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 */
24
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"
69
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>
103
104 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
105 // getrusage() is prepared to handle the associated failure.
106 #ifndef RUSAGE_THREAD
107 #define RUSAGE_THREAD (1) /* only the calling thread */
108 #endif
109
110 #define MAX_PATH (2 * K)
111
112 // for timer info max values which include all bits
113 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
114
115 #define LARGEPAGES_BIT (1 << 6)
116 ////////////////////////////////////////////////////////////////////////////////
117 // global variables
118 julong os::Linux::_physical_memory = 0;
119
120 address os::Linux::_initial_thread_stack_bottom = NULL;
121 uintptr_t os::Linux::_initial_thread_stack_size = 0;
122
123 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
124 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
125 Mutex* os::Linux::_createThread_lock = NULL;
126 pthread_t os::Linux::_main_thread;
127 int os::Linux::_page_size = -1;
128 const int os::Linux::_vm_default_page_size = (8 * K);
129 bool os::Linux::_is_floating_stack = false;
130 bool os::Linux::_is_NPTL = false;
131 bool os::Linux::_supports_fast_thread_cpu_time = false;
132 const char * os::Linux::_glibc_version = NULL;
133 const char * os::Linux::_libpthread_version = NULL;
134 pthread_condattr_t os::Linux::_condattr[1];
135
136 static jlong initial_time_count=0;
137
138 static int clock_tics_per_sec = 100;
139
140 // For diagnostics to print a message once. see run_periodic_checks
141 static sigset_t check_signal_done;
142 static bool check_signals = true;
143
144 static pid_t _initial_pid = 0;
145
146 /* Signal number used to suspend/resume a thread */
147
148 /* do not use any signal number less than SIGSEGV, see 4355769 */
149 static int SR_signum = SIGUSR2;
150 sigset_t SR_sigset;
151
152 /* Used to protect dlsym() calls */
153 static pthread_mutex_t dl_mutex;
154
155 // Declarations
156 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
157
158 #ifdef JAVASE_EMBEDDED
159 class MemNotifyThread: public Thread {
160 friend class VMStructs;
161 public:
162 virtual void run();
163
164 private:
165 static MemNotifyThread* _memnotify_thread;
166 int _fd;
167
168 public:
169
170 // Constructor
171 MemNotifyThread(int fd);
172
173 // Tester
174 bool is_memnotify_thread() const { return true; }
175
176 // Printing
177 char* name() const { return (char*)"Linux MemNotify Thread"; }
178
179 // Returns the single instance of the MemNotifyThread
180 static MemNotifyThread* memnotify_thread() { return _memnotify_thread; }
181
182 // Create and start the single instance of MemNotifyThread
183 static void start();
184 };
185 #endif // JAVASE_EMBEDDED
186
187 // utility functions
188
189 static int SR_initialize();
190
191 julong os::available_memory() {
192 return Linux::available_memory();
193 }
194
195 julong os::Linux::available_memory() {
196 // values in struct sysinfo are "unsigned long"
197 struct sysinfo si;
198 sysinfo(&si);
199
200 return (julong)si.freeram * si.mem_unit;
201 }
202
203 julong os::physical_memory() {
204 return Linux::physical_memory();
205 }
206
207 ////////////////////////////////////////////////////////////////////////////////
208 // environment support
209
210 bool os::getenv(const char* name, char* buf, int len) {
211 const char* val = ::getenv(name);
212 if (val != NULL && strlen(val) < (size_t)len) {
213 strcpy(buf, val);
214 return true;
215 }
216 if (len > 0) buf[0] = 0; // return a null string
217 return false;
218 }
219
220
221 // Return true if user is running as root.
222
223 bool os::have_special_privileges() {
224 static bool init = false;
225 static bool privileges = false;
226 if (!init) {
227 privileges = (getuid() != geteuid()) || (getgid() != getegid());
228 init = true;
229 }
230 return privileges;
231 }
232
233
234 #ifndef SYS_gettid
235 // i386: 224, ia64: 1105, amd64: 186, sparc 143
236 #ifdef __ia64__
237 #define SYS_gettid 1105
238 #elif __i386__
239 #define SYS_gettid 224
240 #elif __amd64__
241 #define SYS_gettid 186
242 #elif __sparc__
243 #define SYS_gettid 143
244 #else
245 #error define gettid for the arch
246 #endif
247 #endif
248
249 // Cpu architecture string
250 #if defined(ZERO)
251 static char cpu_arch[] = ZERO_LIBARCH;
252 #elif defined(IA64)
253 static char cpu_arch[] = "ia64";
254 #elif defined(IA32)
255 static char cpu_arch[] = "i386";
256 #elif defined(AMD64)
257 static char cpu_arch[] = "amd64";
258 #elif defined(ARM)
259 static char cpu_arch[] = "arm";
260 #elif defined(PPC32)
261 static char cpu_arch[] = "ppc";
262 #elif defined(PPC64)
263 static char cpu_arch[] = "ppc64";
264 #elif defined(SPARC)
265 # ifdef _LP64
266 static char cpu_arch[] = "sparcv9";
267 # else
268 static char cpu_arch[] = "sparc";
269 # endif
270 #else
271 #error Add appropriate cpu_arch setting
272 #endif
273
274
275 // pid_t gettid()
276 //
277 // Returns the kernel thread id of the currently running thread. Kernel
278 // thread id is used to access /proc.
279 //
280 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
281 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
282 //
283 pid_t os::Linux::gettid() {
284 int rslt = syscall(SYS_gettid);
285 if (rslt == -1) {
286 // old kernel, no NPTL support
287 return getpid();
288 } else {
289 return (pid_t)rslt;
290 }
291 }
292
293 // Most versions of linux have a bug where the number of processors are
294 // determined by looking at the /proc file system. In a chroot environment,
295 // the system call returns 1. This causes the VM to act as if it is
296 // a single processor and elide locking (see is_MP() call).
297 static bool unsafe_chroot_detected = false;
298 static const char *unstable_chroot_error = "/proc file system not found.\n"
299 "Java may be unstable running multithreaded in a chroot "
300 "environment on Linux when /proc filesystem is not mounted.";
301
302 void os::Linux::initialize_system_info() {
303 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
304 if (processor_count() == 1) {
305 pid_t pid = os::Linux::gettid();
306 char fname[32];
307 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
308 FILE *fp = fopen(fname, "r");
309 if (fp == NULL) {
310 unsafe_chroot_detected = true;
311 } else {
312 fclose(fp);
313 }
314 }
315 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
316 assert(processor_count() > 0, "linux error");
317 }
318
319 void os::init_system_properties_values() {
320 // The next steps are taken in the product version:
321 //
322 // Obtain the JAVA_HOME value from the location of libjvm.so.
323 // This library should be located at:
324 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
325 //
326 // If "/jre/lib/" appears at the right place in the path, then we
327 // assume libjvm.so is installed in a JDK and we use this path.
328 //
329 // Otherwise exit with message: "Could not create the Java virtual machine."
330 //
331 // The following extra steps are taken in the debugging version:
332 //
333 // If "/jre/lib/" does NOT appear at the right place in the path
334 // instead of exit check for $JAVA_HOME environment variable.
335 //
336 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
337 // then we append a fake suffix "hotspot/libjvm.so" to this path so
338 // it looks like libjvm.so is installed there
339 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
340 //
341 // Otherwise exit.
342 //
343 // Important note: if the location of libjvm.so changes this
344 // code needs to be changed accordingly.
345
346 // See ld(1):
347 // The linker uses the following search paths to locate required
348 // shared libraries:
349 // 1: ...
350 // ...
351 // 7: The default directories, normally /lib and /usr/lib.
352 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
353 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
354 #else
355 #define DEFAULT_LIBPATH "/lib:/usr/lib"
356 #endif
357
358 // Base path of extensions installed on the system.
359 #define SYS_EXT_DIR "/usr/java/packages"
360 #define EXTENSIONS_DIR "/lib/ext"
361 #define ENDORSED_DIR "/lib/endorsed"
362
363 // Buffer that fits several sprintfs.
364 // Note that the space for the colon and the trailing null are provided
365 // by the nulls included by the sizeof operator.
366 const size_t bufsize =
367 MAX4((size_t)MAXPATHLEN, // For dll_dir & friends.
368 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH), // invariant ld_library_path
369 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR), // extensions dir
370 (size_t)MAXPATHLEN + sizeof(ENDORSED_DIR)); // endorsed dir
371 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
372
373 // sysclasspath, java_home, dll_dir
374 {
375 char *pslash;
376 os::jvm_path(buf, bufsize);
377
378 // Found the full path to libjvm.so.
379 // Now cut the path to <java_home>/jre if we can.
380 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
381 pslash = strrchr(buf, '/');
382 if (pslash != NULL) {
383 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
384 }
385 Arguments::set_dll_dir(buf);
386
387 if (pslash != NULL) {
388 pslash = strrchr(buf, '/');
389 if (pslash != NULL) {
390 *pslash = '\0'; // Get rid of /<arch>.
391 pslash = strrchr(buf, '/');
392 if (pslash != NULL) {
393 *pslash = '\0'; // Get rid of /lib.
394 }
395 }
396 }
397 Arguments::set_java_home(buf);
398
399 if (!set_boot_path('/', ':')) {
400 return;
401 }
402 }
403
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 char *ld_library_path_inv = buf;
416
417 // Construct the invariant part of ld_library_path.
