld */
1240 &nice, /* 19 %ld */
1241 &junk, /* 20 %ld */
1242 &it_real, /* 21 %ld */
1243 &start, /* 22 UINTX_FORMAT */
1244 &vsize, /* 23 UINTX_FORMAT */
1245 &rss, /* 24 INTX_FORMAT */
1246 &rsslim, /* 25 UINTX_FORMAT */
1247 &scodes, /* 26 UINTX_FORMAT */
1248 &ecode, /* 27 UINTX_FORMAT */
1249 &stack_start); /* 28 UINTX_FORMAT */
1250 }
1252 #undef _UFM
1253 #undef _DFM
1255 if (i != 28 - 2) {
1256 assert(false, "Bad conversion from /proc/self/stat");
1257 // product mode - assume we are the initial thread, good luck in the
1258 // embedded case.
1259 warning("Can't detect initial thread stack location - bad conversion");
1260 stack_start = (uintptr_t) &rlim;
1261 }
1262 } else {
1263 // For some reason we can't open /proc/self/stat (for example, running on
1264 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1265 // most cases, so don't abort:
1266 warning("Can't detect initial thread stack location - no /proc/self/stat");
1267 stack_start = (uintptr_t) &rlim;
1268 }
1269 }
1271 // Now we have a pointer (stack_start) very close to the stack top, the
1272 // next thing to do is to figure out the exact location of stack top. We
1273 // can find out the virtual memory area that contains stack_start by
1274 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1275 // and its upper limit is the real stack top. (again, this would fail if
1276 // running inside chroot, because /proc may not exist.)
1278 uintptr_t stack_top;
1279 address low, high;
1280 if (find_vma((address)stack_start, &low, &high)) {
1281 // success, "high" is the true stack top. (ignore "low", because initial
1282 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1283 stack_top = (uintptr_t)high;
1284 } else {
1285 // failed, likely because /proc/self/maps does not exist
1286 warning("Can't detect initial thread stack location - find_vma failed");
1287 // best effort: stack_start is normally within a few pages below the real
1288 // stack top, use it as stack top, and reduce stack size so we won't put
1289 // guard page outside stack.
1290 stack_top = stack_start;
1291 stack_size -= 16 * page_size();
1292 }
1294 // stack_top could be partially down the page so align it
1295 stack_top = align_size_up(stack_top, page_size());
1297 if (max_size && stack_size > max_size) {
1298 _initial_thread_stack_size = max_size;
1299 } else {
1300 _initial_thread_stack_size = stack_size;
1301 }
1303 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1304 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1305 }
1307 ////////////////////////////////////////////////////////////////////////////////
1308 // time support
1310 // Time since start-up in seconds to a fine granularity.
1311 // Used by VMSelfDestructTimer and the MemProfiler.
1312 double os::elapsedTime() {
1314 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1315 }
1317 jlong os::elapsed_counter() {
1318 return javaTimeNanos() - initial_time_count;
1319 }
1321 jlong os::elapsed_frequency() {
1322 return NANOSECS_PER_SEC; // nanosecond resolution
1323 }
1325 bool os::supports_vtime() { return true; }
1326 bool os::enable_vtime() { return false; }
1327 bool os::vtime_enabled() { return false; }
1329 double os::elapsedVTime() {
1330 struct rusage usage;
1331 int retval = getrusage(RUSAGE_THREAD, &usage);
1332 if (retval == 0) {
1333 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);
1334 } else {
1335 // better than nothing, but not much
1336 return elapsedTime();
1337 }
1338 }
1340 jlong os::javaTimeMillis() {
1341 timeval time;
1342 int status = gettimeofday(&time, NULL);
1343 assert(status != -1, "linux error");
1344 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1345 }
1347 #ifndef CLOCK_MONOTONIC
1348 #define CLOCK_MONOTONIC (1)
1349 #endif
1351 void os::Linux::clock_init() {
1352 // we do dlopen's in this particular order due to bug in linux
1353 // dynamical loader (see 6348968) leading to crash on exit
1354 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1355 if (handle == NULL) {
1356 handle = dlopen("librt.so", RTLD_LAZY);
1357 }
1359 if (handle) {
1360 int (*clock_getres_func)(clockid_t, struct timespec*) =
1361 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1362 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1363 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1364 if (clock_getres_func && clock_gettime_func) {
1365 // See if monotonic clock is supported by the kernel. Note that some
1366 // early implementations simply return kernel jiffies (updated every
1367 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1368 // for nano time (though the monotonic property is still nice to have).
1369 // It's fixed in newer kernels, however clock_getres() still returns
1370 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1371 // resolution for now. Hopefully as people move to new kernels, this
1372 // won't be a problem.
1373 struct timespec res;
1374 struct timespec tp;
1375 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1376 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1377 // yes, monotonic clock is supported
1378 _clock_gettime = clock_gettime_func;
1379 return;
1380 } else {
1381 // close librt if there is no monotonic clock
1382 dlclose(handle);
1383 }
1384 }
1385 }
1386 warning("No monotonic clock was available - timed services may " \
1387 "be adversely affected if the time-of-day clock changes");
1388 }
1390 #ifndef SYS_clock_getres
1392 #if defined(IA32) || defined(AMD64)
1393 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1394 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1395 #else
1396 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1397 #define sys_clock_getres(x,y) -1
1398 #endif
1400 #else
1401 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1402 #endif
1404 void os::Linux::fast_thread_clock_init() {
1405 if (!UseLinuxPosixThreadCPUClocks) {
1406 return;
1407 }
1408 clockid_t clockid;
1409 struct timespec tp;
1410 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1411 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1413 // Switch to using fast clocks for thread cpu time if
1414 // the sys_clock_getres() returns 0 error code.
1415 // Note, that some kernels may support the current thread
1416 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1417 // returned by the pthread_getcpuclockid().
1418 // If the fast Posix clocks are supported then the sys_clock_getres()
1419 // must return at least tp.tv_sec == 0 which means a resolution
1420 // better than 1 sec. This is extra check for reliability.
1422 if(pthread_getcpuclockid_func &&
1423 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1424 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1426 _supports_fast_thread_cpu_time = true;
1427 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1428 }
1429 }
1431 jlong os::javaTimeNanos() {
1432 if (Linux::supports_monotonic_clock()) {
1433 struct timespec tp;
1434 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1435 assert(status == 0, "gettime error");
1436 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1437 return result;
1438 } else {
1439 timeval time;
1440 int status = gettimeofday(&time, NULL);
1441 assert(status != -1, "linux error");
1442 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1443 return 1000 * usecs;
1444 }
1445 }
1447 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1448 if (Linux::supports_monotonic_clock()) {
1449 info_ptr->max_value = ALL_64_BITS;
1451 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1452 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1453 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1454 } else {
1455 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1456 info_ptr->max_value = ALL_64_BITS;
1458 // gettimeofday is a real time clock so it skips
1459 info_ptr->may_skip_backward = true;
1460 info_ptr->may_skip_forward = true;
1461 }
1463 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1464 }
1466 // Return the real, user, and system times in seconds from an
1467 // arbitrary fixed point in the past.
1468 bool os::getTimesSecs(double* process_real_time,
1469 double* process_user_time,
1470 double* process_system_time) {
1471 struct tms ticks;
1472 clock_t real_ticks = times(&ticks);
1474 if (real_ticks == (clock_t) (-1)) {
1475 return false;
1476 } else {
1477 double ticks_per_second = (double) clock_tics_per_sec;
1478 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1479 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1480 *process_real_time = ((double) real_ticks) / ticks_per_second;
1482 return true;
1483 }
1484 }
1487 char * os::local_time_string(char *buf, size_t buflen) {
1488 struct tm t;
1489 time_t long_time;
1490 time(&long_time);
1491 localtime_r(&long_time, &t);
1492 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1493 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1494 t.tm_hour, t.tm_min, t.tm_sec);
1495 return buf;
1496 }
1498 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1499 return localtime_r(clock, res);
1500 }
1502 ////////////////////////////////////////////////////////////////////////////////
1503 // runtime exit support
1505 // Note: os::shutdown() might be called very early during initialization, or
1506 // called from signal handler. Before adding something to os::shutdown(), make
1507 // sure it is async-safe and can handle partially initialized VM.
1508 void os::shutdown() {
1510 // allow PerfMemory to attempt cleanup of any persistent resources
1511 perfMemory_exit();
1513 // needs to remove object in file system
1514 AttachListener::abort();
1516 // flush buffered output, finish log files
1517 ostream_abort();
1519 // Check for abort hook
1520 abort_hook_t abort_hook = Arguments::abort_hook();
1521 if (abort_hook != NULL) {
1522 abort_hook();
1523 }
1525 }
1527 // Note: os::abort() might be called very early during initialization, or
1528 // called from signal handler. Before adding something to os::abort(), make
1529 // sure it is async-safe and can handle partially initialized VM.
1530 void os::abort(bool dump_core) {
1531 os::shutdown();
1532 if (dump_core) {
1533 #ifndef PRODUCT
1534 fdStream out(defaultStream::output_fd());
1535 out.print_raw("Current thread is ");
1536 char buf[16];
1537 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1538 out.print_raw_cr(buf);
1539 out.print_raw_cr("Dumping core ...");
1540 #endif
1541 ::abort(); // dump core
1542 }
1544 ::exit(1);
1545 }
1547 // Die immediately, no exit hook, no abort hook, no cleanup.
1548 void os::die() {
1549 // _exit() on LinuxThreads only kills current thread
1550 ::abort();
1551 }
1554 // This method is a copy of JDK's sysGetLastErrorString
1555 // from src/solaris/hpi/src/system_md.c
1557 size_t os::lasterror(char *buf, size_t len) {
1559 if (errno == 0) return 0;
1561 const char *s = ::strerror(errno);
1562 size_t n = ::strlen(s);
1563 if (n >= len) {
1564 n = len - 1;
1565 }
1566 ::strncpy(buf, s, n);
1567 buf[n] = '\0';
1568 return n;
1569 }
1571 intx os::current_thread_id() { return (intx)pthread_self(); }
1572 int os::current_process_id() {
1574 // Under the old linux thread library, linux gives each thread
1575 // its own process id. Because of this each thread will return
1576 // a different pid if this method were to return the result
1577 // of getpid(2). Linux provides no api that returns the pid
1578 // of the launcher thread for the vm. This implementation
1579 // returns a unique pid, the pid of the launcher thread
1580 // that starts the vm 'process'.
1582 // Under the NPTL, getpid() returns the same pid as the
1583 // launcher thread rather than a unique pid per thread.
1584 // Use gettid() if you want the old pre NPTL behaviour.
1586 // if you are looking for the result of a call to getpid() that
1587 // returns a unique pid for the calling thread, then look at the
1588 // OSThread::thread_id() method in osThread_linux.hpp file
1590 return (int)(_initial_pid ? _initial_pid : getpid());
1591 }
1593 // DLL functions
1595 const char* os::dll_file_extension() { return ".so"; }
1597 // This must be hard coded because it's the system's temporary
1598 // directory not the java application's temp directory, ala java.io.tmpdir.
1599 const char* os::get_temp_directory() { return "/tmp"; }
1601 static bool file_exists(const char* filename) {
1602 struct stat statbuf;
1603 if (filename == NULL || strlen(filename) == 0) {
1604 return false;
1605 }
1606 return os::stat(filename, &statbuf) == 0;
1607 }
1609 bool os::dll_build_name(char* buffer, size_t buflen,
1610 const char* pname, const char* fname) {
1611 bool retval = false;
1612 // Copied from libhpi
1613 const size_t pnamelen = pname ? strlen(pname) : 0;
1615 // Return error on buffer overflow.
1616 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1617 return retval;
1618 }
1620 if (pnamelen == 0) {
1621 snprintf(buffer, buflen, "lib%s.so", fname);
1622 retval = true;
1623 } else if (strchr(pname, *os::path_separator()) != NULL) {
1624 int n;
1625 char** pelements = split_path(pname, &n);
1626 if (pelements == NULL) {
1627 return false;
1628 }
1629 for (int i = 0 ; i < n ; i++) {
1630 // Really shouldn't be NULL, but check can't hurt
1631 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1632 continue; // skip the empty path values
1633 }
1634 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1635 if (file_exists(buffer)) {
1636 retval = true;
1637 break;
1638 }
1639 }
1640 // release the storage
1641 for (int i = 0 ; i < n ; i++) {
1642 if (pelements[i] != NULL) {
1643 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1644 }
1645 }
1646 if (pelements != NULL) {
1647 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1648 }
1649 } else {
1650 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1651 retval = true;
1652 }
1653 return retval;
1654 }
1656 // check if addr is inside libjvm.so
1657 bool os::address_is_in_vm(address addr) {
1658 static address libjvm_base_addr;
1659 Dl_info dlinfo;
1661 if (libjvm_base_addr == NULL) {
1662 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1663 libjvm_base_addr = (address)dlinfo.dli_fbase;
1664 }
1665 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1666 }
1668 if (dladdr((void *)addr, &dlinfo) != 0) {
1669 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1670 }
1672 return false;
1673 }
1675 bool os::dll_address_to_function_name(address addr, char *buf,
1676 int buflen, int *offset) {
1677 // buf is not optional, but offset is optional
1678 assert(buf != NULL, "sanity check");
1680 Dl_info dlinfo;
1682 if (dladdr((void*)addr, &dlinfo) != 0) {
1683 // see if we have a matching symbol
1684 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1685 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1686 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1687 }
1688 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1689 return true;
1690 }
1691 // no matching symbol so try for just file info
1692 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1693 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1694 buf, buflen, offset, dlinfo.dli_fname)) {
1695 return true;
1696 }
1697 }
1698 }
1700 buf[0] = '\0';
1701 if (offset != NULL) *offset = -1;
1702 return false;
1703 }
1705 struct _address_to_library_name {
1706 address addr; // input : memory address
1707 size_t buflen; // size of fname
1708 char* fname; // output: library name
1709 address base; // library base addr
1710 };
1712 static int address_to_library_name_callback(struct dl_phdr_info *info,
1713 size_t size, void *data) {
1714 int i;
1715 bool found = false;
1716 address libbase = NULL;
1717 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1719 // iterate through all loadable segments
1720 for (i = 0; i < info->dlpi_phnum; i++) {
1721 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1722 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1723 // base address of a library is the lowest address of its loaded
1724 // segments.
1725 if (libbase == NULL || libbase > segbase) {
1726 libbase = segbase;
1727 }
1728 // see if 'addr' is within current segment
1729 if (segbase <= d->addr &&
1730 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1731 found = true;
1732 }
1733 }
1734 }
1736 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1737 // so dll_address_to_library_name() can fall through to use dladdr() which
1738 // can figure out executable name from argv[0].
1739 if (found && info->dlpi_name && info->dlpi_name[0]) {
1740 d->base = libbase;
1741 if (d->fname) {
1742 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1743 }
1744 return 1;
1745 }
1746 return 0;
1747 }
1749 bool os::dll_address_to_library_name(address addr, char* buf,
1750 int buflen, int* offset) {
1751 // buf is not optional, but offset is optional
1752 assert(buf != NULL, "sanity check");
1754 Dl_info dlinfo;
1755 struct _address_to_library_name data;
1757 // There is a bug in old glibc dladdr() implementation that it could resolve
1758 // to wrong library name if the .so file has a base address != NULL. Here
1759 // we iterate through the program headers of all loaded libraries to find
1760 // out which library 'addr' really belongs to. This workaround can be
1761 // removed once the minimum requirement for glibc is moved to 2.3.x.
1762 data.addr = addr;
1763 data.fname = buf;
1764 data.buflen = buflen;
1765 data.base = NULL;
1766 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1768 if (rslt) {
1769 // buf already contains library name
1770 if (offset) *offset = addr - data.base;
1771 return true;
1772 }
1773 if (dladdr((void*)addr, &dlinfo) != 0) {
1774 if (dlinfo.dli_fname != NULL) {
1775 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1776 }
1777 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1778 *offset = addr - (address)dlinfo.dli_fbase;
1779 }
1780 return true;
1781 }
1783 buf[0] = '\0';
1784 if (offset) *offset = -1;
1785 return false;
1786 }
1788 // Loads .dll/.so and
1789 // in case of error it checks if .dll/.so was built for the
1790 // same architecture as Hotspot is running on
1793 // Remember the stack's state. The Linux dynamic linker will change
1794 // the stack to 'executable' at most once, so we must safepoint only once.
1795 bool os::Linux::_stack_is_executable = false;
1797 // VM operation that loads a library. This is necessary if stack protection
1798 // of the Java stacks can be lost during loading the library. If we
1799 // do not stop the Java threads, they can stack overflow before the stacks
1800 // are protected again.
1801 class VM_LinuxDllLoad: public VM_Operation {
1802 private:
1803 const char *_filename;
1804 char *_ebuf;
1805 int _ebuflen;
1806 void *_lib;
1807 public:
1808 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1809 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1810 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1811 void doit() {
1812 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1813 os::Linux::_stack_is_executable = true;
1814 }
1815 void* loaded_library() { return _lib; }
1816 };
1818 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1819 {
1820 void * result = NULL;
1821 bool load_attempted = false;
1823 // Check whether the library to load might change execution rights
1824 // of the stack. If they are changed, the protection of the stack
1825 // guard pages will be lost. We need a safepoint to fix this.
1826 //
1827 // See Linux man page execstack(8) for more info.
1828 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1829 ElfFile ef(filename);
1830 if (!ef.specifies_noexecstack()) {
1831 if (!is_init_completed()) {
1832 os::Linux::_stack_is_executable = true;
1833 // This is OK - No Java threads have been created yet, and hence no
1834 // stack guard pages to fix.
1835 //
1836 // This should happen only when you are building JDK7 using a very
1837 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1838 //
1839 // Dynamic loader will make all stacks executable after
1840 // this function returns, and will not do that again.
1841 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1842 } else {
1843 warning("You have loaded library %s which might have disabled stack guard. "
1844 "The VM will try to fix the stack guard now.\n"
1845 "It's highly recommended that you fix the library with "
1846 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1847 filename);
1849 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1850 JavaThread *jt = JavaThread::current();
1851 if (jt->thread_state() != _thread_in_native) {
1852 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1853 // that requires ExecStack. Cannot enter safe point. Let's give up.
1854 warning("Unable to fix stack guard. Giving up.");
1855 } else {
1856 if (!LoadExecStackDllInVMThread) {
1857 // This is for the case where the DLL has an static
1858 // constructor function that executes JNI code. We cannot
1859 // load such DLLs in the VMThread.
1860 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1861 }
1863 ThreadInVMfromNative tiv(jt);
1864 debug_only(VMNativeEntryWrapper vew;)
1866 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1867 VMThread::execute(&op);
1868 if (LoadExecStackDllInVMThread) {
1869 result = op.loaded_library();
1870 }
1871 load_attempted = true;
1872 }
1873 }
1874 }
1875 }
1877 if (!load_attempted) {
1878 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1879 }
1881 if (result != NULL) {
1882 // Successful loading
1883 return result;
1884 }
1886 Elf32_Ehdr elf_head;
1887 int diag_msg_max_length=ebuflen-strlen(ebuf);
1888 char* diag_msg_buf=ebuf+strlen(ebuf);
1890 if (diag_msg_max_length==0) {
1891 // No more space in ebuf for additional diagnostics message
1892 return NULL;
1893 }
1896 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1898 if (file_descriptor < 0) {
1899 // Can't open library, report dlerror() message
1900 return NULL;
1901 }
1903 bool failed_to_read_elf_head=
1904 (sizeof(elf_head)!=
1905 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1907 ::close(file_descriptor);
1908 if (failed_to_read_elf_head) {
1909 // file i/o error - report dlerror() msg
1910 return NULL;
1911 }
1913 typedef struct {
1914 Elf32_Half code; // Actual value as defined in elf.h
1915 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1916 char elf_class; // 32 or 64 bit
1917 char endianess; // MSB or LSB
1918 char* name; // String representation
1919 } arch_t;
1921 #ifndef EM_486
1922 #define EM_486 6 /* Intel 80486 */
1923 #endif
1925 static const arch_t arch_array[]={
1926 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1927 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1928 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1929 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1930 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1931 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1932 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1933 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1934 #if defined(VM_LITTLE_ENDIAN)
1935 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1936 #else
1937 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1938 #endif
1939 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1940 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1941 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1942 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1943 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1944 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1945 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1946 };
1948 #if (defined IA32)
1949 static Elf32_Half running_arch_code=EM_386;
1950 #elif (defined AMD64)
1951 static Elf32_Half running_arch_code=EM_X86_64;
1952 #elif (defined IA64)
1953 static Elf32_Half running_arch_code=EM_IA_64;
1954 #elif (defined __sparc) && (defined _LP64)
1955 static Elf32_Half running_arch_code=EM_SPARCV9;
1956 #elif (defined __sparc) && (!defined _LP64)
1957 static Elf32_Half running_arch_code=EM_SPARC;
1958 #elif (defined __powerpc64__)
1959 static Elf32_Half running_arch_code=EM_PPC64;
1960 #elif (defined __powerpc__)
1961 static Elf32_Half running_arch_code=EM_PPC;
1962 #elif (defined ARM)
1963 static Elf32_Half running_arch_code=EM_ARM;
1964 #elif (defined S390)
1965 static Elf32_Half running_arch_code=EM_S390;
1966 #elif (defined ALPHA)
1967 static Elf32_Half running_arch_code=EM_ALPHA;
1968 #elif (defined MIPSEL)
1969 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1970 #elif (defined PARISC)
1971 static Elf32_Half running_arch_code=EM_PARISC;
1972 #elif (defined MIPS)
1973 static Elf32_Half running_arch_code=EM_MIPS;
1974 #elif (defined M68K)
1975 static Elf32_Half running_arch_code=EM_68K;
1976 #else
1977 #error Method os::dll_load requires that one of following is defined:\
1978 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1979 #endif
1981 // Identify compatability class for VM's architecture and library's architecture
1982 // Obtain string descriptions for architectures
1984 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1985 int running_arch_index=-1;
1987 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1988 if (running_arch_code == arch_array[i].code) {
1989 running_arch_index = i;
1990 }
1991 if (lib_arch.code == arch_array[i].code) {
1992 lib_arch.compat_class = arch_array[i].compat_class;
1993 lib_arch.name = arch_array[i].name;
1994 }
1995 }
1997 assert(running_arch_index != -1,
1998 "Didn't find running architecture code (running_arch_code) in arch_array");
1999 if (running_arch_index == -1) {
2000 // Even though running architecture detection failed
2001 // we may still continue with reporting dlerror() message
2002 return NULL;
2003 }
2005 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
2006 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
2007 return NULL;
2008 }
2010 #ifndef S390
2011 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2012 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2013 return NULL;
2014 }
2015 #endif // !S390
2017 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2018 if ( lib_arch.name!=NULL ) {
2019 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2020 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2021 lib_arch.name, arch_array[running_arch_index].name);
2022 } else {
2023 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2024 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2025 lib_arch.code,
2026 arch_array[running_arch_index].name);
2027 }
2028 }
2030 return NULL;
2031 }
2033 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2034 void * result = ::dlopen(filename, RTLD_LAZY);
2035 if (result == NULL) {
2036 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2037 ebuf[ebuflen-1] = '\0';
2038 }
2039 return result;
2040 }
2042 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2043 void * result = NULL;
2044 if (LoadExecStackDllInVMThread) {
2045 result = dlopen_helper(filename, ebuf, ebuflen);
2046 }
2048 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2049 // library that requires an executable stack, or which does not have this
2050 // stack attribute set, dlopen changes the stack attribute to executable. The
2051 // read protection of the guard pages gets lost.
2052 //
2053 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2054 // may have been queued at the same time.
2056 if (!_stack_is_executable) {
2057 JavaThread *jt = Threads::first();
2059 while (jt) {
2060 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2061 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2062 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2063 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2064 warning("Attempt to reguard stack yellow zone failed.");
2065 }
2066 }
2067 jt = jt->next();
2068 }
2069 }
2071 return result;
2072 }
2074 /*
2075 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2076 * chances are you might want to run the generated bits against glibc-2.0
2077 * libdl.so, so always use locking for any version of glibc.
2078 */
2079 void* os::dll_lookup(void* handle, const char* name) {
2080 pthread_mutex_lock(&dl_mutex);
2081 void* res = dlsym(handle, name);
2082 pthread_mutex_unlock(&dl_mutex);
2083 return res;
2084 }
2086 void* os::get_default_process_handle() {
2087 return (void*)::dlopen(NULL, RTLD_LAZY);
2088 }
2090 static bool _print_ascii_file(const char* filename, outputStream* st) {
2091 int fd = ::open(filename, O_RDONLY);
2092 if (fd == -1) {
2093 return false;
2094 }
2096 char buf[32];
2097 int bytes;
2098 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2099 st->print_raw(buf, bytes);
2100 }
2102 ::close(fd);
2104 return true;
2105 }
2107 void os::print_dll_info(outputStream *st) {
2108 st->print_cr("Dynamic libraries:");
2110 char fname[32];
2111 pid_t pid = os::Linux::gettid();
2113 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2115 if (!_print_ascii_file(fname, st)) {
2116 st->print("Can not get library information for pid = %d\n", pid);
2117 }
2118 }
2120 void os::print_os_info_brief(outputStream* st) {
2121 os::Linux::print_distro_info(st);
2123 os::Posix::print_uname_info(st);
2125 os::Linux::print_libversion_info(st);
2127 }
2129 void os::print_os_info(outputStream* st) {
2130 st->print("OS:");
2132 os::Linux::print_distro_info(st);
2134 os::Posix::print_uname_info(st);
2136 // Print warning if unsafe chroot environment detected
2137 if (unsafe_chroot_detected) {
2138 st->print("WARNING!! ");
2139 st->print_cr("%s", unstable_chroot_error);
2140 }
2142 os::Linux::print_libversion_info(st);
2144 os::Posix::print_rlimit_info(st);
2146 os::Posix::print_load_average(st);
2148 os::Linux::print_full_memory_info(st);
2149 }
2151 // Try to identify popular distros.
2152 // Most Linux distributions have a /etc/XXX-release file, which contains
2153 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2154 // file that also contains the OS version string. Some have more than one
2155 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2156 // /etc/redhat-release.), so the order is important.
2157 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2158 // their own specific XXX-release file as well as a redhat-release file.
2159 // Because of this the XXX-release file needs to be searched for before the
2160 // redhat-release file.
2161 // Since Red Hat has a lsb-release file that is not very descriptive the
2162 // search for redhat-release needs to be before lsb-release.
2163 // Since the lsb-release file is the new standard it needs to be searched
2164 // before the older style release files.
2165 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2166 // next to last resort. The os-release file is a new standard that contains
2167 // distribution information and the system-release file seems to be an old
2168 // standard that has been replaced by the lsb-release and os-release files.
2169 // Searching for the debian_version file is the last resort. It contains
2170 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2171 // "Debian " is printed before the contents of the debian_version file.
2172 void os::Linux::print_distro_info(outputStream* st) {
2173 if (!_print_ascii_file("/etc/oracle-release", st) &&
2174 !_print_ascii_file("/etc/mandriva-release", st) &&
2175 !_print_ascii_file("/etc/mandrake-release", st) &&
2176 !_print_ascii_file("/etc/sun-release", st) &&
2177 !_print_ascii_file("/etc/redhat-release", st) &&
2178 !_print_ascii_file("/etc/lsb-release", st) &&
2179 !_print_ascii_file("/etc/SuSE-release", st) &&
2180 !_print_ascii_file("/etc/turbolinux-release", st) &&
2181 !_print_ascii_file("/etc/gentoo-release", st) &&
2182 !_print_ascii_file("/etc/ltib-release", st) &&
2183 !_print_ascii_file("/etc/angstrom-version", st) &&
2184 !_print_ascii_file("/etc/system-release", st) &&
2185 !_print_ascii_file("/etc/os-release", st)) {
2187 if (file_exists("/etc/debian_version")) {
2188 st->print("Debian ");
2189 _print_ascii_file("/etc/debian_version", st);
2190 } else {
2191 st->print("Linux");
2192 }
2193 }
2194 st->cr();
2195 }
2197 void os::Linux::print_libversion_info(outputStream* st) {
2198 // libc, pthread
2199 st->print("libc:");
2200 st->print("%s ", os::Linux::glibc_version());
2201 st->print("%s ", os::Linux::libpthread_version());
2202 if (os::Linux::is_LinuxThreads()) {
2203 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2204 }
2205 st->cr();
2206 }
2208 void os::Linux::print_full_memory_info(outputStream* st) {
2209 st->print("\n/proc/meminfo:\n");
2210 _print_ascii_file("/proc/meminfo", st);
2211 st->cr();
2212 }
2214 void os::print_memory_info(outputStream* st) {
2216 st->print("Memory:");
2217 st->print(" %dk page", os::vm_page_size()>>10);
2219 // values in struct sysinfo are "unsigned long"
2220 struct sysinfo si;
2221 sysinfo(&si);
2223 st->print(", physical " UINT64_FORMAT "k",
2224 os::physical_memory() >> 10);
2225 st->print("(" UINT64_FORMAT "k free)",
2226 os::available_memory() >> 10);
2227 st->print(", swap " UINT64_FORMAT "k",
2228 ((jlong)si.totalswap * si.mem_unit) >> 10);
2229 st->print("(" UINT64_FORMAT "k free)",
2230 ((jlong)si.freeswap * si.mem_unit) >> 10);
2231 st->cr();
2232 }
2234 void os::pd_print_cpu_info(outputStream* st) {
2235 st->print("\n/proc/cpuinfo:\n");
2236 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2237 st->print(" <Not Available>");
2238 }
2239 st->cr();
2240 }
2242 void os::print_siginfo(outputStream* st, void* siginfo) {
2243 const siginfo_t* si = (const siginfo_t*)siginfo;
2245 os::Posix::print_siginfo_brief(st, si);
2247 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2248 UseSharedSpaces) {
2249 FileMapInfo* mapinfo = FileMapInfo::current_info();
2250 if (mapinfo->is_in_shared_space(si->si_addr)) {
2251 st->print("\n\nError accessing class data sharing archive." \
2252 " Mapped file inaccessible during execution, " \
2253 " possible disk/network problem.");
2254 }
2255 }
2256 st->cr();
2257 }
2260 static void print_signal_handler(outputStream* st, int sig,
2261 char* buf, size_t buflen);
2263 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2264 st->print_cr("Signal Handlers:");
2265 print_signal_handler(st, SIGSEGV, buf, buflen);
2266 print_signal_handler(st, SIGBUS , buf, buflen);
2267 print_signal_handler(st, SIGFPE , buf, buflen);
2268 print_signal_handler(st, SIGPIPE, buf, buflen);
2269 print_signal_handler(st, SIGXFSZ, buf, buflen);
2270 print_signal_handler(st, SIGILL , buf, buflen);
2271 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2272 print_signal_handler(st, SR_signum, buf, buflen);
2273 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2274 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2275 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2276 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2277 #if defined(PPC64)
2278 print_signal_handler(st, SIGTRAP, buf, buflen);
2279 #endif
2280 }
2282 static char saved_jvm_path[MAXPATHLEN] = {0};
2284 // Find the full path to the current module, libjvm.so
2285 void os::jvm_path(char *buf, jint buflen) {
2286 // Error checking.
2287 if (buflen < MAXPATHLEN) {
2288 assert(false, "must use a large-enough buffer");
2289 buf[0] = '\0';
2290 return;
2291 }
2292 // Lazy resolve the path to current module.
2293 if (saved_jvm_path[0] != 0) {
2294 strcpy(buf, saved_jvm_path);
2295 return;
2296 }
2298 char dli_fname[MAXPATHLEN];
2299 bool ret = dll_address_to_library_name(
2300 CAST_FROM_FN_PTR(address, os::jvm_path),
2301 dli_fname, sizeof(dli_fname), NULL);
2302 assert(ret, "cannot locate libjvm");
2303 char *rp = NULL;
2304 if (ret && dli_fname[0] != '\0') {
2305 rp = realpath(dli_fname, buf);
2306 }
2307 if (rp == NULL)
2308 return;
2310 if (Arguments::created_by_gamma_launcher()) {
2311 // Support for the gamma launcher. Typical value for buf is
2312 // "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so". If "/jre/lib/" appears at
2313 // the right place in the string, then assume we are installed in a JDK and
2314 // we're done. Otherwise, check for a JAVA_HOME environment variable and fix
2315 // up the path so it looks like libjvm.so is installed there (append a
2316 // fake suffix hotspot/libjvm.so).
2317 const char *p = buf + strlen(buf) - 1;
2318 for (int count = 0; p > buf && count < 5; ++count) {
2319 for (--p; p > buf && *p != '/'; --p)
2320 /* empty */ ;
2321 }
2323 if (strncmp(p, "/jre/lib/", 9) != 0) {
2324 // Look for JAVA_HOME in the environment.
2325 char* java_home_var = ::getenv("JAVA_HOME");
2326 if (java_home_var != NULL && java_home_var[0] != 0) {
2327 char* jrelib_p;
2328 int len;
2330 // Check the current module name "libjvm.so".
2331 p = strrchr(buf, '/');
2332 assert(strstr(p, "/libjvm") == p, "invalid library name");
2334 rp = realpath(java_home_var, buf);
2335 if (rp == NULL)
2336 return;
2338 // determine if this is a legacy image or modules image
2339 // modules image doesn't have "jre" subdirectory
2340 len = strlen(buf);
2341 assert(len < buflen, "Ran out of buffer room");
2342 jrelib_p = buf + len;
2343 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2344 if (0 != access(buf, F_OK)) {
2345 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2346 }
2348 if (0 == access(buf, F_OK)) {
2349 // Use current module name "libjvm.so"
2350 len = strlen(buf);
2351 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2352 } else {
2353 // Go back to path of .so
2354 rp = realpath(dli_fname, buf);
2355 if (rp == NULL)
2356 return;
2357 }
2358 }
2359 }
2360 }
2362 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2363 }
2365 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2366 // no prefix required, not even "_"
2367 }
2369 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2370 // no suffix required
2371 }
2373 ////////////////////////////////////////////////////////////////////////////////
2374 // sun.misc.Signal support
2376 static volatile jint sigint_count = 0;
2378 static void
2379 UserHandler(int sig, void *siginfo, void *context) {
2380 // 4511530 - sem_post is serialized and handled by the manager thread. When
2381 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2382 // don't want to flood the manager thread with sem_post requests.
2383 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2384 return;
2386 // Ctrl-C is pressed during error reporting, likely because the error
2387 // handler fails to abort. Let VM die immediately.
2388 if (sig == SIGINT && is_error_reported()) {
2389 os::die();
2390 }
2392 os::signal_notify(sig);
2393 }
2395 void* os::user_handler() {
2396 return CAST_FROM_FN_PTR(void*, UserHandler);
2397 }
2399 class Semaphore : public StackObj {
2400 public:
2401 Semaphore();
2402 ~Semaphore();
2403 void signal();
2404 void wait();
2405 bool trywait();
2406 bool timedwait(unsigned int sec, int nsec);
2407 private:
2408 sem_t _semaphore;
2409 };
2411 Semaphore::Semaphore() {
2412 sem_init(&_semaphore, 0, 0);
2413 }
2415 Semaphore::~Semaphore() {
2416 sem_destroy(&_semaphore);
2417 }
2419 void Semaphore::signal() {
2420 sem_post(&_semaphore);
2421 }
2423 void Semaphore::wait() {
2424 sem_wait(&_semaphore);
2425 }
2427 bool Semaphore::trywait() {
2428 return sem_trywait(&_semaphore) == 0;
2429 }
2431 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2433 struct timespec ts;
2434 // Semaphore's are always associated with CLOCK_REALTIME
2435 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2436 // see unpackTime for discussion on overflow checking
2437 if (sec >= MAX_SECS) {
2438 ts.tv_sec += MAX_SECS;
2439 ts.tv_nsec = 0;
2440 } else {
2441 ts.tv_sec += sec;
2442 ts.tv_nsec += nsec;
2443 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2444 ts.tv_nsec -= NANOSECS_PER_SEC;
2445 ++ts.tv_sec; // note: this must be <= max_secs
2446 }
2447 }
2449 while (1) {
2450 int result = sem_timedwait(&_semaphore, &ts);
2451 if (result == 0) {
2452 return true;
2453 } else if (errno == EINTR) {
2454 continue;
2455 } else if (errno == ETIMEDOUT) {
2456 return false;
2457 } else {
2458 return false;
2459 }
2460 }
2461 }
2463 extern "C" {
2464 typedef void (*sa_handler_t)(int);
2465 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2466 }
2468 void* os::signal(int signal_number, void* handler) {
2469 struct sigaction sigAct, oldSigAct;
2471 sigfillset(&(sigAct.sa_mask));
2472 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2473 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2475 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2476 // -1 means registration failed
2477 return (void *)-1;
2478 }
2480 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2481 }
2483 void os::signal_raise(int signal_number) {
2484 ::raise(signal_number);
2485 }
2487 /*
2488 * The following code is moved from os.cpp for making this
2489 * code platform specific, which it is by its very nature.