418 sprintf(ld_library_path_inv, SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
419
420 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
421 // should always exist (until the legacy problem cited above is
422 // addressed).
423 char *v = ::getenv("LD_LIBRARY_PATH");
424 if (v != NULL) {
425 // That's +1 for the colon and +1 for the trailing '\0'.
426 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, strlen(v) + 1 + strlen(ld_library_path_inv) + 1, mtInternal);
427 sprintf(ld_library_path, "%s:%s", v, ld_library_path_inv);
428 Arguments::set_library_path(ld_library_path);
429 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
430 } else {
431 Arguments::set_library_path(ld_library_path_inv);
432 }
433 }
434
435 // Extensions directories.
436 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
437 Arguments::set_ext_dirs(buf);
438
439 // Endorsed standards default directory.
440 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
441 Arguments::set_endorsed_dirs(buf);
442
443 FREE_C_HEAP_ARRAY(char, buf, mtInternal);
444
445 #undef DEFAULT_LIBPATH
446 #undef SYS_EXT_DIR
447 #undef EXTENSIONS_DIR
448 #undef ENDORSED_DIR
449 }
450
451 ////////////////////////////////////////////////////////////////////////////////
452 // breakpoint support
453
454 void os::breakpoint() {
455 BREAKPOINT;
456 }
457
458 extern "C" void breakpoint() {
459 // use debugger to set breakpoint here
460 }
461
462 ////////////////////////////////////////////////////////////////////////////////
463 // signal support
464
465 debug_only(static bool signal_sets_initialized = false);
466 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
467
468 bool os::Linux::is_sig_ignored(int sig) {
469 struct sigaction oact;
470 sigaction(sig, (struct sigaction*)NULL, &oact);
471 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
472 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
473 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
474 return true;
475 else
476 return false;
477 }
478
479 void os::Linux::signal_sets_init() {
480 // Should also have an assertion stating we are still single-threaded.
481 assert(!signal_sets_initialized, "Already initialized");
482 // Fill in signals that are necessarily unblocked for all threads in
483 // the VM. Currently, we unblock the following signals:
484 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
485 // by -Xrs (=ReduceSignalUsage));
486 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
487 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
488 // the dispositions or masks wrt these signals.
489 // Programs embedding the VM that want to use the above signals for their
490 // own purposes must, at this time, use the "-Xrs" option to prevent
491 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
492 // (See bug 4345157, and other related bugs).
493 // In reality, though, unblocking these signals is really a nop, since
494 // these signals are not blocked by default.
495 sigemptyset(&unblocked_sigs);
496 sigemptyset(&allowdebug_blocked_sigs);
497 sigaddset(&unblocked_sigs, SIGILL);
498 sigaddset(&unblocked_sigs, SIGSEGV);
499 sigaddset(&unblocked_sigs, SIGBUS);
500 sigaddset(&unblocked_sigs, SIGFPE);
501 #if defined(PPC64)
502 sigaddset(&unblocked_sigs, SIGTRAP);
503 #endif
504 sigaddset(&unblocked_sigs, SR_signum);
505
506 if (!ReduceSignalUsage) {
507 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
508 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
509 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
510 }
511 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
512 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
513 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
514 }
515 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
516 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
517 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
518 }
519 }
520 // Fill in signals that are blocked by all but the VM thread.
521 sigemptyset(&vm_sigs);
522 if (!ReduceSignalUsage)
523 sigaddset(&vm_sigs, BREAK_SIGNAL);
524 debug_only(signal_sets_initialized = true);
525
526 }
527
528 // These are signals that are unblocked while a thread is running Java.
529 // (For some reason, they get blocked by default.)
530 sigset_t* os::Linux::unblocked_signals() {
531 assert(signal_sets_initialized, "Not initialized");
532 return &unblocked_sigs;
533 }
534
535 // These are the signals that are blocked while a (non-VM) thread is
536 // running Java. Only the VM thread handles these signals.
537 sigset_t* os::Linux::vm_signals() {
538 assert(signal_sets_initialized, "Not initialized");
539 return &vm_sigs;
540 }
541
542 // These are signals that are blocked during cond_wait to allow debugger in
543 sigset_t* os::Linux::allowdebug_blocked_signals() {
544 assert(signal_sets_initialized, "Not initialized");
545 return &allowdebug_blocked_sigs;
546 }
547
548 void os::Linux::hotspot_sigmask(Thread* thread) {
549
550 //Save caller's signal mask before setting VM signal mask
551 sigset_t caller_sigmask;
552 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
553
554 OSThread* osthread = thread->osthread();
555 osthread->set_caller_sigmask(caller_sigmask);
556
557 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
558
559 if (!ReduceSignalUsage) {
560 if (thread->is_VM_thread()) {
561 // Only the VM thread handles BREAK_SIGNAL ...
562 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
563 } else {
564 // ... all other threads block BREAK_SIGNAL
565 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
566 }
567 }
568 }
569
570 //////////////////////////////////////////////////////////////////////////////
571 // detecting pthread library
572
573 void os::Linux::libpthread_init() {
574 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
575 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
576 // generic name for earlier versions.
577 // Define macros here so we can build HotSpot on old systems.
578 # ifndef _CS_GNU_LIBC_VERSION
579 # define _CS_GNU_LIBC_VERSION 2
580 # endif
581 # ifndef _CS_GNU_LIBPTHREAD_VERSION
582 # define _CS_GNU_LIBPTHREAD_VERSION 3
583 # endif
584
585 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
586 if (n > 0) {
587 char *str = (char *)malloc(n, mtInternal);
588 confstr(_CS_GNU_LIBC_VERSION, str, n);
589 os::Linux::set_glibc_version(str);
590 } else {
591 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
592 static char _gnu_libc_version[32];
593 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
594 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
595 os::Linux::set_glibc_version(_gnu_libc_version);
596 }
597
598 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
599 if (n > 0) {
600 char *str = (char *)malloc(n, mtInternal);
601 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
602 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
603 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
604 // is the case. LinuxThreads has a hard limit on max number of threads.
605 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
606 // On the other hand, NPTL does not have such a limit, sysconf()
607 // will return -1 and errno is not changed. Check if it is really NPTL.
608 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
609 strstr(str, "NPTL") &&
610 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
611 free(str);
612 os::Linux::set_libpthread_version("linuxthreads");
613 } else {
614 os::Linux::set_libpthread_version(str);
615 }
616 } else {
617 // glibc before 2.3.2 only has LinuxThreads.
618 os::Linux::set_libpthread_version("linuxthreads");
619 }
620
621 if (strstr(libpthread_version(), "NPTL")) {
622 os::Linux::set_is_NPTL();
623 } else {
624 os::Linux::set_is_LinuxThreads();
625 }
626
627 // LinuxThreads have two flavors: floating-stack mode, which allows variable
628 // stack size; and fixed-stack mode. NPTL is always floating-stack.
629 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
630 os::Linux::set_is_floating_stack();
631 }
632 }
633
634 /////////////////////////////////////////////////////////////////////////////
635 // thread stack
636
637 // Force Linux kernel to expand current thread stack. If "bottom" is close
638 // to the stack guard, caller should block all signals.
639 //
640 // MAP_GROWSDOWN:
641 // A special mmap() flag that is used to implement thread stacks. It tells
642 // kernel that the memory region should extend downwards when needed. This
643 // allows early versions of LinuxThreads to only mmap the first few pages
644 // when creating a new thread. Linux kernel will automatically expand thread
645 // stack as needed (on page faults).
646 //
647 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
648 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
649 // region, it's hard to tell if the fault is due to a legitimate stack
650 // access or because of reading/writing non-exist memory (e.g. buffer
651 // overrun). As a rule, if the fault happens below current stack pointer,
652 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
653 // application (see Linux kernel fault.c).
654 //
655 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
656 // stack overflow detection.
657 //
658 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
659 // not use this flag. However, the stack of initial thread is not created
660 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
661 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
662 // and then attach the thread to JVM.
663 //
664 // To get around the problem and allow stack banging on Linux, we need to
665 // manually expand thread stack after receiving the SIGSEGV.
666 //
667 // There are two ways to expand thread stack to address "bottom", we used
668 // both of them in JVM before 1.5:
669 // 1. adjust stack pointer first so that it is below "bottom", and then
670 // touch "bottom"
671 // 2. mmap() the page in question
672 //
673 // Now alternate signal stack is gone, it's harder to use 2. For instance,
674 // if current sp is already near the lower end of page 101, and we need to
675 // call mmap() to map page 100, it is possible that part of the mmap() frame
676 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
677 // That will destroy the mmap() frame and cause VM to crash.
678 //
679 // The following code works by adjusting sp first, then accessing the "bottom"
680 // page to force a page fault. Linux kernel will then automatically expand the
681 // stack mapping.
682 //
683 // _expand_stack_to() assumes its frame size is less than page size, which
684 // should always be true if the function is not inlined.
685
686 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
687 #define NOINLINE
688 #else
689 #define NOINLINE __attribute__ ((noinline))
690 #endif
691
692 static void _expand_stack_to(address bottom) NOINLINE;
693
694 static void _expand_stack_to(address bottom) {
695 address sp;
696 size_t size;
697 volatile char *p;
698
699 // Adjust bottom to point to the largest address within the same page, it
700 // gives us a one-page buffer if alloca() allocates slightly more memory.