2490 */
2492 // Will be modified when max signal is changed to be dynamic
2493 int os::sigexitnum_pd() {
2494 return NSIG;
2495 }
2497 // a counter for each possible signal value
2498 static volatile jint pending_signals[NSIG+1] = { 0 };
2500 // Linux(POSIX) specific hand shaking semaphore.
2501 static sem_t sig_sem;
2502 static Semaphore sr_semaphore;
2504 void os::signal_init_pd() {
2505 // Initialize signal structures
2506 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2508 // Initialize signal semaphore
2509 ::sem_init(&sig_sem, 0, 0);
2510 }
2512 void os::signal_notify(int sig) {
2513 Atomic::inc(&pending_signals[sig]);
2514 ::sem_post(&sig_sem);
2515 }
2517 static int check_pending_signals(bool wait) {
2518 Atomic::store(0, &sigint_count);
2519 for (;;) {
2520 for (int i = 0; i < NSIG + 1; i++) {
2521 jint n = pending_signals[i];
2522 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2523 return i;
2524 }
2525 }
2526 if (!wait) {
2527 return -1;
2528 }
2529 JavaThread *thread = JavaThread::current();
2530 ThreadBlockInVM tbivm(thread);
2532 bool threadIsSuspended;
2533 do {
2534 thread->set_suspend_equivalent();
2535 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2536 ::sem_wait(&sig_sem);
2538 // were we externally suspended while we were waiting?
2539 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2540 if (threadIsSuspended) {
2541 //
2542 // The semaphore has been incremented, but while we were waiting
2543 // another thread suspended us. We don't want to continue running
2544 // while suspended because that would surprise the thread that
2545 // suspended us.
2546 //
2547 ::sem_post(&sig_sem);
2549 thread->java_suspend_self();
2550 }
2551 } while (threadIsSuspended);
2552 }
2553 }
2555 int os::signal_lookup() {
2556 return check_pending_signals(false);
2557 }
2559 int os::signal_wait() {
2560 return check_pending_signals(true);
2561 }
2563 ////////////////////////////////////////////////////////////////////////////////
2564 // Virtual Memory
2566 int os::vm_page_size() {
2567 // Seems redundant as all get out
2568 assert(os::Linux::page_size() != -1, "must call os::init");
2569 return os::Linux::page_size();
2570 }
2572 // Solaris allocates memory by pages.
2573 int os::vm_allocation_granularity() {
2574 assert(os::Linux::page_size() != -1, "must call os::init");
2575 return os::Linux::page_size();
2576 }
2578 // Rationale behind this function:
2579 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2580 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2581 // samples for JITted code. Here we create private executable mapping over the code cache
2582 // and then we can use standard (well, almost, as mapping can change) way to provide
2583 // info for the reporting script by storing timestamp and location of symbol
2584 void linux_wrap_code(char* base, size_t size) {
2585 static volatile jint cnt = 0;
2587 if (!UseOprofile) {
2588 return;
2589 }
2591 char buf[PATH_MAX+1];
2592 int num = Atomic::add(1, &cnt);
2594 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2595 os::get_temp_directory(), os::current_process_id(), num);
2596 unlink(buf);
2598 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2600 if (fd != -1) {
2601 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2602 if (rv != (off_t)-1) {
2603 if (::write(fd, "", 1) == 1) {
2604 mmap(base, size,
2605 PROT_READ|PROT_WRITE|PROT_EXEC,
2606 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2607 }
2608 }
2609 ::close(fd);
2610 unlink(buf);
2611 }
2612 }
2614 static bool recoverable_mmap_error(int err) {
2615 // See if the error is one we can let the caller handle. This
2616 // list of errno values comes from JBS-6843484. I can't find a
2617 // Linux man page that documents this specific set of errno
2618 // values so while this list currently matches Solaris, it may
2619 // change as we gain experience with this failure mode.
2620 switch (err) {
2621 case EBADF:
2622 case EINVAL:
2623 case ENOTSUP:
2624 // let the caller deal with these errors
2625 return true;
2627 default:
2628 // Any remaining errors on this OS can cause our reserved mapping
2629 // to be lost. That can cause confusion where different data
2630 // structures think they have the same memory mapped. The worst
2631 // scenario is if both the VM and a library think they have the
2632 // same memory mapped.
2633 return false;
2634 }
2635 }
2637 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2638 int err) {
2639 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2640 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2641 strerror(err), err);
2642 }
2644 static void warn_fail_commit_memory(char* addr, size_t size,
2645 size_t alignment_hint, bool exec,
2646 int err) {
2647 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2648 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2649 alignment_hint, exec, strerror(err), err);
2650 }
2652 // NOTE: Linux kernel does not really reserve the pages for us.
2653 // All it does is to check if there are enough free pages
2654 // left at the time of mmap(). This could be a potential
2655 // problem.
2656 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2657 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2658 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2659 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2660 if (res != (uintptr_t) MAP_FAILED) {
2661 if (UseNUMAInterleaving) {
2662 numa_make_global(addr, size);
2663 }
2664 return 0;
2665 }
2667 int err = errno; // save errno from mmap() call above
2669 if (!recoverable_mmap_error(err)) {
2670 warn_fail_commit_memory(addr, size, exec, err);
2671 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2672 }
2674 return err;
2675 }
2677 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2678 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2679 }
2681 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2682 const char* mesg) {
2683 assert(mesg != NULL, "mesg must be specified");
2684 int err = os::Linux::commit_memory_impl(addr, size, exec);
2685 if (err != 0) {
2686 // the caller wants all commit errors to exit with the specified mesg:
2687 warn_fail_commit_memory(addr, size, exec, err);
2688 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2689 }
2690 }
2692 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2693 #ifndef MAP_HUGETLB
2694 #define MAP_HUGETLB 0x40000
2695 #endif
2697 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2698 #ifndef MADV_HUGEPAGE
2699 #define MADV_HUGEPAGE 14
2700 #endif
2702 int os::Linux::commit_memory_impl(char* addr, size_t size,
2703 size_t alignment_hint, bool exec) {
2704 int err = os::Linux::commit_memory_impl(addr, size, exec);
2705 if (err == 0) {
2706 realign_memory(addr, size, alignment_hint);
2707 }
2708 return err;
2709 }
2711 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2712 bool exec) {
2713 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2714 }
2716 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2717 size_t alignment_hint, bool exec,
2718 const char* mesg) {
2719 assert(mesg != NULL, "mesg must be specified");
2720 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2721 if (err != 0) {
2722 // the caller wants all commit errors to exit with the specified mesg:
2723 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2724 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2725 }
2726 }
2728 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2729 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2730 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2731 // be supported or the memory may already be backed by huge pages.
2732 ::madvise(addr, bytes, MADV_HUGEPAGE);
2733 }
2734 }
2736 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2737 // This method works by doing an mmap over an existing mmaping and effectively discarding
2738 // the existing pages. However it won't work for SHM-based large pages that cannot be
2739 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2740 // small pages on top of the SHM segment. This method always works for small pages, so we
2741 // allow that in any case.
2742 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2743 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2744 }
2745 }
2747 void os::numa_make_global(char *addr, size_t bytes) {
2748 Linux::numa_interleave_memory(addr, bytes);
2749 }
2751 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2752 // bind policy to MPOL_PREFERRED for the current thread.
2753 #define USE_MPOL_PREFERRED 0
2755 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2756 // To make NUMA and large pages more robust when both enabled, we need to ease
2757 // the requirements on where the memory should be allocated. MPOL_BIND is the
2758 // default policy and it will force memory to be allocated on the specified
2759 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2760 // the specified node, but will not force it. Using this policy will prevent
2761 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2762 // free large pages.
2763 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2764 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2765 }
2767 bool os::numa_topology_changed() { return false; }
2769 size_t os::numa_get_groups_num() {
2770 int max_node = Linux::numa_max_node();
2771 return max_node > 0 ? max_node + 1 : 1;
2772 }
2774 int os::numa_get_group_id() {
2775 int cpu_id = Linux::sched_getcpu();
2776 if (cpu_id != -1) {
2777 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2778 if (lgrp_id != -1) {
2779 return lgrp_id;
2780 }
2781 }
2782 return 0;
2783 }
2785 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2786 for (size_t i = 0; i < size; i++) {
2787 ids[i] = i;
2788 }
2789 return size;
2790 }
2792 bool os::get_page_info(char *start, page_info* info) {
2793 return false;
2794 }
2796 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2797 return end;
2798 }
2801 int os::Linux::sched_getcpu_syscall(void) {
2802 unsigned int cpu;
2803 int retval = -1;
2805 #if defined(IA32)
2806 # ifndef SYS_getcpu
2807 # define SYS_getcpu 318
2808 # endif
2809 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2810 #elif defined(AMD64)
2811 // Unfortunately we have to bring all these macros here from vsyscall.h
2812 // to be able to compile on old linuxes.
2813 # define __NR_vgetcpu 2
2814 # define VSYSCALL_START (-10UL << 20)
2815 # define VSYSCALL_SIZE 1024
2816 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2817 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2818 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2819 retval = vgetcpu(&cpu, NULL, NULL);
2820 #endif
2822 return (retval == -1) ? retval : cpu;
2823 }
2825 // Something to do with the numa-aware allocator needs these symbols
2826 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2827 extern "C" JNIEXPORT void numa_error(char *where) { }
2828 extern "C" JNIEXPORT int fork1() { return fork(); }
2831 // If we are running with libnuma version > 2, then we should
2832 // be trying to use symbols with versions 1.1
2833 // If we are running with earlier version, which did not have symbol versions,
2834 // we should use the base version.
2835 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2836 void *f = dlvsym(handle, name, "libnuma_1.1");
2837 if (f == NULL) {
2838 f = dlsym(handle, name);
2839 }
2840 return f;
2841 }
2843 bool os::Linux::libnuma_init() {
2844 // sched_getcpu() should be in libc.
2845 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2846 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2848 // If it's not, try a direct syscall.
2849 if (sched_getcpu() == -1)
2850 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2852 if (sched_getcpu() != -1) { // Does it work?
2853 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2854 if (handle != NULL) {
2855 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2856 libnuma_dlsym(handle, "numa_node_to_cpus")));
2857 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2858 libnuma_dlsym(handle, "numa_max_node")));
2859 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2860 libnuma_dlsym(handle, "numa_available")));
2861 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2862 libnuma_dlsym(handle, "numa_tonode_memory")));
2863 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2864 libnuma_dlsym(handle, "numa_interleave_memory")));
2865 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2866 libnuma_dlsym(handle, "numa_set_bind_policy")));
2869 if (numa_available() != -1) {
2870 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2871 // Create a cpu -> node mapping
2872 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2873 rebuild_cpu_to_node_map();
2874 return true;
2875 }
2876 }
2877 }
2878 return false;
2879 }
2881 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2882 // The table is later used in get_node_by_cpu().
2883 void os::Linux::rebuild_cpu_to_node_map() {
2884 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2885 // in libnuma (possible values are starting from 16,
2886 // and continuing up with every other power of 2, but less
2887 // than the maximum number of CPUs supported by kernel), and
2888 // is a subject to change (in libnuma version 2 the requirements
2889 // are more reasonable) we'll just hardcode the number they use
2890 // in the library.
2891 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2893 size_t cpu_num = os::active_processor_count();
2894 size_t cpu_map_size = NCPUS / BitsPerCLong;
2895 size_t cpu_map_valid_size =
2896 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2898 cpu_to_node()->clear();
2899 cpu_to_node()->at_grow(cpu_num - 1);
2900 size_t node_num = numa_get_groups_num();
2902 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2903 for (size_t i = 0; i < node_num; i++) {
2904 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2905 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2906 if (cpu_map[j] != 0) {
2907 for (size_t k = 0; k < BitsPerCLong; k++) {
2908 if (cpu_map[j] & (1UL << k)) {
2909 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2910 }
2911 }
2912 }
2913 }
2914 }
2915 }
2916 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2917 }
2919 int os::Linux::get_node_by_cpu(int cpu_id) {
2920 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2921 return cpu_to_node()->at(cpu_id);
2922 }
2923 return -1;
2924 }
2926 GrowableArray<int>* os::Linux::_cpu_to_node;
2927 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2928 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2929 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2930 os::Linux::numa_available_func_t os::Linux::_numa_available;
2931 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2932 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2933 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2934 unsigned long* os::Linux::_numa_all_nodes;
2936 bool os::pd_uncommit_memory(char* addr, size_t size) {
2937 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2938 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2939 return res != (uintptr_t) MAP_FAILED;
2940 }
2942 static
2943 address get_stack_commited_bottom(address bottom, size_t size) {
2944 address nbot = bottom;
2945 address ntop = bottom + size;
2947 size_t page_sz = os::vm_page_size();
2948 unsigned pages = size / page_sz;
2950 unsigned char vec[1];
2951 unsigned imin = 1, imax = pages + 1, imid;
2952 int mincore_return_value = 0;
2954 assert(imin <= imax, "Unexpected page size");
2956 while (imin < imax) {
2957 imid = (imax + imin) / 2;
2958 nbot = ntop - (imid * page_sz);
2960 // Use a trick with mincore to check whether the page is mapped or not.
2961 // mincore sets vec to 1 if page resides in memory and to 0 if page
2962 // is swapped output but if page we are asking for is unmapped
2963 // it returns -1,ENOMEM
2964 mincore_return_value = mincore(nbot, page_sz, vec);
2966 if (mincore_return_value == -1) {
2967 // Page is not mapped go up
2968 // to find first mapped page
2969 if (errno != EAGAIN) {
2970 assert(errno == ENOMEM, "Unexpected mincore errno");
2971 imax = imid;
2972 }
2973 } else {
2974 // Page is mapped go down
2975 // to find first not mapped page
2976 imin = imid + 1;
2977 }
2978 }
2980 nbot = nbot + page_sz;
2982 // Adjust stack bottom one page up if last checked page is not mapped
2983 if (mincore_return_value == -1) {
2984 nbot = nbot + page_sz;
2985 }
2987 return nbot;
2988 }
2991 // Linux uses a growable mapping for the stack, and if the mapping for
2992 // the stack guard pages is not removed when we detach a thread the
2993 // stack cannot grow beyond the pages where the stack guard was
2994 // mapped. If at some point later in the process the stack expands to
2995 // that point, the Linux kernel cannot expand the stack any further
2996 // because the guard pages are in the way, and a segfault occurs.
2997 //
2998 // However, it's essential not to split the stack region by unmapping
2999 // a region (leaving a hole) that's already part of the stack mapping,
3000 // so if the stack mapping has already grown beyond the guard pages at
3001 // the time we create them, we have to truncate the stack mapping.
3002 // So, we need to know the extent of the stack mapping when
3003 // create_stack_guard_pages() is called.
3005 // We only need this for stacks that are growable: at the time of
3006 // writing thread stacks don't use growable mappings (i.e. those
3007 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3008 // only applies to the main thread.
3010 // If the (growable) stack mapping already extends beyond the point
3011 // where we're going to put our guard pages, truncate the mapping at
3012 // that point by munmap()ping it. This ensures that when we later
3013 // munmap() the guard pages we don't leave a hole in the stack
3014 // mapping. This only affects the main/initial thread
3016 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3018 if (os::Linux::is_initial_thread()) {
3019 // As we manually grow stack up to bottom inside create_attached_thread(),
3020 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3021 // we don't need to do anything special.
3022 // Check it first, before calling heavy function.
3023 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3024 unsigned char vec[1];
3026 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3027 // Fallback to slow path on all errors, including EAGAIN
3028 stack_extent = (uintptr_t) get_stack_commited_bottom(
3029 os::Linux::initial_thread_stack_bottom(),
3030 (size_t)addr - stack_extent);
3031 }
3033 if (stack_extent < (uintptr_t)addr) {
3034 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3035 }
3036 }
3038 return os::commit_memory(addr, size, !ExecMem);
3039 }
3041 // If this is a growable mapping, remove the guard pages entirely by
3042 // munmap()ping them. If not, just call uncommit_memory(). This only
3043 // affects the main/initial thread, but guard against future OS changes
3044 // It's safe to always unmap guard pages for initial thread because we
3045 // always place it right after end of the mapped region
3047 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3048 uintptr_t stack_extent, stack_base;
3050 if (os::Linux::is_initial_thread()) {
3051 return ::munmap(addr, size) == 0;
3052 }
3054 return os::uncommit_memory(addr, size);
3055 }
3057 static address _highest_vm_reserved_address = NULL;
3059 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3060 // at 'requested_addr'. If there are existing memory mappings at the same
3061 // location, however, they will be overwritten. If 'fixed' is false,
3062 // 'requested_addr' is only treated as a hint, the return value may or
3063 // may not start from the requested address. Unlike Linux mmap(), this
3064 // function returns NULL to indicate failure.
3065 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3066 char * addr;
3067 int flags;
3069 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3070 if (fixed) {
3071 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3072 flags |= MAP_FIXED;
3073 }
3075 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3076 // touch an uncommitted page. Otherwise, the read/write might
3077 // succeed if we have enough swap space to back the physical page.
3078 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3079 flags, -1, 0);
3081 if (addr != MAP_FAILED) {
3082 // anon_mmap() should only get called during VM initialization,
3083 // don't need lock (actually we can skip locking even it can be called
3084 // from multiple threads, because _highest_vm_reserved_address is just a
3085 // hint about the upper limit of non-stack memory regions.)
3086 if ((address)addr + bytes > _highest_vm_reserved_address) {
3087 _highest_vm_reserved_address = (address)addr + bytes;
3088 }
3089 }
3091 return addr == MAP_FAILED ? NULL : addr;
3092 }
3094 // Don't update _highest_vm_reserved_address, because there might be memory
3095 // regions above addr + size. If so, releasing a memory region only creates
3096 // a hole in the address space, it doesn't help prevent heap-stack collision.
3097 //
3098 static int anon_munmap(char * addr, size_t size) {
3099 return ::munmap(addr, size) == 0;
3100 }
3102 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3103 size_t alignment_hint) {
3104 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3105 }
3107 bool os::pd_release_memory(char* addr, size_t size) {
3108 return anon_munmap(addr, size);
3109 }
3111 static address highest_vm_reserved_address() {
3112 return _highest_vm_reserved_address;
3113 }
3115 static bool linux_mprotect(char* addr, size_t size, int prot) {
3116 // Linux wants the mprotect address argument to be page aligned.