701 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
702 bottom += os::Linux::page_size() - 1;
703
704 // sp might be slightly above current stack pointer; if that's the case, we
705 // will alloca() a little more space than necessary, which is OK. Don't use
706 // os::current_stack_pointer(), as its result can be slightly below current
707 // stack pointer, causing us to not alloca enough to reach "bottom".
708 sp = (address)&sp;
709
710 if (sp > bottom) {
711 size = sp - bottom;
712 p = (volatile char *)alloca(size);
713 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
714 p[0] = '\0';
715 }
716 }
717
718 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
719 assert(t!=NULL, "just checking");
720 assert(t->osthread()->expanding_stack(), "expand should be set");
721 assert(t->stack_base() != NULL, "stack_base was not initialized");
722
723 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
724 sigset_t mask_all, old_sigset;
725 sigfillset(&mask_all);
726 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
727 _expand_stack_to(addr);
728 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
729 return true;
730 }
731 return false;
732 }
733
734 //////////////////////////////////////////////////////////////////////////////
735 // create new thread
736
737 static address highest_vm_reserved_address();
738
739 // check if it's safe to start a new thread
740 static bool _thread_safety_check(Thread* thread) {
741 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
742 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
743 // Heap is mmap'ed at lower end of memory space. Thread stacks are
744 // allocated (MAP_FIXED) from high address space. Every thread stack
745 // occupies a fixed size slot (usually 2Mbytes, but user can change
746 // it to other values if they rebuild LinuxThreads).
747 //
748 // Problem with MAP_FIXED is that mmap() can still succeed even part of
749 // the memory region has already been mmap'ed. That means if we have too
750 // many threads and/or very large heap, eventually thread stack will
751 // collide with heap.
752 //
753 // Here we try to prevent heap/stack collision by comparing current
754 // stack bottom with the highest address that has been mmap'ed by JVM
755 // plus a safety margin for memory maps created by native code.
756 //
757 // This feature can be disabled by setting ThreadSafetyMargin to 0
758 //
759 if (ThreadSafetyMargin > 0) {
760 address stack_bottom = os::current_stack_base() - os::current_stack_size();
761
762 // not safe if our stack extends below the safety margin
763 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
764 } else {
765 return true;
766 }
767 } else {
768 // Floating stack LinuxThreads or NPTL:
769 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
770 // there's not enough space left, pthread_create() will fail. If we come
771 // here, that means enough space has been reserved for stack.
772 return true;
773 }
774 }
775
776 // Thread start routine for all newly created threads
777 static void *java_start(Thread *thread) {
778 // Try to randomize the cache line index of hot stack frames.
779 // This helps when threads of the same stack traces evict each other's
780 // cache lines. The threads can be either from the same JVM instance, or
781 // from different JVM instances. The benefit is especially true for
782 // processors with hyperthreading technology.
783 static int counter = 0;
784 int pid = os::current_process_id();
785 alloca(((pid ^ counter++) & 7) * 128);
786
787 ThreadLocalStorage::set_thread(thread);
788
789 OSThread* osthread = thread->osthread();
790 Monitor* sync = osthread->startThread_lock();
791
792 // non floating stack LinuxThreads needs extra check, see above
793 if (!_thread_safety_check(thread)) {
794 // notify parent thread
795 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
796 osthread->set_state(ZOMBIE);
797 sync->notify_all();
798 return NULL;
799 }
800
801 // thread_id is kernel thread id (similar to Solaris LWP id)
802 osthread->set_thread_id(os::Linux::gettid());
803
804 if (UseNUMA) {
805 int lgrp_id = os::numa_get_group_id();
806 if (lgrp_id != -1) {
807 thread->set_lgrp_id(lgrp_id);
808 }
809 }
810 // initialize signal mask for this thread
811 os::Linux::hotspot_sigmask(thread);
812
813 // initialize floating point control register
814 os::Linux::init_thread_fpu_state();
815
816 // handshaking with parent thread
817 {
818 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
819
820 // notify parent thread
821 osthread->set_state(INITIALIZED);
822 sync->notify_all();
823
824 // wait until os::start_thread()
825 while (osthread->get_state() == INITIALIZED) {
826 sync->wait(Mutex::_no_safepoint_check_flag);
827 }
828 }
829
830 // call one more level start routine
831 thread->run();
832
833 return 0;
834 }
835
836 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
837 assert(thread->osthread() == NULL, "caller responsible");
838
839 // Allocate the OSThread object
840 OSThread* osthread = new OSThread(NULL, NULL);
841 if (osthread == NULL) {
842 return false;
843 }
844
845 // set the correct thread state
846 osthread->set_thread_type(thr_type);
847
848 // Initial state is ALLOCATED but not INITIALIZED
849 osthread->set_state(ALLOCATED);
850
851 thread->set_osthread(osthread);
852
853 // init thread attributes
854 pthread_attr_t attr;
855 pthread_attr_init(&attr);
856 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
857
858 // stack size
859 if (os::Linux::supports_variable_stack_size()) {
860 // calculate stack size if it's not specified by caller
861 if (stack_size == 0) {
862 stack_size = os::Linux::default_stack_size(thr_type);
863
864 switch (thr_type) {
865 case os::java_thread:
866 // Java threads use ThreadStackSize which default value can be
867 // changed with the flag -Xss
868 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
869 stack_size = JavaThread::stack_size_at_create();
870 break;
871 case os::compiler_thread:
872 if (CompilerThreadStackSize > 0) {
873 stack_size = (size_t)(CompilerThreadStackSize * K);
874 break;
875 } // else fall through:
876 // use VMThreadStackSize if CompilerThreadStackSize is not defined
877 case os::vm_thread:
878 case os::pgc_thread:
879 case os::cgc_thread:
880 case os::watcher_thread:
881 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
882 break;
883 }
884 }
885
886 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
887 pthread_attr_setstacksize(&attr, stack_size);
888 } else {
889 // let pthread_create() pick the default value.
890 }
891
892 // glibc guard page
893 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
894
895 ThreadState state;
896
897 {
898 // Serialize thread creation if we are running with fixed stack LinuxThreads
899 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
900 if (lock) {
901 os::Linux::createThread_lock()->lock_without_safepoint_check();
902 }
903
904 pthread_t tid;
905 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
906
907 pthread_attr_destroy(&attr);
908
909 if (ret != 0) {
910 if (PrintMiscellaneous && (Verbose || WizardMode)) {
911 perror("pthread_create()");
912 }
913 // Need to clean up stuff we've allocated so far
914 thread->set_osthread(NULL);
915 delete osthread;
916 if (lock) os::Linux::createThread_lock()->unlock();
917 return false;
918 }
919
920 // Store pthread info into the OSThread
921 osthread->set_pthread_id(tid);
922
923 // Wait until child thread is either initialized or aborted
924 {
925 Monitor* sync_with_child = osthread->startThread_lock();
926 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
927 while ((state = osthread->get_state()) == ALLOCATED) {
928 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
929 }
930 }
931
932 if (lock) {
933 os::Linux::createThread_lock()->unlock();
934 }
935 }
936
937 // Aborted due to thread limit being reached
938 if (state == ZOMBIE) {
939 thread->set_osthread(NULL);
940 delete osthread;
941 return false;
942 }
943
944 // The thread is returned suspended (in state INITIALIZED),
945 // and is started higher up in the call chain
946 assert(state == INITIALIZED, "race condition");
947 return true;
948 }
949
950 /////////////////////////////////////////////////////////////////////////////
951 // attach existing thread
952
953 // bootstrap the main thread
954 bool os::create_main_thread(JavaThread* thread) {
955 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
956 return create_attached_thread(thread);
957 }
958
959 bool os::create_attached_thread(JavaThread* thread) {
960 #ifdef ASSERT
961 thread->verify_not_published();
962 #endif
963
964 // Allocate the OSThread object
965 OSThread* osthread = new OSThread(NULL, NULL);
966
967 if (osthread == NULL) {
968 return false;
969 }
970
971 // Store pthread info into the OSThread
972 osthread->set_thread_id(os::Linux::gettid());
973 osthread->set_pthread_id(::pthread_self());
974
975 // initialize floating point control register
976 os::Linux::init_thread_fpu_state();
977
978 // Initial thread state is RUNNABLE
979 osthread->set_state(RUNNABLE);
980
981 thread->set_osthread(osthread);
982
983 if (UseNUMA) {
984 int lgrp_id = os::numa_get_group_id();
985 if (lgrp_id != -1) {
986 thread->set_lgrp_id(lgrp_id);
987 }
988 }
989
990 if (os::Linux::is_initial_thread()) {
991 // If current thread is initial thread, its stack is mapped on demand,
992 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
993 // the entire stack region to avoid SEGV in stack banging.
994 // It is also useful to get around the heap-stack-gap problem on SuSE
995 // kernel (see 4821821 for details). We first expand stack to the top
996 // of yellow zone, then enable stack yellow zone (order is significant,
997 // enabling yellow zone first will crash JVM on SuSE Linux), so there
998 // is no gap between the last two virtual memory regions.