3117 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3119 // According to SUSv3, mprotect() should only be used with mappings
3120 // established by mmap(), and mmap() always maps whole pages. Unaligned
3121 // 'addr' likely indicates problem in the VM (e.g. trying to change
3122 // protection of malloc'ed or statically allocated memory). Check the
3123 // caller if you hit this assert.
3124 assert(addr == bottom, "sanity check");
3126 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3127 return ::mprotect(bottom, size, prot) == 0;
3128 }
3130 // Set protections specified
3131 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3132 bool is_committed) {
3133 unsigned int p = 0;
3134 switch (prot) {
3135 case MEM_PROT_NONE: p = PROT_NONE; break;
3136 case MEM_PROT_READ: p = PROT_READ; break;
3137 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3138 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3139 default:
3140 ShouldNotReachHere();
3141 }
3142 // is_committed is unused.
3143 return linux_mprotect(addr, bytes, p);
3144 }
3146 bool os::guard_memory(char* addr, size_t size) {
3147 return linux_mprotect(addr, size, PROT_NONE);
3148 }
3150 bool os::unguard_memory(char* addr, size_t size) {
3151 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3152 }
3154 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3155 bool result = false;
3156 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3157 MAP_ANONYMOUS|MAP_PRIVATE,
3158 -1, 0);
3159 if (p != MAP_FAILED) {
3160 void *aligned_p = align_ptr_up(p, page_size);
3162 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3164 munmap(p, page_size * 2);
3165 }
3167 if (warn && !result) {
3168 warning("TransparentHugePages is not supported by the operating system.");
3169 }
3171 return result;
3172 }
3174 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3175 bool result = false;
3176 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3177 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3178 -1, 0);
3180 if (p != MAP_FAILED) {
3181 // We don't know if this really is a huge page or not.
3182 FILE *fp = fopen("/proc/self/maps", "r");
3183 if (fp) {
3184 while (!feof(fp)) {
3185 char chars[257];
3186 long x = 0;
3187 if (fgets(chars, sizeof(chars), fp)) {
3188 if (sscanf(chars, "%lx-%*x", &x) == 1
3189 && x == (long)p) {
3190 if (strstr (chars, "hugepage")) {
3191 result = true;
3192 break;
3193 }
3194 }
3195 }
3196 }
3197 fclose(fp);
3198 }
3199 munmap(p, page_size);
3200 }
3202 if (warn && !result) {
3203 warning("HugeTLBFS is not supported by the operating system.");
3204 }
3206 return result;
3207 }
3209 /*
3210 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3211 *
3212 * From the coredump_filter documentation:
3213 *
3214 * - (bit 0) anonymous private memory
3215 * - (bit 1) anonymous shared memory
3216 * - (bit 2) file-backed private memory
3217 * - (bit 3) file-backed shared memory
3218 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3219 * effective only if the bit 2 is cleared)
3220 * - (bit 5) hugetlb private memory
3221 * - (bit 6) hugetlb shared memory
3222 */
3223 static void set_coredump_filter(void) {
3224 FILE *f;
3225 long cdm;
3227 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3228 return;
3229 }
3231 if (fscanf(f, "%lx", &cdm) != 1) {
3232 fclose(f);
3233 return;
3234 }
3236 rewind(f);
3238 if ((cdm & LARGEPAGES_BIT) == 0) {
3239 cdm |= LARGEPAGES_BIT;
3240 fprintf(f, "%#lx", cdm);
3241 }
3243 fclose(f);
3244 }
3246 // Large page support
3248 static size_t _large_page_size = 0;
3250 size_t os::Linux::find_large_page_size() {
3251 size_t large_page_size = 0;
3253 // large_page_size on Linux is used to round up heap size. x86 uses either
3254 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3255 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3256 // page as large as 256M.
3257 //
3258 // Here we try to figure out page size by parsing /proc/meminfo and looking
3259 // for a line with the following format:
3260 // Hugepagesize: 2048 kB
3261 //
3262 // If we can't determine the value (e.g. /proc is not mounted, or the text
3263 // format has been changed), we'll use the largest page size supported by
3264 // the processor.
3266 #ifndef ZERO
3267 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3268 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3269 #endif // ZERO
3271 FILE *fp = fopen("/proc/meminfo", "r");
3272 if (fp) {
3273 while (!feof(fp)) {
3274 int x = 0;
3275 char buf[16];
3276 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3277 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3278 large_page_size = x * K;
3279 break;
3280 }
3281 } else {
3282 // skip to next line
3283 for (;;) {
3284 int ch = fgetc(fp);
3285 if (ch == EOF || ch == (int)'\n') break;
3286 }
3287 }
3288 }
3289 fclose(fp);
3290 }
3292 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3293 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3294 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3295 proper_unit_for_byte_size(large_page_size));
3296 }
3298 return large_page_size;
3299 }
3301 size_t os::Linux::setup_large_page_size() {
3302 _large_page_size = Linux::find_large_page_size();
3303 const size_t default_page_size = (size_t)Linux::page_size();
3304 if (_large_page_size > default_page_size) {
3305 _page_sizes[0] = _large_page_size;
3306 _page_sizes[1] = default_page_size;
3307 _page_sizes[2] = 0;
3308 }
3310 return _large_page_size;
3311 }
3313 bool os::Linux::setup_large_page_type(size_t page_size) {
3314 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3315 FLAG_IS_DEFAULT(UseSHM) &&
3316 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3318 // The type of large pages has not been specified by the user.
3320 // Try UseHugeTLBFS and then UseSHM.
3321 UseHugeTLBFS = UseSHM = true;
3323 // Don't try UseTransparentHugePages since there are known
3324 // performance issues with it turned on. This might change in the future.
3325 UseTransparentHugePages = false;
3326 }
3328 if (UseTransparentHugePages) {
3329 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3330 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3331 UseHugeTLBFS = false;
3332 UseSHM = false;
3333 return true;
3334 }
3335 UseTransparentHugePages = false;
3336 }
3338 if (UseHugeTLBFS) {
3339 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3340 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3341 UseSHM = false;
3342 return true;
3343 }
3344 UseHugeTLBFS = false;
3345 }
3347 return UseSHM;
3348 }
3350 void os::large_page_init() {
3351 if (!UseLargePages &&
3352 !UseTransparentHugePages &&
3353 !UseHugeTLBFS &&
3354 !UseSHM) {
3355 // Not using large pages.
3356 return;
3357 }
3359 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3360 // The user explicitly turned off large pages.
3361 // Ignore the rest of the large pages flags.
3362 UseTransparentHugePages = false;
3363 UseHugeTLBFS = false;
3364 UseSHM = false;
3365 return;
3366 }
3368 size_t large_page_size = Linux::setup_large_page_size();
3369 UseLargePages = Linux::setup_large_page_type(large_page_size);
3371 set_coredump_filter();
3372 }
3374 #ifndef SHM_HUGETLB
3375 #define SHM_HUGETLB 04000
3376 #endif
3378 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3379 // "exec" is passed in but not used. Creating the shared image for
3380 // the code cache doesn't have an SHM_X executable permission to check.
3381 assert(UseLargePages && UseSHM, "only for SHM large pages");
3382 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3384 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3385 return NULL; // Fallback to small pages.
3386 }
3388 key_t key = IPC_PRIVATE;
3389 char *addr;
3391 bool warn_on_failure = UseLargePages &&
3392 (!FLAG_IS_DEFAULT(UseLargePages) ||
3393 !FLAG_IS_DEFAULT(UseSHM) ||
3394 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3395 );
3396 char msg[128];
3398 // Create a large shared memory region to attach to based on size.
3399 // Currently, size is the total size of the heap
3400 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3401 if (shmid == -1) {
3402 // Possible reasons for shmget failure:
3403 // 1. shmmax is too small for Java heap.
3404 // > check shmmax value: cat /proc/sys/kernel/shmmax
3405 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3406 // 2. not enough large page memory.
3407 // > check available large pages: cat /proc/meminfo
3408 // > increase amount of large pages:
3409 // echo new_value > /proc/sys/vm/nr_hugepages
3410 // Note 1: different Linux may use different name for this property,
3411 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3412 // Note 2: it's possible there's enough physical memory available but
3413 // they are so fragmented after a long run that they can't
3414 // coalesce into large pages. Try to reserve large pages when
3415 // the system is still "fresh".
3416 if (warn_on_failure) {
3417 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3418 warning("%s", msg);
3419 }
3420 return NULL;
3421 }
3423 // attach to the region
3424 addr = (char*)shmat(shmid, req_addr, 0);
3425 int err = errno;
3427 // Remove shmid. If shmat() is successful, the actual shared memory segment
3428 // will be deleted when it's detached by shmdt() or when the process
3429 // terminates. If shmat() is not successful this will remove the shared
3430 // segment immediately.
3431 shmctl(shmid, IPC_RMID, NULL);
3433 if ((intptr_t)addr == -1) {
3434 if (warn_on_failure) {
3435 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3436 warning("%s", msg);
3437 }
3438 return NULL;
3439 }
3441 return addr;
3442 }
3444 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3445 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3447 bool warn_on_failure = UseLargePages &&
3448 (!FLAG_IS_DEFAULT(UseLargePages) ||
3449 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3450 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3452 if (warn_on_failure) {
3453 char msg[128];
3454 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3455 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3456 warning("%s", msg);
3457 }
3458 }
3460 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3461 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3462 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3463 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3465 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3466 char* addr = (char*)::mmap(req_addr, bytes, prot,
3467 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3468 -1, 0);
3470 if (addr == MAP_FAILED) {
3471 warn_on_large_pages_failure(req_addr, bytes, errno);
3472 return NULL;
3473 }
3475 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3477 return addr;
3478 }
3480 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3481 size_t large_page_size = os::large_page_size();
3483 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3485 // Allocate small pages.
3487 char* start;
3488 if (req_addr != NULL) {
3489 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3490 assert(is_size_aligned(bytes, alignment), "Must be");
3491 start = os::reserve_memory(bytes, req_addr);
3492 assert(start == NULL || start == req_addr, "Must be");
3493 } else {
3494 start = os::reserve_memory_aligned(bytes, alignment);
3495 }
3497 if (start == NULL) {
3498 return NULL;
3499 }
3501 assert(is_ptr_aligned(start, alignment), "Must be");
3503 // os::reserve_memory_special will record this memory area.
3504 // Need to release it here to prevent overlapping reservations.
3505 MemTracker::record_virtual_memory_release((address)start, bytes);
3507 char* end = start + bytes;
3509 // Find the regions of the allocated chunk that can be promoted to large pages.
3510 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3511 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3513 size_t lp_bytes = lp_end - lp_start;
3515 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3517 if (lp_bytes == 0) {
3518 // The mapped region doesn't even span the start and the end of a large page.
3519 // Fall back to allocate a non-special area.
3520 ::munmap(start, end - start);
3521 return NULL;
3522 }
3524 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3527 void* result;
3529 if (start != lp_start) {
3530 result = ::mmap(start, lp_start - start, prot,
3531 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3532 -1, 0);
3533 if (result == MAP_FAILED) {
3534 ::munmap(lp_start, end - lp_start);
3535 return NULL;
3536 }
3537 }
3539 result = ::mmap(lp_start, lp_bytes, prot,
3540 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3541 -1, 0);
3542 if (result == MAP_FAILED) {
3543 warn_on_large_pages_failure(req_addr, bytes, errno);
3544 // If the mmap above fails, the large pages region will be unmapped and we
3545 // have regions before and after with small pages. Release these regions.
3546 //
3547 // | mapped | unmapped | mapped |
3548 // ^ ^ ^ ^
3549 // start lp_start lp_end end
3550 //
3551 ::munmap(start, lp_start - start);
3552 ::munmap(lp_end, end - lp_end);
3553 return NULL;
3554 }
3556 if (lp_end != end) {
3557 result = ::mmap(lp_end, end - lp_end, prot,
3558 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3559 -1, 0);
3560 if (result == MAP_FAILED) {
3561 ::munmap(start, lp_end - start);
3562 return NULL;
3563 }
3564 }
3566 return start;
3567 }
3569 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3570 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3571 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3572 assert(is_power_of_2(alignment), "Must be");
3573 assert(is_power_of_2(os::large_page_size()), "Must be");
3574 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3576 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3577 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3578 } else {
3579 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3580 }
3581 }
3583 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3584 assert(UseLargePages, "only for large pages");
3586 char* addr;
3587 if (UseSHM) {
3588 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3589 } else {
3590 assert(UseHugeTLBFS, "must be");
3591 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3592 }
3594 if (addr != NULL) {
3595 if (UseNUMAInterleaving) {
3596 numa_make_global(addr, bytes);
3597 }
3599 // The memory is committed
3600 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3601 }
3603 return addr;
3604 }
3606 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3607 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3608 return shmdt(base) == 0;
3609 }
3611 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3612 return pd_release_memory(base, bytes);
3613 }
3615 bool os::release_memory_special(char* base, size_t bytes) {
3616 assert(UseLargePages, "only for large pages");
3618 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3620 bool res;
3621 if (UseSHM) {
3622 res = os::Linux::release_memory_special_shm(base, bytes);
3623 } else {
3624 assert(UseHugeTLBFS, "must be");
3625 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3626 }
3628 if (res) {
3629 tkr.record((address)base, bytes);
3630 } else {
3631 tkr.discard();
3632 }
3634 return res;
3635 }
3637 size_t os::large_page_size() {
3638 return _large_page_size;
3639 }
3641 // With SysV SHM the entire memory region must be allocated as shared
3642 // memory.
3643 // HugeTLBFS allows application to commit large page memory on demand.
3644 // However, when committing memory with HugeTLBFS fails, the region
3645 // that was supposed to be committed will lose the old reservation
3646 // and allow other threads to steal that memory region. Because of this
3647 // behavior we can't commit HugeTLBFS memory.
3648 bool os::can_commit_large_page_memory() {
3649 return UseTransparentHugePages;
3650 }
3652 bool os::can_execute_large_page_memory() {
3653 return UseTransparentHugePages || UseHugeTLBFS;
3654 }
3656 // Reserve memory at an arbitrary address, only if that area is
3657 // available (and not reserved for something else).
3659 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3660 const int max_tries = 10;
3661 char* base[max_tries];
3662 size_t size[max_tries];
3663 const size_t gap = 0x000000;
3665 // Assert only that the size is a multiple of the page size, since
3666 // that's all that mmap requires, and since that's all we really know
3667 // about at this low abstraction level. If we need higher alignment,
3668 // we can either pass an alignment to this method or verify alignment
3669 // in one of the methods further up the call chain. See bug 5044738.
3670 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3672 // Repeatedly allocate blocks until the block is allocated at the
3673 // right spot. Give up after max_tries. Note that reserve_memory() will
3674 // automatically update _highest_vm_reserved_address if the call is
3675 // successful. The variable tracks the highest memory address every reserved
3676 // by JVM. It is used to detect heap-stack collision if running with
3677 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3678 // space than needed, it could confuse the collision detecting code. To
3679 // solve the problem, save current _highest_vm_reserved_address and
3680 // calculate the correct value before return.
3681 address old_highest = _highest_vm_reserved_address;
3683 // Linux mmap allows caller to pass an address as hint; give it a try first,
3684 // if kernel honors the hint then we can return immediately.
3685 char * addr = anon_mmap(requested_addr, bytes, false);
3686 if (addr == requested_addr) {
3687 return requested_addr;
3688 }
3690 if (addr != NULL) {
3691 // mmap() is successful but it fails to reserve at the requested address
3692 anon_munmap(addr, bytes);
3693 }
3695 int i;
3696 for (i = 0; i < max_tries; ++i) {
3697 base[i] = reserve_memory(bytes);
3699 if (base[i] != NULL) {
3700 // Is this the block we wanted?
3701 if (base[i] == requested_addr) {
3702 size[i] = bytes;
3703 break;
3704 }
3706 // Does this overlap the block we wanted? Give back the overlapped
3707 // parts and try again.
3709 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3710 if (top_overlap >= 0 && top_overlap < bytes) {
3711 unmap_memory(base[i], top_overlap);
3712 base[i] += top_overlap;
3713 size[i] = bytes - top_overlap;
3714 } else {
3715 size_t bottom_overlap = base[i] + bytes - requested_addr;
3716 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3717 unmap_memory(requested_addr, bottom_overlap);
3718 size[i] = bytes - bottom_overlap;
3719 } else {
3720 size[i] = bytes;
3721 }
3722 }
3723 }
3724 }
3726 // Give back the unused reserved pieces.
3728 for (int j = 0; j < i; ++j) {
3729 if (base[j] != NULL) {
3730 unmap_memory(base[j], size[j]);
3731 }
3732 }
3734 if (i < max_tries) {
3735 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3736 return requested_addr;
3737 } else {
3738 _highest_vm_reserved_address = old_highest;
3739 return NULL;
3740 }
3741 }
3743 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3744 return ::read(fd, buf, nBytes);
3745 }
3747 // TODO-FIXME: reconcile Solaris' os::sleep with the linux variation.
3748 // Solaris uses poll(), linux uses park().
3749 // Poll() is likely a better choice, assuming that Thread.interrupt()
3750 // generates a SIGUSRx signal. Note that SIGUSR1 can interfere with
3751 // SIGSEGV, see 4355769.
3753 int os::sleep(Thread* thread, jlong millis, bool interruptible) {
3754 assert(thread == Thread::current(), "thread consistency check");
3756 ParkEvent * const slp = thread->_SleepEvent ;
3757 slp->reset() ;
3758 OrderAccess::fence() ;
3760 if (interruptible) {
3761 jlong prevtime = javaTimeNanos();
3763 for (;;) {
3764 if (os::is_interrupted(thread, true)) {
3765 return OS_INTRPT;
3766 }
3768 jlong newtime = javaTimeNanos();
3770 if (newtime - prevtime < 0) {
3771 // time moving backwards, should only happen if no monotonic clock
3772 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3773 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3774 } else {
3775 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3776 }
3778 if(millis <= 0) {
3779 return OS_OK;
3780 }
3782 prevtime = newtime;
3784 {
3785 assert(thread->is_Java_thread(), "sanity check");
3786 JavaThread *jt = (JavaThread *) thread;
3787 ThreadBlockInVM tbivm(jt);
3788 OSThreadWaitState osts(jt->osthread(), false /* not Object.wait() */);
3790 jt->set_suspend_equivalent();
3791 // cleared by handle_special_suspend_equivalent_condition() or
3792 // java_suspend_self() via check_and_wait_while_suspended()
3794 slp->park(millis);
3796 // were we externally suspended while we were waiting?