999
1000 JavaThread *jt = (JavaThread *)thread;
1001 address addr = jt->stack_yellow_zone_base();
1002 assert(addr != NULL, "initialization problem?");
1003 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1004
1005 osthread->set_expanding_stack();
1006 os::Linux::manually_expand_stack(jt, addr);
1007 osthread->clear_expanding_stack();
1008 }
1009
1010 // initialize signal mask for this thread
1011 // and save the caller's signal mask
1012 os::Linux::hotspot_sigmask(thread);
1013
1014 return true;
1015 }
1016
1017 void os::pd_start_thread(Thread* thread) {
1018 OSThread * osthread = thread->osthread();
1019 assert(osthread->get_state() != INITIALIZED, "just checking");
1020 Monitor* sync_with_child = osthread->startThread_lock();
1021 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1022 sync_with_child->notify();
1023 }
1024
1025 // Free Linux resources related to the OSThread
1026 void os::free_thread(OSThread* osthread) {
1027 assert(osthread != NULL, "osthread not set");
1028
1029 if (Thread::current()->osthread() == osthread) {
1030 // Restore caller's signal mask
1031 sigset_t sigmask = osthread->caller_sigmask();
1032 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1033 }
1034
1035 delete osthread;
1036 }
1037
1038 //////////////////////////////////////////////////////////////////////////////
1039 // thread local storage
1040
1041 int os::allocate_thread_local_storage() {
1042 pthread_key_t key;
1043 int rslt = pthread_key_create(&key, NULL);
1044 assert(rslt == 0, "cannot allocate thread local storage");
1045 return (int)key;
1046 }
1047
1048 // Note: This is currently not used by VM, as we don't destroy TLS key
1049 // on VM exit.
1050 void os::free_thread_local_storage(int index) {
1051 int rslt = pthread_key_delete((pthread_key_t)index);
1052 assert(rslt == 0, "invalid index");
1053 }
1054
1055 void os::thread_local_storage_at_put(int index, void* value) {
1056 int rslt = pthread_setspecific((pthread_key_t)index, value);
1057 assert(rslt == 0, "pthread_setspecific failed");
1058 }
1059
1060 extern "C" Thread* get_thread() {
1061 return ThreadLocalStorage::thread();
1062 }
1063
1064 //////////////////////////////////////////////////////////////////////////////
1065 // initial thread
1066
1067 // Check if current thread is the initial thread, similar to Solaris thr_main.
1068 bool os::Linux::is_initial_thread(void) {
1069 char dummy;
1070 // If called before init complete, thread stack bottom will be null.
1071 // Can be called if fatal error occurs before initialization.
1072 if (initial_thread_stack_bottom() == NULL) return false;
1073 assert(initial_thread_stack_bottom() != NULL &&
1074 initial_thread_stack_size() != 0,
1075 "os::init did not locate initial thread's stack region");
1076 if ((address)&dummy >= initial_thread_stack_bottom() &&
1077 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1078 return true;
1079 else return false;
1080 }
1081
1082 // Find the virtual memory area that contains addr
1083 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1084 FILE *fp = fopen("/proc/self/maps", "r");
1085 if (fp) {
1086 address low, high;
1087 while (!feof(fp)) {
1088 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1089 if (low <= addr && addr < high) {
1090 if (vma_low) *vma_low = low;
1091 if (vma_high) *vma_high = high;
1092 fclose (fp);
1093 return true;
1094 }
1095 }
1096 for (;;) {
1097 int ch = fgetc(fp);
1098 if (ch == EOF || ch == (int)'\n') break;
1099 }
1100 }
1101 fclose(fp);
1102 }
1103 return false;
1104 }
1105
1106 // Locate initial thread stack. This special handling of initial thread stack
1107 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1108 // bogus value for initial thread.
1109 void os::Linux::capture_initial_stack(size_t max_size) {
1110 // stack size is the easy part, get it from RLIMIT_STACK
1111 size_t stack_size;
1112 struct rlimit rlim;
1113 getrlimit(RLIMIT_STACK, &rlim);
1114 stack_size = rlim.rlim_cur;
1115
1116 // 6308388: a bug in ld.so will relocate its own .data section to the
1117 // lower end of primordial stack; reduce ulimit -s value a little bit
1118 // so we won't install guard page on ld.so's data section.
1119 stack_size -= 2 * page_size();
1120
1121 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1122 // 7.1, in both cases we will get 2G in return value.
1123 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1124 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1125 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1126 // in case other parts in glibc still assumes 2M max stack size.
1127 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1128 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1129 if (stack_size > 2 * K * K IA64_ONLY(*2))
1130 stack_size = 2 * K * K IA64_ONLY(*2);
1131 // Try to figure out where the stack base (top) is. This is harder.
1132 //
1133 // When an application is started, glibc saves the initial stack pointer in
1134 // a global variable "__libc_stack_end", which is then used by system
1135 // libraries. __libc_stack_end should be pretty close to stack top. The
1136 // variable is available since the very early days. However, because it is
1137 // a private interface, it could disappear in the future.
1138 //
1139 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1140 // to __libc_stack_end, it is very close to stack top, but isn't the real
1141 // stack top. Note that /proc may not exist if VM is running as a chroot
1142 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1143 // /proc/<pid>/stat could change in the future (though unlikely).
1144 //
1145 // We try __libc_stack_end first. If that doesn't work, look for
1146 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1147 // as a hint, which should work well in most cases.
1148
1149 uintptr_t stack_start;
1150
1151 // try __libc_stack_end first
1152 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1153 if (p && *p) {
1154 stack_start = *p;
1155 } else {
1156 // see if we can get the start_stack field from /proc/self/stat
1157 FILE *fp;
1158 int pid;
1159 char state;
1160 int ppid;
1161 int pgrp;
1162 int session;
1163 int nr;
1164 int tpgrp;
1165 unsigned long flags;
1166 unsigned long minflt;
1167 unsigned long cminflt;
1168 unsigned long majflt;
1169 unsigned long cmajflt;
1170 unsigned long utime;
1171 unsigned long stime;
1172 long cutime;
1173 long cstime;
1174 long prio;
1175 long nice;
1176 long junk;
1177 long it_real;
1178 uintptr_t start;
1179 uintptr_t vsize;
1180 intptr_t rss;
1181 uintptr_t rsslim;
1182 uintptr_t scodes;
1183 uintptr_t ecode;
1184 int i;
1185
1186 // Figure what the primordial thread stack base is. Code is inspired
1187 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1188 // followed by command name surrounded by parentheses, state, etc.
1189 char stat[2048];
1190 int statlen;
1191
1192 fp = fopen("/proc/self/stat", "r");
1193 if (fp) {
1194 statlen = fread(stat, 1, 2047, fp);
1195 stat[statlen] = '\0';
1196 fclose(fp);
1197
1198 // Skip pid and the command string. Note that we could be dealing with
1199 // weird command names, e.g. user could decide to rename java launcher
1200 // to "java 1.4.2 :)", then the stat file would look like
1201 // 1234 (java 1.4.2 :)) R ... ...
1202 // We don't really need to know the command string, just find the last
1203 // occurrence of ")" and then start parsing from there. See bug 4726580.
1204 char * s = strrchr(stat, ')');
1205
1206 i = 0;
1207 if (s) {
1208 // Skip blank chars
1209 do s++; while (isspace(*s));
1210
1211 #define _UFM UINTX_FORMAT
1212 #define _DFM INTX_FORMAT
1213
1214 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1215 /* 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 */
1216 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,
1217 &state, /* 3 %c */
1218 &ppid, /* 4 %d */
1219 &pgrp, /* 5 %d */
1220 &session, /* 6 %d */
1221 &nr, /* 7 %d */
1222 &tpgrp, /* 8 %d */
1223 &flags, /* 9 %lu */
1224 &minflt, /* 10 %lu */
1225 &cminflt, /* 11 %lu */
1226 &majflt, /* 12 %lu */
1227 &cmajflt, /* 13 %lu */
1228 &utime, /* 14 %lu */
1229 &stime, /* 15 %lu */
1230 &cutime, /* 16 %ld */
1231 &cstime, /* 17 %ld */
1232 &prio, /* 18 %ld */
1233 &nice, /* 19 %ld */
1234 &junk, /* 20 %ld */
1235 &it_real, /* 21 %ld */
1236 &start, /* 22 UINTX_FORMAT */
1237 &vsize, /* 23 UINTX_FORMAT */
1238 &rss, /* 24 INTX_FORMAT */
1239 &rsslim, /* 25 UINTX_FORMAT */
1240 &scodes, /* 26 UINTX_FORMAT */
1241 &ecode, /* 27 UINTX_FORMAT */
1242 &stack_start); /* 28 UINTX_FORMAT */
1243 }
1244
1245 #undef _UFM
1246 #undef _DFM
1247
1248 if (i != 28 - 2) {
1249 assert(false, "Bad conversion from /proc/self/stat");
1250 // product mode - assume we are the initial thread, good luck in the
1251 // embedded case.
1252 warning("Can't detect initial thread stack location - bad conversion");
1253 stack_start = (uintptr_t) &rlim;
1254 }
1255 } else {
1256 // For some reason we can't open /proc/self/stat (for example, running on
1257 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1258 // most cases, so don't abort:
1259 warning("Can't detect initial thread stack location - no /proc/self/stat");
1260 stack_start = (uintptr_t) &rlim;
1261 }
1262 }
1263
1264 // Now we have a pointer (stack_start) very close to the stack top, the
1265 // next thing to do is to figure out the exact location of stack top. We
1266 // can find out the virtual memory area that contains stack_start by
1267 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1268 // and its upper limit is the real stack top. (again, this would fail if
1269 // running inside chroot, because /proc may not exist.)