3797 jt->check_and_wait_while_suspended();
3798 }
3799 }
3800 } else {
3801 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
3802 jlong prevtime = javaTimeNanos();
3804 for (;;) {
3805 // It'd be nice to avoid the back-to-back javaTimeNanos() calls on
3806 // the 1st iteration ...
3807 jlong newtime = javaTimeNanos();
3809 if (newtime - prevtime < 0) {
3810 // time moving backwards, should only happen if no monotonic clock
3811 // not a guarantee() because JVM should not abort on kernel/glibc bugs
3812 assert(!Linux::supports_monotonic_clock(), "time moving backwards");
3813 } else {
3814 millis -= (newtime - prevtime) / NANOSECS_PER_MILLISEC;
3815 }
3817 if(millis <= 0) break ;
3819 prevtime = newtime;
3820 slp->park(millis);
3821 }
3822 return OS_OK ;
3823 }
3824 }
3826 //
3827 // Short sleep, direct OS call.
3828 //
3829 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3830 // sched_yield(2) will actually give up the CPU:
3831 //
3832 // * Alone on this pariticular CPU, keeps running.
3833 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3834 // (pre 2.6.39).
3835 //
3836 // So calling this with 0 is an alternative.
3837 //
3838 void os::naked_short_sleep(jlong ms) {
3839 struct timespec req;
3841 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3842 req.tv_sec = 0;
3843 if (ms > 0) {
3844 req.tv_nsec = (ms % 1000) * 1000000;
3845 }
3846 else {
3847 req.tv_nsec = 1;
3848 }
3850 nanosleep(&req, NULL);
3852 return;
3853 }
3855 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3856 void os::infinite_sleep() {
3857 while (true) { // sleep forever ...
3858 ::sleep(100); // ... 100 seconds at a time
3859 }
3860 }
3862 // Used to convert frequent JVM_Yield() to nops
3863 bool os::dont_yield() {
3864 return DontYieldALot;
3865 }
3867 void os::yield() {
3868 sched_yield();
3869 }
3871 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3873 void os::yield_all(int attempts) {
3874 // Yields to all threads, including threads with lower priorities
3875 // Threads on Linux are all with same priority. The Solaris style
3876 // os::yield_all() with nanosleep(1ms) is not necessary.
3877 sched_yield();
3878 }
3880 // Called from the tight loops to possibly influence time-sharing heuristics
3881 void os::loop_breaker(int attempts) {
3882 os::yield_all(attempts);
3883 }
3885 ////////////////////////////////////////////////////////////////////////////////
3886 // thread priority support
3888 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3889 // only supports dynamic priority, static priority must be zero. For real-time
3890 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3891 // However, for large multi-threaded applications, SCHED_RR is not only slower
3892 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3893 // of 5 runs - Sep 2005).
3894 //
3895 // The following code actually changes the niceness of kernel-thread/LWP. It
3896 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3897 // not the entire user process, and user level threads are 1:1 mapped to kernel
3898 // threads. It has always been the case, but could change in the future. For
3899 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3900 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3902 int os::java_to_os_priority[CriticalPriority + 1] = {
3903 19, // 0 Entry should never be used
3905 4, // 1 MinPriority
3906 3, // 2
3907 2, // 3
3909 1, // 4
3910 0, // 5 NormPriority
3911 -1, // 6
3913 -2, // 7
3914 -3, // 8
3915 -4, // 9 NearMaxPriority
3917 -5, // 10 MaxPriority
3919 -5 // 11 CriticalPriority
3920 };
3922 static int prio_init() {
3923 if (ThreadPriorityPolicy == 1) {
3924 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3925 // if effective uid is not root. Perhaps, a more elegant way of doing
3926 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3927 if (geteuid() != 0) {
3928 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3929 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3930 }
3931 ThreadPriorityPolicy = 0;
3932 }
3933 }
3934 if (UseCriticalJavaThreadPriority) {
3935 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3936 }
3937 return 0;
3938 }
3940 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3941 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3943 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3944 return (ret == 0) ? OS_OK : OS_ERR;
3945 }
3947 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3948 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3949 *priority_ptr = java_to_os_priority[NormPriority];
3950 return OS_OK;
3951 }
3953 errno = 0;
3954 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3955 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3956 }
3958 // Hint to the underlying OS that a task switch would not be good.
3959 // Void return because it's a hint and can fail.
3960 void os::hint_no_preempt() {}
3962 ////////////////////////////////////////////////////////////////////////////////
3963 // suspend/resume support
3965 // the low-level signal-based suspend/resume support is a remnant from the
3966 // old VM-suspension that used to be for java-suspension, safepoints etc,
3967 // within hotspot. Now there is a single use-case for this:
3968 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3969 // that runs in the watcher thread.
3970 // The remaining code is greatly simplified from the more general suspension
3971 // code that used to be used.
3972 //
3973 // The protocol is quite simple:
3974 // - suspend:
3975 // - sends a signal to the target thread
3976 // - polls the suspend state of the osthread using a yield loop
3977 // - target thread signal handler (SR_handler) sets suspend state
3978 // and blocks in sigsuspend until continued
3979 // - resume:
3980 // - sets target osthread state to continue
3981 // - sends signal to end the sigsuspend loop in the SR_handler
3982 //
3983 // Note that the SR_lock plays no role in this suspend/resume protocol.
3984 //
3986 static void resume_clear_context(OSThread *osthread) {
3987 osthread->set_ucontext(NULL);
3988 osthread->set_siginfo(NULL);
3989 }
3991 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3992 osthread->set_ucontext(context);
3993 osthread->set_siginfo(siginfo);
3994 }
3996 //
3997 // Handler function invoked when a thread's execution is suspended or
3998 // resumed. We have to be careful that only async-safe functions are
3999 // called here (Note: most pthread functions are not async safe and
4000 // should be avoided.)
4001 //
4002 // Note: sigwait() is a more natural fit than sigsuspend() from an
4003 // interface point of view, but sigwait() prevents the signal hander
4004 // from being run. libpthread would get very confused by not having
4005 // its signal handlers run and prevents sigwait()'s use with the
4006 // mutex granting granting signal.
4007 //
4008 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
4009 //
4010 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
4011 // Save and restore errno to avoid confusing native code with EINTR
4012 // after sigsuspend.
4013 int old_errno = errno;
4015 Thread* thread = Thread::current();
4016 OSThread* osthread = thread->osthread();
4017 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
4019 os::SuspendResume::State current = osthread->sr.state();
4020 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
4021 suspend_save_context(osthread, siginfo, context);
4023 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
4024 os::SuspendResume::State state = osthread->sr.suspended();
4025 if (state == os::SuspendResume::SR_SUSPENDED) {
4026 sigset_t suspend_set; // signals for sigsuspend()
4028 // get current set of blocked signals and unblock resume signal
4029 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4030 sigdelset(&suspend_set, SR_signum);
4032 sr_semaphore.signal();
4033 // wait here until we are resumed
4034 while (1) {
4035 sigsuspend(&suspend_set);
4037 os::SuspendResume::State result = osthread->sr.running();
4038 if (result == os::SuspendResume::SR_RUNNING) {
4039 sr_semaphore.signal();
4040 break;
4041 }
4042 }
4044 } else if (state == os::SuspendResume::SR_RUNNING) {
4045 // request was cancelled, continue
4046 } else {
4047 ShouldNotReachHere();
4048 }
4050 resume_clear_context(osthread);
4051 } else if (current == os::SuspendResume::SR_RUNNING) {
4052 // request was cancelled, continue
4053 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4054 // ignore
4055 } else {
4056 // ignore
4057 }
4059 errno = old_errno;
4060 }
4063 static int SR_initialize() {
4064 struct sigaction act;
4065 char *s;
4066 /* Get signal number to use for suspend/resume */
4067 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4068 int sig = ::strtol(s, 0, 10);
4069 if (sig > 0 || sig < _NSIG) {
4070 SR_signum = sig;
4071 }
4072 }
4074 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4075 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4077 sigemptyset(&SR_sigset);
4078 sigaddset(&SR_sigset, SR_signum);
4080 /* Set up signal handler for suspend/resume */
4081 act.sa_flags = SA_RESTART|SA_SIGINFO;
4082 act.sa_handler = (void (*)(int)) SR_handler;
4084 // SR_signum is blocked by default.
4085 // 4528190 - We also need to block pthread restart signal (32 on all
4086 // supported Linux platforms). Note that LinuxThreads need to block
4087 // this signal for all threads to work properly. So we don't have
4088 // to use hard-coded signal number when setting up the mask.
4089 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4091 if (sigaction(SR_signum, &act, 0) == -1) {
4092 return -1;
4093 }
4095 // Save signal flag
4096 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4097 return 0;
4098 }
4100 static int sr_notify(OSThread* osthread) {
4101 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4102 assert_status(status == 0, status, "pthread_kill");
4103 return status;
4104 }
4106 // "Randomly" selected value for how long we want to spin
4107 // before bailing out on suspending a thread, also how often
4108 // we send a signal to a thread we want to resume
4109 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4110 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4112 // returns true on success and false on error - really an error is fatal
4113 // but this seems the normal response to library errors
4114 static bool do_suspend(OSThread* osthread) {
4115 assert(osthread->sr.is_running(), "thread should be running");
4116 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4118 // mark as suspended and send signal
4119 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4120 // failed to switch, state wasn't running?
4121 ShouldNotReachHere();
4122 return false;
4123 }
4125 if (sr_notify(osthread) != 0) {
4126 ShouldNotReachHere();
4127 }
4129 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4130 while (true) {
4131 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4132 break;
4133 } else {
4134 // timeout
4135 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4136 if (cancelled == os::SuspendResume::SR_RUNNING) {
4137 return false;
4138 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4139 // make sure that we consume the signal on the semaphore as well
4140 sr_semaphore.wait();
4141 break;
4142 } else {
4143 ShouldNotReachHere();
4144 return false;
4145 }
4146 }
4147 }
4149 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4150 return true;
4151 }
4153 static void do_resume(OSThread* osthread) {
4154 assert(osthread->sr.is_suspended(), "thread should be suspended");
4155 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4157 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4158 // failed to switch to WAKEUP_REQUEST
4159 ShouldNotReachHere();
4160 return;
4161 }
4163 while (true) {
4164 if (sr_notify(osthread) == 0) {
4165 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4166 if (osthread->sr.is_running()) {
4167 return;
4168 }
4169 }
4170 } else {
4171 ShouldNotReachHere();
4172 }
4173 }
4175 guarantee(osthread->sr.is_running(), "Must be running!");
4176 }
4178 ////////////////////////////////////////////////////////////////////////////////
4179 // interrupt support
4181 void os::interrupt(Thread* thread) {
4182 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4183 "possibility of dangling Thread pointer");
4185 OSThread* osthread = thread->osthread();
4187 if (!osthread->interrupted()) {
4188 osthread->set_interrupted(true);
4189 // More than one thread can get here with the same value of osthread,
4190 // resulting in multiple notifications. We do, however, want the store
4191 // to interrupted() to be visible to other threads before we execute unpark().
4192 OrderAccess::fence();
4193 ParkEvent * const slp = thread->_SleepEvent ;
4194 if (slp != NULL) slp->unpark() ;
4195 }
4197 // For JSR166. Unpark even if interrupt status already was set
4198 if (thread->is_Java_thread())
4199 ((JavaThread*)thread)->parker()->unpark();
4201 ParkEvent * ev = thread->_ParkEvent ;
4202 if (ev != NULL) ev->unpark() ;
4204 }
4206 bool os::is_interrupted(Thread* thread, bool clear_interrupted) {
4207 assert(Thread::current() == thread || Threads_lock->owned_by_self(),
4208 "possibility of dangling Thread pointer");
4210 OSThread* osthread = thread->osthread();
4212 bool interrupted = osthread->interrupted();
4214 if (interrupted && clear_interrupted) {
4215 osthread->set_interrupted(false);
4216 // consider thread->_SleepEvent->reset() ... optional optimization
4217 }
4219 return interrupted;
4220 }
4222 ///////////////////////////////////////////////////////////////////////////////////
4223 // signal handling (except suspend/resume)
4225 // This routine may be used by user applications as a "hook" to catch signals.
4226 // The user-defined signal handler must pass unrecognized signals to this
4227 // routine, and if it returns true (non-zero), then the signal handler must
4228 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4229 // routine will never retun false (zero), but instead will execute a VM panic
4230 // routine kill the process.
4231 //
4232 // If this routine returns false, it is OK to call it again. This allows
4233 // the user-defined signal handler to perform checks either before or after
4234 // the VM performs its own checks. Naturally, the user code would be making
4235 // a serious error if it tried to handle an exception (such as a null check
4236 // or breakpoint) that the VM was generating for its own correct operation.
4237 //
4238 // This routine may recognize any of the following kinds of signals:
4239 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4240 // It should be consulted by handlers for any of those signals.
4241 //
4242 // The caller of this routine must pass in the three arguments supplied
4243 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4244 // field of the structure passed to sigaction(). This routine assumes that
4245 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4246 //
4247 // Note that the VM will print warnings if it detects conflicting signal
4248 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4249 //
4250 extern "C" JNIEXPORT int
4251 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4252 void* ucontext, int abort_if_unrecognized);
4254 void signalHandler(int sig, siginfo_t* info, void* uc) {
4255 assert(info != NULL && uc != NULL, "it must be old kernel");
4256 int orig_errno = errno; // Preserve errno value over signal handler.
4257 JVM_handle_linux_signal(sig, info, uc, true);
4258 errno = orig_errno;
4259 }
4262 // This boolean allows users to forward their own non-matching signals
4263 // to JVM_handle_linux_signal, harmlessly.
4264 bool os::Linux::signal_handlers_are_installed = false;
4266 // For signal-chaining
4267 struct sigaction os::Linux::sigact[MAXSIGNUM];
4268 unsigned int os::Linux::sigs = 0;
4269 bool os::Linux::libjsig_is_loaded = false;
4270 typedef struct sigaction *(*get_signal_t)(int);
4271 get_signal_t os::Linux::get_signal_action = NULL;
4273 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4274 struct sigaction *actp = NULL;
4276 if (libjsig_is_loaded) {
4277 // Retrieve the old signal handler from libjsig
4278 actp = (*get_signal_action)(sig);
4279 }
4280 if (actp == NULL) {
4281 // Retrieve the preinstalled signal handler from jvm
4282 actp = get_preinstalled_handler(sig);
4283 }
4285 return actp;
4286 }
4288 static bool call_chained_handler(struct sigaction *actp, int sig,
4289 siginfo_t *siginfo, void *context) {
4290 // Call the old signal handler
4291 if (actp->sa_handler == SIG_DFL) {
4292 // It's more reasonable to let jvm treat it as an unexpected exception
4293 // instead of taking the default action.
4294 return false;
4295 } else if (actp->sa_handler != SIG_IGN) {
4296 if ((actp->sa_flags & SA_NODEFER) == 0) {
4297 // automaticlly block the signal
4298 sigaddset(&(actp->sa_mask), sig);
4299 }
4301 sa_handler_t hand;
4302 sa_sigaction_t sa;
4303 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4304 // retrieve the chained handler
4305 if (siginfo_flag_set) {
4306 sa = actp->sa_sigaction;
4307 } else {
4308 hand = actp->sa_handler;
4309 }
4311 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4312 actp->sa_handler = SIG_DFL;
4313 }
4315 // try to honor the signal mask
4316 sigset_t oset;
4317 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4319 // call into the chained handler
4320 if (siginfo_flag_set) {
4321 (*sa)(sig, siginfo, context);
4322 } else {
4323 (*hand)(sig);
4324 }
4326 // restore the signal mask
4327 pthread_sigmask(SIG_SETMASK, &oset, 0);
4328 }
4329 // Tell jvm's signal handler the signal is taken care of.
4330 return true;
4331 }
4333 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4334 bool chained = false;
4335 // signal-chaining
4336 if (UseSignalChaining) {
4337 struct sigaction *actp = get_chained_signal_action(sig);
4338 if (actp != NULL) {
4339 chained = call_chained_handler(actp, sig, siginfo, context);
4340 }
4341 }
4342 return chained;
4343 }
4345 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4346 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4347 return &sigact[sig];
4348 }
4349 return NULL;
4350 }
4352 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4353 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4354 sigact[sig] = oldAct;
4355 sigs |= (unsigned int)1 << sig;
4356 }
4358 // for diagnostic
4359 int os::Linux::sigflags[MAXSIGNUM];
4361 int os::Linux::get_our_sigflags(int sig) {
4362 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4363 return sigflags[sig];
4364 }
4366 void os::Linux::set_our_sigflags(int sig, int flags) {
4367 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4368 sigflags[sig] = flags;
4369 }
4371 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4372 // Check for overwrite.
4373 struct sigaction oldAct;
4374 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4376 void* oldhand = oldAct.sa_sigaction
4377 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4378 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4379 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4380 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4381 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4382 if (AllowUserSignalHandlers || !set_installed) {
4383 // Do not overwrite; user takes responsibility to forward to us.
4384 return;
4385 } else if (UseSignalChaining) {
4386 // save the old handler in jvm
4387 save_preinstalled_handler(sig, oldAct);
4388 // libjsig also interposes the sigaction() call below and saves the
4389 // old sigaction on it own.
4390 } else {
4391 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4392 "%#lx for signal %d.", (long)oldhand, sig));
4393 }
4394 }
4396 struct sigaction sigAct;
4397 sigfillset(&(sigAct.sa_mask));
4398 sigAct.sa_handler = SIG_DFL;
4399 if (!set_installed) {
4400 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4401 } else {
4402 sigAct.sa_sigaction = signalHandler;
4403 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4404 }
4405 // Save flags, which are set by ours
4406 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4407 sigflags[sig] = sigAct.sa_flags;
4409 int ret = sigaction(sig, &sigAct, &oldAct);
4410 assert(ret == 0, "check");
4412 void* oldhand2 = oldAct.sa_sigaction
4413 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4414 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4415 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4416 }
4418 // install signal handlers for signals that HotSpot needs to
4419 // handle in order to support Java-level exception handling.