1270
1271 uintptr_t stack_top;
1272 address low, high;
1273 if (find_vma((address)stack_start, &low, &high)) {
1274 // success, "high" is the true stack top. (ignore "low", because initial
1275 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1276 stack_top = (uintptr_t)high;
1277 } else {
1278 // failed, likely because /proc/self/maps does not exist
1279 warning("Can't detect initial thread stack location - find_vma failed");
1280 // best effort: stack_start is normally within a few pages below the real
1281 // stack top, use it as stack top, and reduce stack size so we won't put
1282 // guard page outside stack.
1283 stack_top = stack_start;
1284 stack_size -= 16 * page_size();
1285 }
1286
1287 // stack_top could be partially down the page so align it
1288 stack_top = align_size_up(stack_top, page_size());
1289
1290 if (max_size && stack_size > max_size) {
1291 _initial_thread_stack_size = max_size;
1292 } else {
1293 _initial_thread_stack_size = stack_size;
1294 }
1295
1296 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1297 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1298 }
1299
1300 ////////////////////////////////////////////////////////////////////////////////
1301 // time support
1302
1303 // Time since start-up in seconds to a fine granularity.
1304 // Used by VMSelfDestructTimer and the MemProfiler.
1305 double os::elapsedTime() {
1306
1307 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1308 }
1309
1310 jlong os::elapsed_counter() {
1311 return javaTimeNanos() - initial_time_count;
1312 }
1313
1314 jlong os::elapsed_frequency() {
1315 return NANOSECS_PER_SEC; // nanosecond resolution
1316 }
1317
1318 bool os::supports_vtime() { return true; }
1319 bool os::enable_vtime() { return false; }
1320 bool os::vtime_enabled() { return false; }
1321
1322 double os::elapsedVTime() {
1323 struct rusage usage;
1324 int retval = getrusage(RUSAGE_THREAD, &usage);
1325 if (retval == 0) {
1326 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);
1327 } else {
1328 // better than nothing, but not much
1329 return elapsedTime();
1330 }
1331 }
1332
1333 jlong os::javaTimeMillis() {
1334 timeval time;
1335 int status = gettimeofday(&time, NULL);
1336 assert(status != -1, "linux error");
1337 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1338 }
1339
1340 #ifndef CLOCK_MONOTONIC
1341 #define CLOCK_MONOTONIC (1)
1342 #endif
1343
1344 void os::Linux::clock_init() {
1345 // we do dlopen's in this particular order due to bug in linux
1346 // dynamical loader (see 6348968) leading to crash on exit
1347 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1348 if (handle == NULL) {
1349 handle = dlopen("librt.so", RTLD_LAZY);
1350 }
1351
1352 if (handle) {
1353 int (*clock_getres_func)(clockid_t, struct timespec*) =
1354 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1355 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1356 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1357 if (clock_getres_func && clock_gettime_func) {
1358 // See if monotonic clock is supported by the kernel. Note that some
1359 // early implementations simply return kernel jiffies (updated every
1360 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1361 // for nano time (though the monotonic property is still nice to have).
1362 // It's fixed in newer kernels, however clock_getres() still returns
1363 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1364 // resolution for now. Hopefully as people move to new kernels, this
1365 // won't be a problem.
1366 struct timespec res;
1367 struct timespec tp;
1368 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1369 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1370 // yes, monotonic clock is supported
1371 _clock_gettime = clock_gettime_func;
1372 return;
1373 } else {
1374 // close librt if there is no monotonic clock
1375 dlclose(handle);
1376 }
1377 }
1378 }
1379 warning("No monotonic clock was available - timed services may " \
1380 "be adversely affected if the time-of-day clock changes");
1381 }
1382
1383 #ifndef SYS_clock_getres
1384
1385 #if defined(IA32) || defined(AMD64)
1386 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1387 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1388 #else
1389 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1390 #define sys_clock_getres(x,y) -1
1391 #endif
1392
1393 #else
1394 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1395 #endif
1396
1397 void os::Linux::fast_thread_clock_init() {
1398 if (!UseLinuxPosixThreadCPUClocks) {
1399 return;
1400 }
1401 clockid_t clockid;
1402 struct timespec tp;
1403 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1404 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1405
1406 // Switch to using fast clocks for thread cpu time if
1407 // the sys_clock_getres() returns 0 error code.
1408 // Note, that some kernels may support the current thread
1409 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1410 // returned by the pthread_getcpuclockid().
1411 // If the fast Posix clocks are supported then the sys_clock_getres()
1412 // must return at least tp.tv_sec == 0 which means a resolution
1413 // better than 1 sec. This is extra check for reliability.
1414
1415 if(pthread_getcpuclockid_func &&
1416 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1417 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1418
1419 _supports_fast_thread_cpu_time = true;
1420 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1421 }
1422 }
1423
1424 jlong os::javaTimeNanos() {
1425 if (os::supports_monotonic_clock()) {
1426 struct timespec tp;
1427 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1428 assert(status == 0, "gettime error");
1429 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1430 return result;
1431 } else {
1432 timeval time;
1433 int status = gettimeofday(&time, NULL);
1434 assert(status != -1, "linux error");
1435 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1436 return 1000 * usecs;
1437 }
1438 }
1439
1440 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1441 if (os::supports_monotonic_clock()) {
1442 info_ptr->max_value = ALL_64_BITS;
1443
1444 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1445 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1446 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1447 } else {
1448 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1449 info_ptr->max_value = ALL_64_BITS;
1450
1451 // gettimeofday is a real time clock so it skips
1452 info_ptr->may_skip_backward = true;
1453 info_ptr->may_skip_forward = true;
1454 }
1455
1456 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1457 }
1458
1459 // Return the real, user, and system times in seconds from an
1460 // arbitrary fixed point in the past.
1461 bool os::getTimesSecs(double* process_real_time,
1462 double* process_user_time,
1463 double* process_system_time) {
1464 struct tms ticks;
1465 clock_t real_ticks = times(&ticks);
1466
1467 if (real_ticks == (clock_t) (-1)) {
1468 return false;
1469 } else {
1470 double ticks_per_second = (double) clock_tics_per_sec;
1471 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1472 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1473 *process_real_time = ((double) real_ticks) / ticks_per_second;
1474
1475 return true;
1476 }
1477 }
1478
1479
1480 char * os::local_time_string(char *buf, size_t buflen) {
1481 struct tm t;
1482 time_t long_time;
1483 time(&long_time);
1484 localtime_r(&long_time, &t);
1485 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1486 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1487 t.tm_hour, t.tm_min, t.tm_sec);
1488 return buf;
1489 }
1490
1491 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1492 return localtime_r(clock, res);
1493 }
1494
1495 ////////////////////////////////////////////////////////////////////////////////
1496 // runtime exit support
1497
1498 // Note: os::shutdown() might be called very early during initialization, or
1499 // called from signal handler. Before adding something to os::shutdown(), make
1500 // sure it is async-safe and can handle partially initialized VM.
1501 void os::shutdown() {
1502
1503 // allow PerfMemory to attempt cleanup of any persistent resources
1504 perfMemory_exit();
1505
1506 // needs to remove object in file system
1507 AttachListener::abort();
1508
1509 // flush buffered output, finish log files
1510 ostream_abort();
1511
1512 // Check for abort hook
1513 abort_hook_t abort_hook = Arguments::abort_hook();
1514 if (abort_hook != NULL) {
1515 abort_hook();
1516 }
1517
1518 }
1519
1520 // Note: os::abort() might be called very early during initialization, or
1521 // called from signal handler. Before adding something to os::abort(), make
1522 // sure it is async-safe and can handle partially initialized VM.
1523 void os::abort(bool dump_core) {
1524 os::shutdown();
1525 if (dump_core) {
1526 #ifndef PRODUCT
1527 fdStream out(defaultStream::output_fd());
1528 out.print_raw("Current thread is ");
1529 char buf[16];
1530 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1531 out.print_raw_cr(buf);
1532 out.print_raw_cr("Dumping core ...");
1533 #endif
1534 ::abort(); // dump core
1535 }
1536
1537 ::exit(1);
1538 }
1539
1540 // Die immediately, no exit hook, no abort hook, no cleanup.
1541 void os::die() {
1542 // _exit() on LinuxThreads only kills current thread
1543 ::abort();
1544 }
1545
1546 // unused on linux for now.
1547 void os::set_error_file(const char *logfile) {}
1548
1549
1550 // This method is a copy of JDK's sysGetLastErrorString
1551 // from src/solaris/hpi/src/system_md.c
1552
1553 size_t os::lasterror(char *buf, size_t len) {
1554
1555 if (errno == 0) return 0;
1556
1557 const char *s = ::strerror(errno);
1558 size_t n = ::strlen(s);
1559 if (n >= len) {
1560 n = len - 1;
1561 }
1562 ::strncpy(buf, s, n);
1563 buf[n] = '\0';
1564 return n;
1565 }
1566
1567 intx os::current_thread_id() { return (intx)pthread_self(); }
1568 int os::current_process_id() {
1569
1570 // Under the old linux thread library, linux gives each thread
1571 // its own process id. Because of this each thread will return
1572 // a different pid if this method were to return the result
1573 // of getpid(2). Linux provides no api that returns the pid
1574 // of the launcher thread for the vm. This implementation
1575 // returns a unique pid, the pid of the launcher thread
1576 // that starts the vm 'process'.