4421 void os::Linux::install_signal_handlers() {
4422 if (!signal_handlers_are_installed) {
4423 signal_handlers_are_installed = true;
4425 // signal-chaining
4426 typedef void (*signal_setting_t)();
4427 signal_setting_t begin_signal_setting = NULL;
4428 signal_setting_t end_signal_setting = NULL;
4429 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4430 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4431 if (begin_signal_setting != NULL) {
4432 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4433 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4434 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4435 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4436 libjsig_is_loaded = true;
4437 assert(UseSignalChaining, "should enable signal-chaining");
4438 }
4439 if (libjsig_is_loaded) {
4440 // Tell libjsig jvm is setting signal handlers
4441 (*begin_signal_setting)();
4442 }
4444 set_signal_handler(SIGSEGV, true);
4445 set_signal_handler(SIGPIPE, true);
4446 set_signal_handler(SIGBUS, true);
4447 set_signal_handler(SIGILL, true);
4448 set_signal_handler(SIGFPE, true);
4449 #if defined(PPC64)
4450 set_signal_handler(SIGTRAP, true);
4451 #endif
4452 set_signal_handler(SIGXFSZ, true);
4454 if (libjsig_is_loaded) {
4455 // Tell libjsig jvm finishes setting signal handlers
4456 (*end_signal_setting)();
4457 }
4459 // We don't activate signal checker if libjsig is in place, we trust ourselves
4460 // and if UserSignalHandler is installed all bets are off.
4461 // Log that signal checking is off only if -verbose:jni is specified.
4462 if (CheckJNICalls) {
4463 if (libjsig_is_loaded) {
4464 if (PrintJNIResolving) {
4465 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4466 }
4467 check_signals = false;
4468 }
4469 if (AllowUserSignalHandlers) {
4470 if (PrintJNIResolving) {
4471 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4472 }
4473 check_signals = false;
4474 }
4475 }
4476 }
4477 }
4479 // This is the fastest way to get thread cpu time on Linux.
4480 // Returns cpu time (user+sys) for any thread, not only for current.
4481 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4482 // It might work on 2.6.10+ with a special kernel/glibc patch.
4483 // For reference, please, see IEEE Std 1003.1-2004:
4484 // http://www.unix.org/single_unix_specification
4486 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4487 struct timespec tp;
4488 int rc = os::Linux::clock_gettime(clockid, &tp);
4489 assert(rc == 0, "clock_gettime is expected to return 0 code");
4491 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4492 }
4494 /////
4495 // glibc on Linux platform uses non-documented flag
4496 // to indicate, that some special sort of signal
4497 // trampoline is used.
4498 // We will never set this flag, and we should
4499 // ignore this flag in our diagnostic
4500 #ifdef SIGNIFICANT_SIGNAL_MASK
4501 #undef SIGNIFICANT_SIGNAL_MASK
4502 #endif
4503 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4505 static const char* get_signal_handler_name(address handler,
4506 char* buf, int buflen) {
4507 int offset;
4508 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4509 if (found) {
4510 // skip directory names
4511 const char *p1, *p2;
4512 p1 = buf;
4513 size_t len = strlen(os::file_separator());
4514 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4515 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4516 } else {
4517 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4518 }
4519 return buf;
4520 }
4522 static void print_signal_handler(outputStream* st, int sig,
4523 char* buf, size_t buflen) {
4524 struct sigaction sa;
4526 sigaction(sig, NULL, &sa);
4528 // See comment for SIGNIFICANT_SIGNAL_MASK define
4529 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4531 st->print("%s: ", os::exception_name(sig, buf, buflen));
4533 address handler = (sa.sa_flags & SA_SIGINFO)
4534 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4535 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4537 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4538 st->print("SIG_DFL");
4539 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4540 st->print("SIG_IGN");
4541 } else {
4542 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4543 }
4545 st->print(", sa_mask[0]=");
4546 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4548 address rh = VMError::get_resetted_sighandler(sig);
4549 // May be, handler was resetted by VMError?
4550 if(rh != NULL) {
4551 handler = rh;
4552 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4553 }
4555 st->print(", sa_flags=");
4556 os::Posix::print_sa_flags(st, sa.sa_flags);
4558 // Check: is it our handler?
4559 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4560 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4561 // It is our signal handler
4562 // check for flags, reset system-used one!
4563 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4564 st->print(
4565 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4566 os::Linux::get_our_sigflags(sig));
4567 }
4568 }
4569 st->cr();
4570 }
4573 #define DO_SIGNAL_CHECK(sig) \
4574 if (!sigismember(&check_signal_done, sig)) \
4575 os::Linux::check_signal_handler(sig)
4577 // This method is a periodic task to check for misbehaving JNI applications
4578 // under CheckJNI, we can add any periodic checks here
4580 void os::run_periodic_checks() {
4582 if (check_signals == false) return;
4584 // SEGV and BUS if overridden could potentially prevent
4585 // generation of hs*.log in the event of a crash, debugging
4586 // such a case can be very challenging, so we absolutely
4587 // check the following for a good measure:
4588 DO_SIGNAL_CHECK(SIGSEGV);
4589 DO_SIGNAL_CHECK(SIGILL);
4590 DO_SIGNAL_CHECK(SIGFPE);
4591 DO_SIGNAL_CHECK(SIGBUS);
4592 DO_SIGNAL_CHECK(SIGPIPE);
4593 DO_SIGNAL_CHECK(SIGXFSZ);
4594 #if defined(PPC64)
4595 DO_SIGNAL_CHECK(SIGTRAP);
4596 #endif
4598 // ReduceSignalUsage allows the user to override these handlers
4599 // see comments at the very top and jvm_solaris.h
4600 if (!ReduceSignalUsage) {
4601 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4602 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4603 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4604 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4605 }
4607 DO_SIGNAL_CHECK(SR_signum);
4608 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4609 }
4611 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4613 static os_sigaction_t os_sigaction = NULL;
4615 void os::Linux::check_signal_handler(int sig) {
4616 char buf[O_BUFLEN];
4617 address jvmHandler = NULL;
4620 struct sigaction act;
4621 if (os_sigaction == NULL) {
4622 // only trust the default sigaction, in case it has been interposed
4623 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4624 if (os_sigaction == NULL) return;
4625 }
4627 os_sigaction(sig, (struct sigaction*)NULL, &act);
4630 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4632 address thisHandler = (act.sa_flags & SA_SIGINFO)
4633 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4634 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4637 switch(sig) {
4638 case SIGSEGV:
4639 case SIGBUS:
4640 case SIGFPE:
4641 case SIGPIPE:
4642 case SIGILL:
4643 case SIGXFSZ:
4644 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4645 break;
4647 case SHUTDOWN1_SIGNAL:
4648 case SHUTDOWN2_SIGNAL:
4649 case SHUTDOWN3_SIGNAL:
4650 case BREAK_SIGNAL:
4651 jvmHandler = (address)user_handler();
4652 break;
4654 case INTERRUPT_SIGNAL:
4655 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4656 break;
4658 default:
4659 if (sig == SR_signum) {
4660 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4661 } else {
4662 return;
4663 }
4664 break;
4665 }
4667 if (thisHandler != jvmHandler) {
4668 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4669 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4670 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4671 // No need to check this sig any longer
4672 sigaddset(&check_signal_done, sig);
4673 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4674 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4675 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4676 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4677 // No need to check this sig any longer
4678 sigaddset(&check_signal_done, sig);
4679 }
4681 // Dump all the signal
4682 if (sigismember(&check_signal_done, sig)) {
4683 print_signal_handlers(tty, buf, O_BUFLEN);
4684 }
4685 }
4687 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4689 extern bool signal_name(int signo, char* buf, size_t len);
4691 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4692 if (0 < exception_code && exception_code <= SIGRTMAX) {
4693 // signal
4694 if (!signal_name(exception_code, buf, size)) {
4695 jio_snprintf(buf, size, "SIG%d", exception_code);
4696 }
4697 return buf;
4698 } else {
4699 return NULL;
4700 }
4701 }
4703 // this is called _before_ the most of global arguments have been parsed
4704 void os::init(void) {
4705 char dummy; /* used to get a guess on initial stack address */
4706 // first_hrtime = gethrtime();
4708 // With LinuxThreads the JavaMain thread pid (primordial thread)
4709 // is different than the pid of the java launcher thread.
4710 // So, on Linux, the launcher thread pid is passed to the VM
4711 // via the sun.java.launcher.pid property.
4712 // Use this property instead of getpid() if it was correctly passed.
4713 // See bug 6351349.
4714 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4716 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4718 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4720 init_random(1234567);
4722 ThreadCritical::initialize();
4724 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4725 if (Linux::page_size() == -1) {
4726 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4727 strerror(errno)));
4728 }
4729 init_page_sizes((size_t) Linux::page_size());
4731 Linux::initialize_system_info();
4733 // main_thread points to the aboriginal thread
4734 Linux::_main_thread = pthread_self();
4736 Linux::clock_init();
4737 initial_time_count = javaTimeNanos();
4739 // pthread_condattr initialization for monotonic clock
4740 int status;
4741 pthread_condattr_t* _condattr = os::Linux::condAttr();
4742 if ((status = pthread_condattr_init(_condattr)) != 0) {
4743 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4744 }
4745 // Only set the clock if CLOCK_MONOTONIC is available
4746 if (Linux::supports_monotonic_clock()) {
4747 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4748 if (status == EINVAL) {
4749 warning("Unable to use monotonic clock with relative timed-waits" \
4750 " - changes to the time-of-day clock may have adverse affects");
4751 } else {
4752 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4753 }
4754 }
4755 }
4756 // else it defaults to CLOCK_REALTIME
4758 pthread_mutex_init(&dl_mutex, NULL);
4760 // If the pagesize of the VM is greater than 8K determine the appropriate
4761 // number of initial guard pages. The user can change this with the
4762 // command line arguments, if needed.
4763 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4764 StackYellowPages = 1;
4765 StackRedPages = 1;
4766 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4767 }
4768 }
4770 // To install functions for atexit system call
4771 extern "C" {
4772 static void perfMemory_exit_helper() {
4773 perfMemory_exit();
4774 }
4775 }
4777 // this is called _after_ the global arguments have been parsed
4778 jint os::init_2(void)
4779 {
4780 Linux::fast_thread_clock_init();
4782 // Allocate a single page and mark it as readable for safepoint polling
4783 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4784 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4786 os::set_polling_page( polling_page );
4788 #ifndef PRODUCT
4789 if(Verbose && PrintMiscellaneous)
4790 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4791 #endif
4793 if (!UseMembar) {
4794 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4795 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4796 os::set_memory_serialize_page( mem_serialize_page );
4798 #ifndef PRODUCT
4799 if(Verbose && PrintMiscellaneous)
4800 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4801 #endif
4802 }
4804 // initialize suspend/resume support - must do this before signal_sets_init()
4805 if (SR_initialize() != 0) {
4806 perror("SR_initialize failed");
4807 return JNI_ERR;
4808 }
4810 Linux::signal_sets_init();
4811 Linux::install_signal_handlers();
4813 // Check minimum allowable stack size for thread creation and to initialize
4814 // the java system classes, including StackOverflowError - depends on page
4815 // size. Add a page for compiler2 recursion in main thread.
4816 // Add in 2*BytesPerWord times page size to account for VM stack during
4817 // class initialization depending on 32 or 64 bit VM.
4818 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4819 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4820 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4822 size_t threadStackSizeInBytes = ThreadStackSize * K;
4823 if (threadStackSizeInBytes != 0 &&
4824 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4825 tty->print_cr("\nThe stack size specified is too small, "
4826 "Specify at least %dk",
4827 os::Linux::min_stack_allowed/ K);
4828 return JNI_ERR;
4829 }
4831 // Make the stack size a multiple of the page size so that
4832 // the yellow/red zones can be guarded.
4833 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4834 vm_page_size()));
4836 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4838 #if defined(IA32)
4839 workaround_expand_exec_shield_cs_limit();
4840 #endif
4842 Linux::libpthread_init();
4843 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4844 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4845 Linux::glibc_version(), Linux::libpthread_version(),
4846 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4847 }
4849 if (UseNUMA) {
4850 if (!Linux::libnuma_init()) {
4851 UseNUMA = false;
4852 } else {
4853 if ((Linux::numa_max_node() < 1)) {
4854 // There's only one node(they start from 0), disable NUMA.
4855 UseNUMA = false;
4856 }
4857 }
4858 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4859 // we can make the adaptive lgrp chunk resizing work. If the user specified
4860 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4861 // disable adaptive resizing.
4862 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4863 if (FLAG_IS_DEFAULT(UseNUMA)) {
4864 UseNUMA = false;
4865 } else {
4866 if (FLAG_IS_DEFAULT(UseLargePages) &&
4867 FLAG_IS_DEFAULT(UseSHM) &&
4868 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4869 UseLargePages = false;
4870 } else {
4871 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4872 UseAdaptiveSizePolicy = false;
4873 UseAdaptiveNUMAChunkSizing = false;
4874 }
4875 }
4876 }
4877 if (!UseNUMA && ForceNUMA) {
4878 UseNUMA = true;
4879 }
4880 }
4882 if (MaxFDLimit) {
4883 // set the number of file descriptors to max. print out error
4884 // if getrlimit/setrlimit fails but continue regardless.
4885 struct rlimit nbr_files;
4886 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4887 if (status != 0) {
4888 if (PrintMiscellaneous && (Verbose || WizardMode))
4889 perror("os::init_2 getrlimit failed");
4890 } else {
4891 nbr_files.rlim_cur = nbr_files.rlim_max;
4892 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4893 if (status != 0) {
4894 if (PrintMiscellaneous && (Verbose || WizardMode))
4895 perror("os::init_2 setrlimit failed");
4896 }
4897 }
4898 }
4900 // Initialize lock used to serialize thread creation (see os::create_thread)
4901 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4903 // at-exit methods are called in the reverse order of their registration.
4904 // atexit functions are called on return from main or as a result of a
4905 // call to exit(3C). There can be only 32 of these functions registered
4906 // and atexit() does not set errno.
4908 if (PerfAllowAtExitRegistration) {
4909 // only register atexit functions if PerfAllowAtExitRegistration is set.
4910 // atexit functions can be delayed until process exit time, which
4911 // can be problematic for embedded VM situations. Embedded VMs should
4912 // call DestroyJavaVM() to assure that VM resources are released.
4914 // note: perfMemory_exit_helper atexit function may be removed in
4915 // the future if the appropriate cleanup code can be added to the
4916 // VM_Exit VMOperation's doit method.
4917 if (atexit(perfMemory_exit_helper) != 0) {
4918 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4919 }
4920 }
4922 // initialize thread priority policy
4923 prio_init();
4925 return JNI_OK;
4926 }
4928 // this is called at the end of vm_initialization
4929 void os::init_3(void) {
4930 #ifdef JAVASE_EMBEDDED
4931 // Start the MemNotifyThread
4932 if (LowMemoryProtection) {
4933 MemNotifyThread::start();
4934 }
4935 return;
4936 #endif
4937 }
4939 // Mark the polling page as unreadable
4940 void os::make_polling_page_unreadable(void) {
4941 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4942 fatal("Could not disable polling page");
4943 };
4945 // Mark the polling page as readable
4946 void os::make_polling_page_readable(void) {
4947 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4948 fatal("Could not enable polling page");
4949 }
4950 };
4952 int os::active_processor_count() {
4953 // Linux doesn't yet have a (official) notion of processor sets,
4954 // so just return the number of online processors.
4955 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4956 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4957 return online_cpus;
4958 }
4960 void os::set_native_thread_name(const char *name) {
4961 // Not yet implemented.
4962 return;
4963 }
4965 bool os::distribute_processes(uint length, uint* distribution) {
4966 // Not yet implemented.
4967 return false;
4968 }
4970 bool os::bind_to_processor(uint processor_id) {
4971 // Not yet implemented.
4972 return false;
4973 }
4975 ///
4977 void os::SuspendedThreadTask::internal_do_task() {
4978 if (do_suspend(_thread->osthread())) {
4979 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4980 do_task(context);
4981 do_resume(_thread->osthread());
4982 }
4983 }
4985 class PcFetcher : public os::SuspendedThreadTask {
4986 public:
4987 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4988 ExtendedPC result();
4989 protected:
4990 void do_task(const os::SuspendedThreadTaskContext& context);
4991 private:
4992 ExtendedPC _epc;
4993 };
4995 ExtendedPC PcFetcher::result() {
4996 guarantee(is_done(), "task is not done yet.");
4997 return _epc;
4998 }
5000 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
5001 Thread* thread = context.thread();
5002 OSThread* osthread = thread->osthread();
5003 if (osthread->ucontext() != NULL) {
5004 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
5005 } else {
5006 // NULL context is unexpected, double-check this is the VMThread
5007 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
5008 }
5009 }
5011 // Suspends the target using the signal mechanism and then grabs the PC before
5012 // resuming the target. Used by the flat-profiler only
5013 ExtendedPC os::get_thread_pc(Thread* thread) {
5014 // Make sure that it is called by the watcher for the VMThread
5015 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
5016 assert(thread->is_VM_thread(), "Can only be called for VMThread");
5018 PcFetcher fetcher(thread);
5019 fetcher.run();
5020 return fetcher.result();
5021 }
5023 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
5024 {
5025 if (is_NPTL()) {
5026 return pthread_cond_timedwait(_cond, _mutex, _abstime);
5027 } else {
5028 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
5029 // word back to default 64bit precision if condvar is signaled. Java
5030 // wants 53bit precision. Save and restore current value.
5031 int fpu = get_fpu_control_word();
5032 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
5033 set_fpu_control_word(fpu);
5034 return status;
5035 }
5036 }
5038 ////////////////////////////////////////////////////////////////////////////////
5039 // debug support
5041 bool os::find(address addr, outputStream* st) {
5042 Dl_info dlinfo;
5043 memset(&dlinfo, 0, sizeof(dlinfo));
5044 if (dladdr(addr, &dlinfo) != 0) {
5045 st->print(PTR_FORMAT ": ", addr);
5046 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
5047 st->print("%s+%#x", dlinfo.dli_sname,
5048 addr - (intptr_t)dlinfo.dli_saddr);
5049 } else if (dlinfo.dli_fbase != NULL) {
5050 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
5051 } else {
5052 st->print("<absolute address>");
5053 }
5054 if (dlinfo.dli_fname != NULL) {
5055 st->print(" in %s", dlinfo.dli_fname);
5056 }
5057 if (dlinfo.dli_fbase != NULL) {
5058 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5059 }
5060 st->cr();
5062 if (Verbose) {
5063 // decode some bytes around the PC
5064 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5065 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5066 address lowest = (address) dlinfo.dli_sname;
5067 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5068 if (begin < lowest) begin = lowest;
5069 Dl_info dlinfo2;
5070 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5071 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
5072 end = (address) dlinfo2.dli_saddr;
5073 Disassembler::decode(begin, end, st);
5074 }
5075 return true;
5076 }
5077 return false;
5078 }
5080 ////////////////////////////////////////////////////////////////////////////////
5081 // misc
5083 // This does not do anything on Linux. This is basically a hook for being
5084 // able to use structured exception handling (thread-local exception filters)
5085 // on, e.g., Win32.