1577
1578 // Under the NPTL, getpid() returns the same pid as the
1579 // launcher thread rather than a unique pid per thread.
1580 // Use gettid() if you want the old pre NPTL behaviour.
1581
1582 // if you are looking for the result of a call to getpid() that
1583 // returns a unique pid for the calling thread, then look at the
1584 // OSThread::thread_id() method in osThread_linux.hpp file
1585
1586 return (int)(_initial_pid ? _initial_pid : getpid());
1587 }
1588
1589 // DLL functions
1590
1591 const char* os::dll_file_extension() { return ".so"; }
1592
1593 // This must be hard coded because it's the system's temporary
1594 // directory not the java application's temp directory, ala java.io.tmpdir.
1595 const char* os::get_temp_directory() { return "/tmp"; }
1596
1597 static bool file_exists(const char* filename) {
1598 struct stat statbuf;
1599 if (filename == NULL || strlen(filename) == 0) {
1600 return false;
1601 }
1602 return os::stat(filename, &statbuf) == 0;
1603 }
1604
1605 bool os::dll_build_name(char* buffer, size_t buflen,
1606 const char* pname, const char* fname) {
1607 bool retval = false;
1608 // Copied from libhpi
1609 const size_t pnamelen = pname ? strlen(pname) : 0;
1610
1611 // Return error on buffer overflow.
1612 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1613 return retval;
1614 }
1615
1616 if (pnamelen == 0) {
1617 snprintf(buffer, buflen, "lib%s.so", fname);
1618 retval = true;
1619 } else if (strchr(pname, *os::path_separator()) != NULL) {
1620 int n;
1621 char** pelements = split_path(pname, &n);
1622 if (pelements == NULL) {
1623 return false;
1624 }
1625 for (int i = 0 ; i < n ; i++) {
1626 // Really shouldn't be NULL, but check can't hurt
1627 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1628 continue; // skip the empty path values
1629 }
1630 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1631 if (file_exists(buffer)) {
1632 retval = true;
1633 break;
1634 }
1635 }
1636 // release the storage
1637 for (int i = 0 ; i < n ; i++) {
1638 if (pelements[i] != NULL) {
1639 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1640 }
1641 }
1642 if (pelements != NULL) {
1643 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1644 }
1645 } else {
1646 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1647 retval = true;
1648 }
1649 return retval;
1650 }
1651
1652 // check if addr is inside libjvm.so
1653 bool os::address_is_in_vm(address addr) {
1654 static address libjvm_base_addr;
1655 Dl_info dlinfo;
1656
1657 if (libjvm_base_addr == NULL) {
1658 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1659 libjvm_base_addr = (address)dlinfo.dli_fbase;
1660 }
1661 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1662 }
1663
1664 if (dladdr((void *)addr, &dlinfo) != 0) {
1665 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1666 }
1667
1668 return false;
1669 }
1670
1671 bool os::dll_address_to_function_name(address addr, char *buf,
1672 int buflen, int *offset) {
1673 // buf is not optional, but offset is optional
1674 assert(buf != NULL, "sanity check");
1675
1676 Dl_info dlinfo;
1677
1678 if (dladdr((void*)addr, &dlinfo) != 0) {
1679 // see if we have a matching symbol
1680 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1681 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1682 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1683 }
1684 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1685 return true;
1686 }
1687 // no matching symbol so try for just file info
1688 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1689 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1690 buf, buflen, offset, dlinfo.dli_fname)) {
1691 return true;
1692 }
1693 }
1694 }
1695
1696 buf[0] = '\0';
1697 if (offset != NULL) *offset = -1;
1698 return false;
1699 }
1700
1701 struct _address_to_library_name {
1702 address addr; // input : memory address
1703 size_t buflen; // size of fname
1704 char* fname; // output: library name
1705 address base; // library base addr
1706 };
1707
1708 static int address_to_library_name_callback(struct dl_phdr_info *info,
1709 size_t size, void *data) {
1710 int i;
1711 bool found = false;
1712 address libbase = NULL;
1713 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1714
1715 // iterate through all loadable segments
1716 for (i = 0; i < info->dlpi_phnum; i++) {
1717 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1718 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1719 // base address of a library is the lowest address of its loaded
1720 // segments.
1721 if (libbase == NULL || libbase > segbase) {
1722 libbase = segbase;
1723 }
1724 // see if 'addr' is within current segment
1725 if (segbase <= d->addr &&
1726 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1727 found = true;
1728 }
1729 }
1730 }
1731
1732 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1733 // so dll_address_to_library_name() can fall through to use dladdr() which
1734 // can figure out executable name from argv[0].
1735 if (found && info->dlpi_name && info->dlpi_name[0]) {
1736 d->base = libbase;
1737 if (d->fname) {
1738 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1739 }
1740 return 1;
1741 }
1742 return 0;
1743 }
1744
1745 bool os::dll_address_to_library_name(address addr, char* buf,
1746 int buflen, int* offset) {
1747 // buf is not optional, but offset is optional
1748 assert(buf != NULL, "sanity check");
1749
1750 Dl_info dlinfo;
1751 struct _address_to_library_name data;
1752
1753 // There is a bug in old glibc dladdr() implementation that it could resolve
1754 // to wrong library name if the .so file has a base address != NULL. Here
1755 // we iterate through the program headers of all loaded libraries to find
1756 // out which library 'addr' really belongs to. This workaround can be
1757 // removed once the minimum requirement for glibc is moved to 2.3.x.
1758 data.addr = addr;
1759 data.fname = buf;
1760 data.buflen = buflen;
1761 data.base = NULL;
1762 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1763
1764 if (rslt) {
1765 // buf already contains library name
1766 if (offset) *offset = addr - data.base;
1767 return true;
1768 }
1769 if (dladdr((void*)addr, &dlinfo) != 0) {
1770 if (dlinfo.dli_fname != NULL) {
1771 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1772 }
1773 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1774 *offset = addr - (address)dlinfo.dli_fbase;
1775 }
1776 return true;
1777 }
1778
1779 buf[0] = '\0';
1780 if (offset) *offset = -1;
1781 return false;
1782 }
1783
1784 // Loads .dll/.so and
1785 // in case of error it checks if .dll/.so was built for the
1786 // same architecture as Hotspot is running on
1787
1788
1789 // Remember the stack's state. The Linux dynamic linker will change
1790 // the stack to 'executable' at most once, so we must safepoint only once.
1791 bool os::Linux::_stack_is_executable = false;
1792
1793 // VM operation that loads a library. This is necessary if stack protection
1794 // of the Java stacks can be lost during loading the library. If we
1795 // do not stop the Java threads, they can stack overflow before the stacks
1796 // are protected again.
1797 class VM_LinuxDllLoad: public VM_Operation {
1798 private:
1799 const char *_filename;
1800 char *_ebuf;
1801 int _ebuflen;
1802 void *_lib;
1803 public:
1804 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1805 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1806 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1807 void doit() {
1808 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1809 os::Linux::_stack_is_executable = true;
1810 }
1811 void* loaded_library() { return _lib; }
1812 };
1813
1814 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1815 {
1816 void * result = NULL;
1817 bool load_attempted = false;
1818
1819 // Check whether the library to load might change execution rights
1820 // of the stack. If they are changed, the protection of the stack
1821 // guard pages will be lost. We need a safepoint to fix this.
1822 //
1823 // See Linux man page execstack(8) for more info.
1824 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1825 ElfFile ef(filename);
1826 if (!ef.specifies_noexecstack()) {
1827 if (!is_init_completed()) {
1828 os::Linux::_stack_is_executable = true;
1829 // This is OK - No Java threads have been created yet, and hence no
1830 // stack guard pages to fix.
1831 //
1832 // This should happen only when you are building JDK7 using a very
1833 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1834 //
1835 // Dynamic loader will make all stacks executable after
1836 // this function returns, and will not do that again.
1837 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1838 } else {
1839 warning("You have loaded library %s which might have disabled stack guard. "
1840 "The VM will try to fix the stack guard now.\n"
1841 "It's highly recommended that you fix the library with "
1842 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1843 filename);
1844
1845 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1846 JavaThread *jt = JavaThread::current();
1847 if (jt->thread_state() != _thread_in_native) {
1848 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1849 // that requires ExecStack. Cannot enter safe point. Let's give up.
1850 warning("Unable to fix stack guard. Giving up.");
1851 } else {
1852 if (!LoadExecStackDllInVMThread) {
1853 // This is for the case where the DLL has an static
1854 // constructor function that executes JNI code. We cannot
1855 // load such DLLs in the VMThread.
1856 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1857 }
1858
1859 ThreadInVMfromNative tiv(jt);
1860 debug_only(VMNativeEntryWrapper vew;)
1861
1862 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1863 VMThread::execute(&op);
1864 if (LoadExecStackDllInVMThread) {
1865 result = op.loaded_library();
1866 }
1867 load_attempted = true;
1868 }
1869 }
1870 }
1871 }
1872
1873 if (!load_attempted) {
1874 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1875 }
1876
1877 if (result != NULL) {
1878 // Successful loading
1879 return result;
1880 }
1881
1882 Elf32_Ehdr elf_head;
1883 int diag_msg_max_length=ebuflen-strlen(ebuf);
1884 char* diag_msg_buf=ebuf+strlen(ebuf);
1885
1886 if (diag_msg_max_length==0) {
1887 // No more space in ebuf for additional diagnostics message
1888 return NULL;
1889 }
1890
1891
1892 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1893
1894 if (file_descriptor < 0) {
1895 // Can't open library, report dlerror() message
1896 return NULL;
1897 }
1898
1899 bool failed_to_read_elf_head=
1900 (sizeof(elf_head)!=
1901 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1902
1903 ::close(file_descriptor);
1904 if (failed_to_read_elf_head) {
1905 // file i/o error - report dlerror() msg
1906 return NULL;
1907 }
1908
1909 typedef struct {
1910 Elf32_Half code; // Actual value as defined in elf.h
1911 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1912 char elf_class; // 32 or 64 bit
1913 char endianess; // MSB or LSB
1914 char* name; // String representation
1915 } arch_t;
1916
1917 #ifndef EM_486
1918 #define EM_486 6 /* Intel 80486 */
1919 #endif
1920
1921 static const arch_t arch_array[]={
1922 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1923 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1924 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1925 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1926 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1927 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1928 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1929 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1930 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1931 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1932 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1933 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1934 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1935 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1936 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1937 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1938 };
1939
1940 #if (defined IA32)
1941 static Elf32_Half running_arch_code=EM_386;
1942 #elif (defined AMD64)
1943 static Elf32_Half running_arch_code=EM_X86_64;
1944 #elif (defined IA64)
1945 static Elf32_Half running_arch_code=EM_IA_64;
1946 #elif (defined __sparc) && (defined _LP64)
1947 static Elf32_Half running_arch_code=EM_SPARCV9;
1948 #elif (defined __sparc) && (!defined _LP64)
1949 static Elf32_Half running_arch_code=EM_SPARC;
1950 #elif (defined __powerpc64__)
1951 static Elf32_Half running_arch_code=EM_PPC64;
1952 #elif (defined __powerpc__)
1953 static Elf32_Half running_arch_code=EM_PPC;
1954 #elif (defined ARM)
1955 static Elf32_Half running_arch_code=EM_ARM;
1956 #elif (defined S390)
1957 static Elf32_Half running_arch_code=EM_S390;
1958 #elif (defined ALPHA)
1959 static Elf32_Half running_arch_code=EM_ALPHA;
1960 #elif (defined MIPSEL)
1961 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1962 #elif (defined PARISC)
1963 static Elf32_Half running_arch_code=EM_PARISC;
1964 #elif (defined MIPS)
1965 static Elf32_Half running_arch_code=EM_MIPS;
1966 #elif (defined M68K)
1967 static Elf32_Half running_arch_code=EM_68K;
1968 #else
1969 #error Method os::dll_load requires that one of following is defined:\
1970 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1971 #endif
1972
1973 // Identify compatability class for VM's architecture and library's architecture
1974 // Obtain string descriptions for architectures
1975
1976 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1977 int running_arch_index=-1;
1978
1979 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1980 if (running_arch_code == arch_array[i].code) {
1981 running_arch_index = i;
1982 }
1983 if (lib_arch.code == arch_array[i].code) {
1984 lib_arch.compat_class = arch_array[i].compat_class;
1985 lib_arch.name = arch_array[i].name;
1986 }
1987 }
1988
1989 assert(running_arch_index != -1,
1990 "Didn't find running architecture code (running_arch_code) in arch_array");
1991 if (running_arch_index == -1) {
1992 // Even though running architecture detection failed
1993 // we may still continue with reporting dlerror() message
1994 return NULL;
1995 }
1996
1997 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1998 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1999 return NULL;
2000 }
2001
2002 #ifndef S390
2003 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2004 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2005 return NULL;
2006 }
2007 #endif // !S390
2008
2009 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2010 if ( lib_arch.name!=NULL ) {
2011 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2012 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2013 lib_arch.name, arch_array[running_arch_index].name);
2014 } else {
2015 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2016 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2017 lib_arch.code,
2018 arch_array[running_arch_index].name);
2019 }
2020 }
2021
2022 return NULL;
2023 }
2024
2025 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2026 void * result = ::dlopen(filename, RTLD_LAZY);
2027 if (result == NULL) {
2028 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2029 ebuf[ebuflen-1] = '\0';
2030 }
2031 return result;
2032 }
2033
2034 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2035 void * result = NULL;
2036 if (LoadExecStackDllInVMThread) {
2037 result = dlopen_helper(filename, ebuf, ebuflen);
2038 }
2039
2040 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2041 // library that requires an executable stack, or which does not have this
2042 // stack attribute set, dlopen changes the stack attribute to executable. The
2043 // read protection of the guard pages gets lost.
2044 //
2045 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2046 // may have been queued at the same time.
2047
2048 if (!_stack_is_executable) {
2049 JavaThread *jt = Threads::first();
2050
2051 while (jt) {
2052 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2053 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2054 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2055 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2056 warning("Attempt to reguard stack yellow zone failed.");
2057 }
2058 }
2059 jt = jt->next();
2060 }
2061 }
2062
2063 return result;
2064 }
2065
2066 /*
2067 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2068 * chances are you might want to run the generated bits against glibc-2.0
2069 * libdl.so, so always use locking for any version of glibc.
2070 */
2071 void* os::dll_lookup(void* handle, const char* name) {
2072 pthread_mutex_lock(&dl_mutex);
2073 void* res = dlsym(handle, name);
2074 pthread_mutex_unlock(&dl_mutex);
2075 return res;
2076 }
2077
2078 void* os::get_default_process_handle() {
2079 return (void*)::dlopen(NULL, RTLD_LAZY);
2080 }
2081
2082 static bool _print_ascii_file(const char* filename, outputStream* st) {
2083 int fd = ::open(filename, O_RDONLY);
2084 if (fd == -1) {
2085 return false;
2086 }
2087
2088 char buf[32];
2089 int bytes;
2090 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2091 st->print_raw(buf, bytes);
2092 }
2093
2094 ::close(fd);
2095
2096 return true;
2097 }
2098
2099 void os::print_dll_info(outputStream *st) {
2100 st->print_cr("Dynamic libraries:");
2101
2102 char fname[32];
2103 pid_t pid = os::Linux::gettid();
2104
2105 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2106
2107 if (!_print_ascii_file(fname, st)) {
2108 st->print("Can not get library information for pid = %d\n", pid);
2109 }
2110 }
2111
2112 void os::print_os_info_brief(outputStream* st) {
2113 os::Linux::print_distro_info(st);
2114
2115 os::Posix::print_uname_info(st);
2116
2117 os::Linux::print_libversion_info(st);
2118
2119 }
2120
2121 void os::print_os_info(outputStream* st) {
2122 st->print("OS:");
2123
2124 os::Linux::print_distro_info(st);
2125
2126 os::Posix::print_uname_info(st);
2127
2128 // Print warning if unsafe chroot environment detected
2129 if (unsafe_chroot_detected) {
2130 st->print("WARNING!! ");
2131 st->print_cr(unstable_chroot_error);
2132 }
2133
2134 os::Linux::print_libversion_info(st);
2135
2136 os::Posix::print_rlimit_info(st);
2137
2138 os::Posix::print_load_average(st);
2139
2140 os::Linux::print_full_memory_info(st);
2141 }
2142
2143 // Try to identify popular distros.
2144 // Most Linux distributions have a /etc/XXX-release file, which contains
2145 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2146 // file that also contains the OS version string. Some have more than one
2147 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2148 // /etc/redhat-release.), so the order is important.
2149 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2150 // their own specific XXX-release file as well as a redhat-release file.
2151 // Because of this the XXX-release file needs to be searched for before the
2152 // redhat-release file.
2153 // Since Red Hat has a lsb-release file that is not very descriptive the
2154 // search for redhat-release needs to be before lsb-release.
2155 // Since the lsb-release file is the new standard it needs to be searched
2156 // before the older style release files.
2157 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2158 // next to last resort. The os-release file is a new standard that contains
2159 // distribution information and the system-release file seems to be an old
2160 // standard that has been replaced by the lsb-release and os-release files.
2161 // Searching for the debian_version file is the last resort. It contains
2162 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2163 // "Debian " is printed before the contents of the debian_version file.