5086 void
5087 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5088 JavaCallArguments* args, Thread* thread) {
5089 f(value, method, args, thread);
5090 }
5092 void os::print_statistics() {
5093 }
5095 int os::message_box(const char* title, const char* message) {
5096 int i;
5097 fdStream err(defaultStream::error_fd());
5098 for (i = 0; i < 78; i++) err.print_raw("=");
5099 err.cr();
5100 err.print_raw_cr(title);
5101 for (i = 0; i < 78; i++) err.print_raw("-");
5102 err.cr();
5103 err.print_raw_cr(message);
5104 for (i = 0; i < 78; i++) err.print_raw("=");
5105 err.cr();
5107 char buf[16];
5108 // Prevent process from exiting upon "read error" without consuming all CPU
5109 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5111 return buf[0] == 'y' || buf[0] == 'Y';
5112 }
5114 int os::stat(const char *path, struct stat *sbuf) {
5115 char pathbuf[MAX_PATH];
5116 if (strlen(path) > MAX_PATH - 1) {
5117 errno = ENAMETOOLONG;
5118 return -1;
5119 }
5120 os::native_path(strcpy(pathbuf, path));
5121 return ::stat(pathbuf, sbuf);
5122 }
5124 bool os::check_heap(bool force) {
5125 return true;
5126 }
5128 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
5129 return ::vsnprintf(buf, count, format, args);
5130 }
5132 // Is a (classpath) directory empty?
5133 bool os::dir_is_empty(const char* path) {
5134 DIR *dir = NULL;
5135 struct dirent *ptr;
5137 dir = opendir(path);
5138 if (dir == NULL) return true;
5140 /* Scan the directory */
5141 bool result = true;
5142 char buf[sizeof(struct dirent) + MAX_PATH];
5143 while (result && (ptr = ::readdir(dir)) != NULL) {
5144 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5145 result = false;
5146 }
5147 }
5148 closedir(dir);
5149 return result;
5150 }
5152 // This code originates from JDK's sysOpen and open64_w
5153 // from src/solaris/hpi/src/system_md.c
5155 #ifndef O_DELETE
5156 #define O_DELETE 0x10000
5157 #endif
5159 // Open a file. Unlink the file immediately after open returns
5160 // if the specified oflag has the O_DELETE flag set.
5161 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5163 int os::open(const char *path, int oflag, int mode) {
5165 if (strlen(path) > MAX_PATH - 1) {
5166 errno = ENAMETOOLONG;
5167 return -1;
5168 }
5169 int fd;
5170 int o_delete = (oflag & O_DELETE);
5171 oflag = oflag & ~O_DELETE;
5173 fd = ::open64(path, oflag, mode);
5174 if (fd == -1) return -1;
5176 //If the open succeeded, the file might still be a directory
5177 {
5178 struct stat64 buf64;
5179 int ret = ::fstat64(fd, &buf64);
5180 int st_mode = buf64.st_mode;
5182 if (ret != -1) {
5183 if ((st_mode & S_IFMT) == S_IFDIR) {
5184 errno = EISDIR;
5185 ::close(fd);
5186 return -1;
5187 }
5188 } else {
5189 ::close(fd);
5190 return -1;
5191 }
5192 }
5194 /*
5195 * All file descriptors that are opened in the JVM and not
5196 * specifically destined for a subprocess should have the
5197 * close-on-exec flag set. If we don't set it, then careless 3rd
5198 * party native code might fork and exec without closing all
5199 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5200 * UNIXProcess.c), and this in turn might:
5201 *
5202 * - cause end-of-file to fail to be detected on some file
5203 * descriptors, resulting in mysterious hangs, or
5204 *
5205 * - might cause an fopen in the subprocess to fail on a system
5206 * suffering from bug 1085341.
5207 *
5208 * (Yes, the default setting of the close-on-exec flag is a Unix
5209 * design flaw)
5210 *
5211 * See:
5212 * 1085341: 32-bit stdio routines should support file descriptors >255
5213 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5214 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5215 */
5216 #ifdef FD_CLOEXEC
5217 {
5218 int flags = ::fcntl(fd, F_GETFD);
5219 if (flags != -1)
5220 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5221 }
5222 #endif
5224 if (o_delete != 0) {
5225 ::unlink(path);
5226 }
5227 return fd;
5228 }
5231 // create binary file, rewriting existing file if required
5232 int os::create_binary_file(const char* path, bool rewrite_existing) {
5233 int oflags = O_WRONLY | O_CREAT;
5234 if (!rewrite_existing) {
5235 oflags |= O_EXCL;
5236 }
5237 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5238 }
5240 // return current position of file pointer
5241 jlong os::current_file_offset(int fd) {
5242 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5243 }
5245 // move file pointer to the specified offset
5246 jlong os::seek_to_file_offset(int fd, jlong offset) {
5247 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5248 }
5250 // This code originates from JDK's sysAvailable
5251 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5253 int os::available(int fd, jlong *bytes) {
5254 jlong cur, end;
5255 int mode;
5256 struct stat64 buf64;
5258 if (::fstat64(fd, &buf64) >= 0) {
5259 mode = buf64.st_mode;
5260 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5261 /*
5262 * XXX: is the following call interruptible? If so, this might
5263 * need to go through the INTERRUPT_IO() wrapper as for other
5264 * blocking, interruptible calls in this file.
5265 */
5266 int n;
5267 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5268 *bytes = n;
5269 return 1;
5270 }
5271 }
5272 }
5273 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5274 return 0;
5275 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5276 return 0;
5277 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5278 return 0;
5279 }
5280 *bytes = end - cur;
5281 return 1;
5282 }
5284 int os::socket_available(int fd, jint *pbytes) {
5285 // Linux doc says EINTR not returned, unlike Solaris
5286 int ret = ::ioctl(fd, FIONREAD, pbytes);
5288 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5289 // is expected to return 0 on failure and 1 on success to the jdk.
5290 return (ret < 0) ? 0 : 1;
5291 }
5293 // Map a block of memory.
5294 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5295 char *addr, size_t bytes, bool read_only,
5296 bool allow_exec) {
5297 int prot;
5298 int flags = MAP_PRIVATE;
5300 if (read_only) {
5301 prot = PROT_READ;
5302 } else {
5303 prot = PROT_READ | PROT_WRITE;
5304 }
5306 if (allow_exec) {
5307 prot |= PROT_EXEC;
5308 }
5310 if (addr != NULL) {
5311 flags |= MAP_FIXED;
5312 }
5314 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5315 fd, file_offset);
5316 if (mapped_address == MAP_FAILED) {
5317 return NULL;
5318 }
5319 return mapped_address;
5320 }
5323 // Remap a block of memory.
5324 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5325 char *addr, size_t bytes, bool read_only,
5326 bool allow_exec) {
5327 // same as map_memory() on this OS
5328 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5329 allow_exec);
5330 }
5333 // Unmap a block of memory.
5334 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5335 return munmap(addr, bytes) == 0;
5336 }
5338 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5340 static clockid_t thread_cpu_clockid(Thread* thread) {
5341 pthread_t tid = thread->osthread()->pthread_id();
5342 clockid_t clockid;
5344 // Get thread clockid
5345 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5346 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5347 return clockid;
5348 }
5350 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5351 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5352 // of a thread.
5353 //
5354 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5355 // the fast estimate available on the platform.
5357 jlong os::current_thread_cpu_time() {
5358 if (os::Linux::supports_fast_thread_cpu_time()) {
5359 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5360 } else {
5361 // return user + sys since the cost is the same
5362 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5363 }
5364 }
5366 jlong os::thread_cpu_time(Thread* thread) {
5367 // consistent with what current_thread_cpu_time() returns
5368 if (os::Linux::supports_fast_thread_cpu_time()) {
5369 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5370 } else {
5371 return slow_thread_cpu_time(thread, true /* user + sys */);
5372 }
5373 }
5375 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5376 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5377 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5378 } else {
5379 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5380 }
5381 }
5383 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5384 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5385 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5386 } else {
5387 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5388 }
5389 }
5391 //
5392 // -1 on error.
5393 //
5395 PRAGMA_DIAG_PUSH
5396 PRAGMA_FORMAT_NONLITERAL_IGNORED
5397 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5398 static bool proc_task_unchecked = true;
5399 static const char *proc_stat_path = "/proc/%d/stat";
5400 pid_t tid = thread->osthread()->thread_id();
5401 char *s;
5402 char stat[2048];
5403 int statlen;
5404 char proc_name[64];
5405 int count;
5406 long sys_time, user_time;
5407 char cdummy;
5408 int idummy;
5409 long ldummy;
5410 FILE *fp;
5412 // The /proc/<tid>/stat aggregates per-process usage on
5413 // new Linux kernels 2.6+ where NPTL is supported.
5414 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5415 // See bug 6328462.
5416 // There possibly can be cases where there is no directory
5417 // /proc/self/task, so we check its availability.
5418 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5419 // This is executed only once
5420 proc_task_unchecked = false;
5421 fp = fopen("/proc/self/task", "r");
5422 if (fp != NULL) {
5423 proc_stat_path = "/proc/self/task/%d/stat";
5424 fclose(fp);
5425 }
5426 }
5428 sprintf(proc_name, proc_stat_path, tid);
5429 fp = fopen(proc_name, "r");
5430 if ( fp == NULL ) return -1;
5431 statlen = fread(stat, 1, 2047, fp);
5432 stat[statlen] = '\0';
5433 fclose(fp);
5435 // Skip pid and the command string. Note that we could be dealing with
5436 // weird command names, e.g. user could decide to rename java launcher
5437 // to "java 1.4.2 :)", then the stat file would look like
5438 // 1234 (java 1.4.2 :)) R ... ...
5439 // We don't really need to know the command string, just find the last
5440 // occurrence of ")" and then start parsing from there. See bug 4726580.
5441 s = strrchr(stat, ')');
5442 if (s == NULL ) return -1;
5444 // Skip blank chars
5445 do s++; while (isspace(*s));
5447 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5448 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5449 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5450 &user_time, &sys_time);
5451 if ( count != 13 ) return -1;
5452 if (user_sys_cpu_time) {
5453 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5454 } else {
5455 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5456 }
5457 }
5458 PRAGMA_DIAG_POP
5460 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5461 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5462 info_ptr->may_skip_backward = false; // elapsed time not wall time
5463 info_ptr->may_skip_forward = false; // elapsed time not wall time
5464 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5465 }
5467 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5468 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5469 info_ptr->may_skip_backward = false; // elapsed time not wall time
5470 info_ptr->may_skip_forward = false; // elapsed time not wall time
5471 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5472 }
5474 bool os::is_thread_cpu_time_supported() {
5475 return true;
5476 }
5478 // System loadavg support. Returns -1 if load average cannot be obtained.
5479 // Linux doesn't yet have a (official) notion of processor sets,
5480 // so just return the system wide load average.
5481 int os::loadavg(double loadavg[], int nelem) {
5482 return ::getloadavg(loadavg, nelem);
5483 }
5485 void os::pause() {
5486 char filename[MAX_PATH];
5487 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5488 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5489 } else {
5490 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5491 }
5493 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5494 if (fd != -1) {
5495 struct stat buf;
5496 ::close(fd);
5497 while (::stat(filename, &buf) == 0) {
5498 (void)::poll(NULL, 0, 100);
5499 }
5500 } else {
5501 jio_fprintf(stderr,
5502 "Could not open pause file '%s', continuing immediately.\n", filename);
5503 }
5504 }
5507 // Refer to the comments in os_solaris.cpp park-unpark.
5508 //
5509 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5510 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5511 // For specifics regarding the bug see GLIBC BUGID 261237 :
5512 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5513 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5514 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5515 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5516 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5517 // and monitorenter when we're using 1-0 locking. All those operations may result in
5518 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5519 // of libpthread avoids the problem, but isn't practical.
5520 //
5521 // Possible remedies:
5522 //
5523 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5524 // This is palliative and probabilistic, however. If the thread is preempted
5525 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5526 // than the minimum period may have passed, and the abstime may be stale (in the
5527 // past) resultin in a hang. Using this technique reduces the odds of a hang
5528 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5529 //
5530 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5531 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5532 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5533 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5534 // thread.
5535 //
5536 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5537 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5538 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5539 // This also works well. In fact it avoids kernel-level scalability impediments
5540 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5541 // timers in a graceful fashion.
5542 //
5543 // 4. When the abstime value is in the past it appears that control returns
5544 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5545 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5546 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5547 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5548 // It may be possible to avoid reinitialization by checking the return
5549 // value from pthread_cond_timedwait(). In addition to reinitializing the
5550 // condvar we must establish the invariant that cond_signal() is only called
5551 // within critical sections protected by the adjunct mutex. This prevents
5552 // cond_signal() from "seeing" a condvar that's in the midst of being
5553 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5554 // desirable signal-after-unlock optimization that avoids futile context switching.
5555 //
5556 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5557 // structure when a condvar is used or initialized. cond_destroy() would
5558 // release the helper structure. Our reinitialize-after-timedwait fix
5559 // put excessive stress on malloc/free and locks protecting the c-heap.
5560 //
5561 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5562 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5563 // and only enabling the work-around for vulnerable environments.
5565 // utility to compute the abstime argument to timedwait:
5566 // millis is the relative timeout time
5567 // abstime will be the absolute timeout time
5568 // TODO: replace compute_abstime() with unpackTime()
5570 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5571 if (millis < 0) millis = 0;
5573 jlong seconds = millis / 1000;
5574 millis %= 1000;
5575 if (seconds > 50000000) { // see man cond_timedwait(3T)
5576 seconds = 50000000;
5577 }
5579 if (os::Linux::supports_monotonic_clock()) {
5580 struct timespec now;
5581 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5582 assert_status(status == 0, status, "clock_gettime");
5583 abstime->tv_sec = now.tv_sec + seconds;
5584 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5585 if (nanos >= NANOSECS_PER_SEC) {
5586 abstime->tv_sec += 1;
5587 nanos -= NANOSECS_PER_SEC;
5588 }
5589 abstime->tv_nsec = nanos;
5590 } else {
5591 struct timeval now;
5592 int status = gettimeofday(&now, NULL);
5593 assert(status == 0, "gettimeofday");
5594 abstime->tv_sec = now.tv_sec + seconds;
5595 long usec = now.tv_usec + millis * 1000;
5596 if (usec >= 1000000) {
5597 abstime->tv_sec += 1;
5598 usec -= 1000000;
5599 }
5600 abstime->tv_nsec = usec * 1000;
5601 }
5602 return abstime;
5603 }
5606 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5607 // Conceptually TryPark() should be equivalent to park(0).
5609 int os::PlatformEvent::TryPark() {
5610 for (;;) {
5611 const int v = _Event ;
5612 guarantee ((v == 0) || (v == 1), "invariant") ;
5613 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5614 }
5615 }
5617 void os::PlatformEvent::park() { // AKA "down()"
5618 // Invariant: Only the thread associated with the Event/PlatformEvent
5619 // may call park().
5620 // TODO: assert that _Assoc != NULL or _Assoc == Self
5621 int v ;
5622 for (;;) {
5623 v = _Event ;
5624 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5625 }
5626 guarantee (v >= 0, "invariant") ;
5627 if (v == 0) {
5628 // Do this the hard way by blocking ...
5629 int status = pthread_mutex_lock(_mutex);
5630 assert_status(status == 0, status, "mutex_lock");
5631 guarantee (_nParked == 0, "invariant") ;
5632 ++ _nParked ;
5633 while (_Event < 0) {
5634 status = pthread_cond_wait(_cond, _mutex);
5635 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5636 // Treat this the same as if the wait was interrupted
5637 if (status == ETIME) { status = EINTR; }
5638 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5639 }
5640 -- _nParked ;
5642 _Event = 0 ;
5643 status = pthread_mutex_unlock(_mutex);
5644 assert_status(status == 0, status, "mutex_unlock");
5645 // Paranoia to ensure our locked and lock-free paths interact
5646 // correctly with each other.
5647 OrderAccess::fence();
5648 }
5649 guarantee (_Event >= 0, "invariant") ;
5650 }
5652 int os::PlatformEvent::park(jlong millis) {
5653 guarantee (_nParked == 0, "invariant") ;
5655 int v ;
5656 for (;;) {
5657 v = _Event ;
5658 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5659 }
5660 guarantee (v >= 0, "invariant") ;
5661 if (v != 0) return OS_OK ;
5663 // We do this the hard way, by blocking the thread.
5664 // Consider enforcing a minimum timeout value.
5665 struct timespec abst;
5666 compute_abstime(&abst, millis);
5668 int ret = OS_TIMEOUT;
5669 int status = pthread_mutex_lock(_mutex);
5670 assert_status(status == 0, status, "mutex_lock");
5671 guarantee (_nParked == 0, "invariant") ;
5672 ++_nParked ;
5674 // Object.wait(timo) will return because of
5675 // (a) notification
5676 // (b) timeout
5677 // (c) thread.interrupt
5678 //
5679 // Thread.interrupt and object.notify{All} both call Event::set.
5680 // That is, we treat thread.interrupt as a special case of notification.
5681 // The underlying Solaris implementation, cond_timedwait, admits
5682 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5683 // JVM from making those visible to Java code. As such, we must
5684 // filter out spurious wakeups. We assume all ETIME returns are valid.
5685 //
5686 // TODO: properly differentiate simultaneous notify+interrupt.
5687 // In that case, we should propagate the notify to another waiter.
5689 while (_Event < 0) {
5690 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5691 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5692 pthread_cond_destroy (_cond);
5693 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5694 }
5695 assert_status(status == 0 || status == EINTR ||
5696 status == ETIME || status == ETIMEDOUT,
5697 status, "cond_timedwait");
5698 if (!FilterSpuriousWakeups) break ; // previous semantics
5699 if (status == ETIME || status == ETIMEDOUT) break ;
5700 // We consume and ignore EINTR and spurious wakeups.
5701 }
5702 --_nParked ;
5703 if (_Event >= 0) {
5704 ret = OS_OK;
5705 }
5706 _Event = 0 ;
5707 status = pthread_mutex_unlock(_mutex);
5708 assert_status(status == 0, status, "mutex_unlock");
5709 assert (_nParked == 0, "invariant") ;
5710 // Paranoia to ensure our locked and lock-free paths interact
5711 // correctly with each other.
5712 OrderAccess::fence();
5713 return ret;
5714 }
5716 void os::PlatformEvent::unpark() {
5717 // Transitions for _Event:
5718 // 0 :=> 1
5719 // 1 :=> 1
5720 // -1 :=> either 0 or 1; must signal target thread
5721 // That is, we can safely transition _Event from -1 to either
5722 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5723 // unpark() calls.
5724 // See also: "Semaphores in Plan 9" by Mullender & Cox
5725 //
5726 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5727 // that it will take two back-to-back park() calls for the owning
5728 // thread to block. This has the benefit of forcing a spurious return
5729 // from the first park() call after an unpark() call which will help
5730 // shake out uses of park() and unpark() without condition variables.
5732 if (Atomic::xchg(1, &_Event) >= 0) return;
5734 // Wait for the thread associated with the event to vacate
5735 int status = pthread_mutex_lock(_mutex);
5736 assert_status(status == 0, status, "mutex_lock");
5737 int AnyWaiters = _nParked;
5738 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5739 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5740 AnyWaiters = 0;
5741 pthread_cond_signal(_cond);
5742 }
5743 status = pthread_mutex_unlock(_mutex);
5744 assert_status(status == 0, status, "mutex_unlock");
5745 if (AnyWaiters != 0) {
5746 status = pthread_cond_signal(_cond);
5747 assert_status(status == 0, status, "cond_signal");
5748 }
5750 // Note that we signal() _after dropping the lock for "immortal" Events.
5751 // This is safe and avoids a common class of futile wakeups. In rare
5752 // circumstances this can cause a thread to return prematurely from
5753 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5754 // simply re-test the condition and re-park itself.
5755 }
5758 // JSR166
5759 // -------------------------------------------------------
5761 /*
5762 * The solaris and linux implementations of park/unpark are fairly
5763 * conservative for now, but can be improved. They currently use a
5764 * mutex/condvar pair, plus a a count.
5765 * Park decrements count if > 0, else does a condvar wait. Unpark
5766 * sets count to 1 and signals condvar. Only one thread ever waits
5767 * on the condvar. Contention seen when trying to park implies that someone
5768 * is unparking you, so don't wait. And spurious returns are fine, so there
5769 * is no need to track notifications.
5770 */
5772 /*
5773 * This code is common to linux and solaris and will be moved to a
5774 * common place in dolphin.
5775 *
5776 * The passed in time value is either a relative time in nanoseconds
5777 * or an absolute time in milliseconds. Either way it has to be unpacked
5778 * into suitable seconds and nanoseconds components and stored in the
5779 * given timespec structure.
5780 * Given time is a 64-bit value and the time_t used in the timespec is only
5781 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5782 * overflow if times way in the future are given. Further on Solaris versions
5783 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5784 * number of seconds, in abstime, is less than current_time + 100,000,000.
5785 * As it will be 28 years before "now + 100000000" will overflow we can
5786 * ignore overflow and just impose a hard-limit on seconds using the value
5787 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5788 * years from "now".
5789 */
5791 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5792 assert (time > 0, "convertTime");
5793 time_t max_secs = 0;
5795 if (!os::Linux::supports_monotonic_clock() || isAbsolute) {
5796 struct timeval now;
5797 int status = gettimeofday(&now, NULL);
5798 assert(status == 0, "gettimeofday");
5800 max_secs = now.tv_sec + MAX_SECS;
5802 if (isAbsolute) {
5803 jlong secs = time / 1000;
5804 if (secs > max_secs) {
5805 absTime->tv_sec = max_secs;
5806 } else {
5807 absTime->tv_sec = secs;
5808 }
5809 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5810 } else {
5811 jlong secs = time / NANOSECS_PER_SEC;
5812 if (secs >= MAX_SECS) {
5813 absTime->tv_sec = max_secs;
5814 absTime->tv_nsec = 0;
5815 } else {
5816 absTime->tv_sec = now.tv_sec + secs;
5817 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5818 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5819 absTime->tv_nsec -= NANOSECS_PER_SEC;
5820 ++absTime->tv_sec; // note: this must be <= max_secs
5821 }
5822 }
5823 }
5824 } else {
5825 // must be relative using monotonic clock
5826 struct timespec now;
5827 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5828 assert_status(status == 0, status, "clock_gettime");
5829 max_secs = now.tv_sec + MAX_SECS;
5830 jlong secs = time / NANOSECS_PER_SEC;
5831 if (secs >= MAX_SECS) {
5832 absTime->tv_sec = max_secs;
5833 absTime->tv_nsec = 0;
5834 } else {
5835 absTime->tv_sec = now.tv_sec + secs;
5836 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5837 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5838 absTime->tv_nsec -= NANOSECS_PER_SEC;
5839 ++absTime->tv_sec; // note: this must be <= max_secs
5840 }
5841 }
5842 }
5843 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5844 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5845 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5846 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5847 }
5849 void Parker::park(bool isAbsolute, jlong time) {
5850 // Ideally we'd do something useful while spinning, such
5851 // as calling unpackTime().
5853 // Optional fast-path check:
5854 // Return immediately if a permit is available.
5855 // We depend on Atomic::xchg() having full barrier semantics
5856 // since we are doing a lock-free update to _counter.
5857 if (Atomic::xchg(0, &_counter) > 0) return;
5859 Thread* thread = Thread::current();
5860 assert(thread->is_Java_thread(), "Must be JavaThread");
5861 JavaThread *jt = (JavaThread *)thread;
5863 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5864 // Check interrupt before trying to wait
5865 if (Thread::is_interrupted(thread, false)) {
5866 return;
5867 }
5869 // Next, demultiplex/decode time arguments
5870 timespec absTime;
5871 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5872 return;
5873 }
5874 if (time > 0) {
5875 unpackTime(&absTime, isAbsolute, time);
5876 }
5879 // Enter safepoint region
5880 // Beware of deadlocks such as 6317397.
5881 // The per-thread Parker:: mutex is a classic leaf-lock.
5882 // In particular a thread must never block on the Threads_lock while
5883 // holding the Parker:: mutex. If safepoints are pending both the
5884 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5885 ThreadBlockInVM tbivm(jt);
5887 // Don't wait if cannot get lock since interference arises from
5888 // unblocking. Also. check interrupt before trying wait
5889 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5890 return;
5891 }
5893 int status ;
5894 if (_counter > 0) { // no wait needed
5895 _counter = 0;
5896 status = pthread_mutex_unlock(_mutex);
5897 assert (status == 0, "invariant") ;
5898 // Paranoia to ensure our locked and lock-free paths interact
5899 // correctly with each other and Java-level accesses.
5900 OrderAccess::fence();
5901 return;
5902 }
5904 #ifdef ASSERT
5905 // Don't catch signals while blocked; let the running threads have the signals.
5906 // (This allows a debugger to break into the running thread.)
5907 sigset_t oldsigs;
5908 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5909 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5910 #endif
5912 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5913 jt->set_suspend_equivalent();
5914 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5916 assert(_cur_index == -1, "invariant");
5917 if (time == 0) {
5918 _cur_index = REL_INDEX; // arbitrary choice when not timed
5919 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5920 } else {
5921 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5922 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5923 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5924 pthread_cond_destroy (&_cond[_cur_index]) ;
5925 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5926 }
5927 }
5928 _cur_index = -1;
5929 assert_status(status == 0 || status == EINTR ||
5930 status == ETIME || status == ETIMEDOUT,
5931 status, "cond_timedwait");
5933 #ifdef ASSERT
5934 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5935 #endif
5937 _counter = 0 ;
5938 status = pthread_mutex_unlock(_mutex) ;
5939 assert_status(status == 0, status, "invariant") ;
5940 // Paranoia to ensure our locked and lock-free paths interact
5941 // correctly with each other and Java-level accesses.
5942 OrderAccess::fence();
5944 // If externally suspended while waiting, re-suspend
5945 if (jt->handle_special_suspend_equivalent_condition()) {
5946 jt->java_suspend_self();
5947 }
5948 }
5950 void Parker::unpark() {
5951 int s, status ;
5952 status = pthread_mutex_lock(_mutex);
5953 assert (status == 0, "invariant") ;
5954 s = _counter;
5955 _counter = 1;
5956 if (s < 1) {
5957 // thread might be parked
5958 if (_cur_index != -1) {
5959 // thread is definitely parked
5960 if (WorkAroundNPTLTimedWaitHang) {
5961 status = pthread_cond_signal (&_cond[_cur_index]);
5962 assert (status == 0, "invariant");
5963 status = pthread_mutex_unlock(_mutex);
5964 assert (status == 0, "invariant");
5965 } else {
5966 status = pthread_mutex_unlock(_mutex);
5967 assert (status == 0, "invariant");
5968 status = pthread_cond_signal (&_cond[_cur_index]);
5969 assert (status == 0, "invariant");
5970 }
5971 } else {
5972 pthread_mutex_unlock(_mutex);
5973 assert (status == 0, "invariant") ;
5974 }
5975 } else {
5976 pthread_mutex_unlock(_mutex);
5977 assert (status == 0, "invariant") ;
5978 }
5979 }
5982 extern char** environ;
5984 #ifndef __NR_fork
5985 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5986 #endif
5988 #ifndef __NR_execve
5989 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5990 #endif
5992 // Run the specified command in a separate process. Return its exit value,
5993 // or -1 on failure (e.g. can't fork a new process).
5994 // Unlike system(), this function can be called from signal handler. It
5995 // doesn't block SIGINT et al.
5996 int os::fork_and_exec(char* cmd) {
5997 const char * argv[4] = {"sh", "-c", cmd, NULL};
5999 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
6000 // pthread_atfork handlers and reset pthread library. All we need is a
6001 // separate process to execve. Make a direct syscall to fork process.
6002 // On IA64 there's no fork syscall, we have to use fork() and hope for
6003 // the best...
6004 pid_t pid = NOT_IA64(syscall(__NR_fork);)
6005 IA64_ONLY(fork();)
6007 if (pid < 0) {
6008 // fork failed
6009 return -1;
6011 } else if (pid == 0) {
6012 // child process
6014 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
6015 // first to kill every thread on the thread list. Because this list is
6016 // not reset by fork() (see notes above), execve() will instead kill
6017 // every thread in the parent process. We know this is the only thread
6018 // in the new process, so make a system call directly.
6019 // IA64 should use normal execve() from glibc to match the glibc fork()
6020 // above.
6021 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
6022 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
6024 // execve failed
6025 _exit(-1);
6027 } else {
6028 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
6029 // care about the actual exit code, for now.
6031 int status;
6033 // Wait for the child process to exit. This returns immediately if
6034 // the child has already exited. */
6035 while (waitpid(pid, &status, 0) < 0) {
6036 switch (errno) {
6037 case ECHILD: return 0;
6038 case EINTR: break;
6039 default: return -1;
6040 }
6041 }
6043 if (WIFEXITED(status)) {
6044 // The child exited normally; get its exit code.
6045 return WEXITSTATUS(status);
6046 } else if (WIFSIGNALED(status)) {
6047 // The child exited because of a signal
6048 // The best value to return is 0x80 + signal number,
6049 // because that is what all Unix shells do, and because
6050 // it allows callers to distinguish between process exit and
6051 // process death by signal.
6052 return 0x80 + WTERMSIG(status);
6053 } else {
6054 // Unknown exit code; pass it through
6055 return status;
6056 }
6057 }
6058 }
6060 // is_headless_jre()
6061 //
6062 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
6063 // in order to report if we are running in a headless jre
6064 //
6065 // Since JDK8 xawt/libmawt.so was moved into the same directory
6066 // as libawt.so, and renamed libawt_xawt.so
6067 //
6068 bool os::is_headless_jre() {
6069 struct stat statbuf;
6070 char buf[MAXPATHLEN];
6071 char libmawtpath[MAXPATHLEN];
6072 const char *xawtstr = "/xawt/libmawt.so";
6073 const char *new_xawtstr = "/libawt_xawt.so";
6074 char *p;
6076 // Get path to libjvm.so
6077 os::jvm_path(buf, sizeof(buf));
6079 // Get rid of libjvm.so
6080 p = strrchr(buf, '/');
6081 if (p == NULL) return false;
6082 else *p = '\0';
6084 // Get rid of client or server
6085 p = strrchr(buf, '/');
6086 if (p == NULL) return false;
6087 else *p = '\0';
6089 // check xawt/libmawt.so
6090 strcpy(libmawtpath, buf);
6091 strcat(libmawtpath, xawtstr);
6092 if (::stat(libmawtpath, &statbuf) == 0) return false;
6094 // check libawt_xawt.so
6095 strcpy(libmawtpath, buf);
6096 strcat(libmawtpath, new_xawtstr);
6097 if (::stat(libmawtpath, &statbuf) == 0) return false;
6099 return true;
6100 }
6102 // Get the default path to the core file
6103 // Returns the length of the string
6104 int os::get_core_path(char* buffer, size_t bufferSize) {
6105 const char* p = get_current_directory(buffer, bufferSize);
6107 if (p == NULL) {
6108 assert(p != NULL, "failed to get current directory");
6109 return 0;
6110 }
6112 return strlen(buffer);
6113 }
6115 #ifdef JAVASE_EMBEDDED
6116 //
6117 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
6118 //
6119 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
6121 // ctor
6122 //
6123 MemNotifyThread::MemNotifyThread(int fd): Thread() {
6124 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
6125 _fd = fd;
6127 if (os::create_thread(this, os::os_thread)) {
6128 _memnotify_thread = this;
6129 os::set_priority(this, NearMaxPriority);
6130 os::start_thread(this);
6131 }
6132 }
6134 // Where all the work gets done
6135 //
6136 void MemNotifyThread::run() {
6137 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
6139 // Set up the select arguments
6140 fd_set rfds;
6141 if (_fd != -1) {
6142 FD_ZERO(&rfds);
6143 FD_SET(_fd, &rfds);
6144 }
6146 // Now wait for the mem_notify device to wake up
6147 while (1) {
6148 // Wait for the mem_notify device to signal us..
6149 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6150 if (rc == -1) {
6151 perror("select!\n");
6152 break;
6153 } else if (rc) {
6154 //ssize_t free_before = os::available_memory();
6155 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6157 // The kernel is telling us there is not much memory left...
6158 // try to do something about that
6160 // If we are not already in a GC, try one.
6161 if (!Universe::heap()->is_gc_active()) {
6162 Universe::heap()->collect(GCCause::_allocation_failure);
6164 //ssize_t free_after = os::available_memory();
6165 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6166 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6167 }
6168 // We might want to do something like the following if we find the GC's are not helping...
6169 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6170 }
6171 }
6172 }
6174 //
6175 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6176 //
6177 void MemNotifyThread::start() {
6178 int fd;
6179 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6180 if (fd < 0) {
6181 return;
6182 }
6184 if (memnotify_thread() == NULL) {
6185 new MemNotifyThread(fd);
6186 }
6187 }
6189 #endif // JAVASE_EMBEDDED
6192 /////////////// Unit tests ///////////////
6194 #ifndef PRODUCT
6196 #define test_log(...) \
6197 do {\
6198 if (VerboseInternalVMTests) { \
6199 tty->print_cr(__VA_ARGS__); \
6200 tty->flush(); \
6201 }\
6202 } while (false)
6204 class TestReserveMemorySpecial : AllStatic {
6205 public:
6206 static void small_page_write(void* addr, size_t size) {
6207 size_t page_size = os::vm_page_size();
6209 char* end = (char*)addr + size;
6210 for (char* p = (char*)addr; p < end; p += page_size) {
6211 *p = 1;
6212 }
6213 }
6215 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6216 if (!UseHugeTLBFS) {
6217 return;
6218 }
6220 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6222 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6224 if (addr != NULL) {
6225 small_page_write(addr, size);
6227 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6228 }
6229 }
6231 static void test_reserve_memory_special_huge_tlbfs_only() {
6232 if (!UseHugeTLBFS) {
6233 return;
6234 }
6236 size_t lp = os::large_page_size();
6238 for (size_t size = lp; size <= lp * 10; size += lp) {
6239 test_reserve_memory_special_huge_tlbfs_only(size);
6240 }
6241 }
6243 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6244 if (!UseHugeTLBFS) {
6245 return;
6246 }
6248 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6249 size, alignment);
6251 assert(size >= os::large_page_size(), "Incorrect input to test");
6253 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6255 if (addr != NULL) {
6256 small_page_write(addr, size);
6258 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6259 }
6260 }
6262 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6263 size_t lp = os::large_page_size();
6264 size_t ag = os::vm_allocation_granularity();
6266 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6267 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6268 }
6269 }
6271 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6272 size_t lp = os::large_page_size();
6273 size_t ag = os::vm_allocation_granularity();
6275 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6276 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6277 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6278 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6279 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6280 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6281 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6282 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6283 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6284 }
6286 static void test_reserve_memory_special_huge_tlbfs() {
6287 if (!UseHugeTLBFS) {
6288 return;
6289 }
6291 test_reserve_memory_special_huge_tlbfs_only();
6292 test_reserve_memory_special_huge_tlbfs_mixed();
6293 }
6295 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6296 if (!UseSHM) {
6297 return;
6298 }
6300 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6302 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6304 if (addr != NULL) {
6305 assert(is_ptr_aligned(addr, alignment), "Check");
6306 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6308 small_page_write(addr, size);
6310 os::Linux::release_memory_special_shm(addr, size);
6311 }
6312 }
6314 static void test_reserve_memory_special_shm() {
6315 size_t lp = os::large_page_size();
6316 size_t ag = os::vm_allocation_granularity();
6318 for (size_t size = ag; size < lp * 3; size += ag) {
6319 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6320 test_reserve_memory_special_shm(size, alignment);
6321 }
6322 }
6323 }
6325 static void test() {
6326 test_reserve_memory_special_huge_tlbfs();
6327 test_reserve_memory_special_shm();
6328 }
6329 };
6331 void TestReserveMemorySpecial_test() {
6332 TestReserveMemorySpecial::test();
6333 }
6335 #endif