2164 void os::Linux::print_distro_info(outputStream* st) {
2165 if (!_print_ascii_file("/etc/oracle-release", st) &&
2166 !_print_ascii_file("/etc/mandriva-release", st) &&
2167 !_print_ascii_file("/etc/mandrake-release", st) &&
2168 !_print_ascii_file("/etc/sun-release", st) &&
2169 !_print_ascii_file("/etc/redhat-release", st) &&
2170 !_print_ascii_file("/etc/lsb-release", st) &&
2171 !_print_ascii_file("/etc/SuSE-release", st) &&
2172 !_print_ascii_file("/etc/turbolinux-release", st) &&
2173 !_print_ascii_file("/etc/gentoo-release", st) &&
2174 !_print_ascii_file("/etc/ltib-release", st) &&
2175 !_print_ascii_file("/etc/angstrom-version", st) &&
2176 !_print_ascii_file("/etc/system-release", st) &&
2177 !_print_ascii_file("/etc/os-release", st)) {
2178
2179 if (file_exists("/etc/debian_version")) {
2180 st->print("Debian ");
2181 _print_ascii_file("/etc/debian_version", st);
2182 } else {
2183 st->print("Linux");
2184 }
2185 }
2186 st->cr();
2187 }
2188
2189 void os::Linux::print_libversion_info(outputStream* st) {
2190 // libc, pthread
2191 st->print("libc:");
2192 st->print(os::Linux::glibc_version()); st->print(" ");
2193 st->print(os::Linux::libpthread_version()); st->print(" ");
2194 if (os::Linux::is_LinuxThreads()) {
2195 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2196 }
2197 st->cr();
2198 }
2199
2200 void os::Linux::print_full_memory_info(outputStream* st) {
2201 st->print("\n/proc/meminfo:\n");
2202 _print_ascii_file("/proc/meminfo", st);
2203 st->cr();
2204 }
2205
2206 void os::print_memory_info(outputStream* st) {
2207
2208 st->print("Memory:");
2209 st->print(" %dk page", os::vm_page_size()>>10);
2210
2211 // values in struct sysinfo are "unsigned long"
2212 struct sysinfo si;
2213 sysinfo(&si);
2214
2215 st->print(", physical " UINT64_FORMAT "k",
2216 os::physical_memory() >> 10);
2217 st->print("(" UINT64_FORMAT "k free)",
2218 os::available_memory() >> 10);
2219 st->print(", swap " UINT64_FORMAT "k",
2220 ((jlong)si.totalswap * si.mem_unit) >> 10);
2221 st->print("(" UINT64_FORMAT "k free)",
2222 ((jlong)si.freeswap * si.mem_unit) >> 10);
2223 st->cr();
2224 }
2225
2226 void os::pd_print_cpu_info(outputStream* st) {
2227 st->print("\n/proc/cpuinfo:\n");
2228 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2229 st->print(" <Not Available>");
2230 }
2231 st->cr();
2232 }
2233
2234 void os::print_siginfo(outputStream* st, void* siginfo) {
2235 const siginfo_t* si = (const siginfo_t*)siginfo;
2236
2237 os::Posix::print_siginfo_brief(st, si);
2238
2239 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2240 UseSharedSpaces) {
2241 FileMapInfo* mapinfo = FileMapInfo::current_info();
2242 if (mapinfo->is_in_shared_space(si->si_addr)) {
2243 st->print("\n\nError accessing class data sharing archive." \
2244 " Mapped file inaccessible during execution, " \
2245 " possible disk/network problem.");
2246 }
2247 }
2248 st->cr();
2249 }
2250
2251
2252 static void print_signal_handler(outputStream* st, int sig,
2253 char* buf, size_t buflen);
2254
2255 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2256 st->print_cr("Signal Handlers:");
2257 print_signal_handler(st, SIGSEGV, buf, buflen);
2258 print_signal_handler(st, SIGBUS , buf, buflen);
2259 print_signal_handler(st, SIGFPE , buf, buflen);
2260 print_signal_handler(st, SIGPIPE, buf, buflen);
2261 print_signal_handler(st, SIGXFSZ, buf, buflen);
2262 print_signal_handler(st, SIGILL , buf, buflen);
2263 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2264 print_signal_handler(st, SR_signum, buf, buflen);
2265 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2266 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2267 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2268 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2269 #if defined(PPC64)
2270 print_signal_handler(st, SIGTRAP, buf, buflen);
2271 #endif
2272 }
2273
2274 static char saved_jvm_path[MAXPATHLEN] = {0};
2275
2276 // Find the full path to the current module, libjvm.so
2277 void os::jvm_path(char *buf, jint buflen) {
2278 // Error checking.
2279 if (buflen < MAXPATHLEN) {
2280 assert(false, "must use a large-enough buffer");
2281 buf[0] = '\0';
2282 return;
2283 }
2284 // Lazy resolve the path to current module.
2285 if (saved_jvm_path[0] != 0) {
2286 strcpy(buf, saved_jvm_path);
2287 return;
2288 }
2289
2290 char dli_fname[MAXPATHLEN];
2291 bool ret = dll_address_to_library_name(
2292 CAST_FROM_FN_PTR(address, os::jvm_path),
2293 dli_fname, sizeof(dli_fname), NULL);
2294 assert(ret, "cannot locate libjvm");
2295 char *rp = NULL;
2296 if (ret && dli_fname[0] != '\0') {
2297 rp = realpath(dli_fname, buf);
2298 }
2299 if (rp == NULL)
2300 return;
2301
2302 if (Arguments::sun_java_launcher_is_altjvm()) {
2303 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2304 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2305 // If "/jre/lib/" appears at the right place in the string, then
2306 // assume we are installed in a JDK and we're done. Otherwise, check
2307 // for a JAVA_HOME environment variable and fix up the path so it
2308 // looks like libjvm.so is installed there (append a fake suffix
2309 // hotspot/libjvm.so).
2310 const char *p = buf + strlen(buf) - 1;
2311 for (int count = 0; p > buf && count < 5; ++count) {
2312 for (--p; p > buf && *p != '/'; --p)
2313 /* empty */ ;
2314 }
2315
2316 if (strncmp(p, "/jre/lib/", 9) != 0) {
2317 // Look for JAVA_HOME in the environment.
2318 char* java_home_var = ::getenv("JAVA_HOME");
2319 if (java_home_var != NULL && java_home_var[0] != 0) {
2320 char* jrelib_p;
2321 int len;
2322
2323 // Check the current module name "libjvm.so".
2324 p = strrchr(buf, '/');
2325 assert(strstr(p, "/libjvm") == p, "invalid library name");
2326
2327 rp = realpath(java_home_var, buf);
2328 if (rp == NULL)
2329 return;
2330
2331 // determine if this is a legacy image or modules image
2332 // modules image doesn't have "jre" subdirectory
2333 len = strlen(buf);
2334 jrelib_p = buf + len;
2335 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2336 if (0 != access(buf, F_OK)) {
2337 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2338 }
2339
2340 if (0 == access(buf, F_OK)) {
2341 // Use current module name "libjvm.so"
2342 len = strlen(buf);
2343 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2344 } else {
2345 // Go back to path of .so
2346 rp = realpath(dli_fname, buf);
2347 if (rp == NULL)
2348 return;
2349 }
2350 }
2351 }
2352 }
2353
2354 strcpy(saved_jvm_path, buf);
2355 }
2356
2357 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2358 // no prefix required, not even "_"
2359 }
2360
2361 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2362 // no suffix required
2363 }
2364
2365 ////////////////////////////////////////////////////////////////////////////////
2366 // sun.misc.Signal support
2367
2368 static volatile jint sigint_count = 0;
2369
2370 static void
2371 UserHandler(int sig, void *siginfo, void *context) {
2372 // 4511530 - sem_post is serialized and handled by the manager thread. When
2373 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2374 // don't want to flood the manager thread with sem_post requests.
2375 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2376 return;
2377
2378 // Ctrl-C is pressed during error reporting, likely because the error
2379 // handler fails to abort. Let VM die immediately.
2380 if (sig == SIGINT && is_error_reported()) {
2381 os::die();
2382 }
2383
2384 os::signal_notify(sig);
2385 }
2386
2387 void* os::user_handler() {
2388 return CAST_FROM_FN_PTR(void*, UserHandler);
2389 }
2390
2391 class Semaphore : public StackObj {
2392 public:
2393 Semaphore();
2394 ~Semaphore();
2395 void signal();
2396 void wait();
2397 bool trywait();
2398 bool timedwait(unsigned int sec, int nsec);
2399 private:
2400 sem_t _semaphore;
2401 };
2402
2403
2404 Semaphore::Semaphore() {
2405 sem_init(&_semaphore, 0, 0);
2406 }
2407
2408 Semaphore::~Semaphore() {
2409 sem_destroy(&_semaphore);
2410 }
2411
2412 void Semaphore::signal() {
2413 sem_post(&_semaphore);
2414 }
2415
2416 void Semaphore::wait() {
2417 sem_wait(&_semaphore);
2418 }
2419
2420 bool Semaphore::trywait() {
2421 return sem_trywait(&_semaphore) == 0;
2422 }
2423
2424 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2425 struct timespec ts;
2426 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2427
2428 while (1) {
2429 int result = sem_timedwait(&_semaphore, &ts);
2430 if (result == 0) {
2431 return true;
2432 } else if (errno == EINTR) {
2433 continue;
2434 } else if (errno == ETIMEDOUT) {
2435 return false;
2436 } else {
2437 return false;
2438 }
2439 }
2440 }
2441
2442 extern "C" {
2443 typedef void (*sa_handler_t)(int);
2444 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2445 }
2446
2447 void* os::signal(int signal_number, void* handler) {
2448 struct sigaction sigAct, oldSigAct;
2449
2450 sigfillset(&(sigAct.sa_mask));
2451 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2452 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2453
2454 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2455 // -1 means registration failed
2456 return (void *)-1;
2457 }
2458
2459 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2460 }
2461
2462 void os::signal_raise(int signal_number) {
2463 ::raise(signal_number);
2464 }
2465
2466 /*
2467 * The following code is moved from os.cpp for making this
2468 * code platform specific, which it is by its very nature.
2469 */
2470
2471 // Will be modified when max signal is changed to be dynamic
2472 int os::sigexitnum_pd() {
2473 return NSIG;
2474 }
2475
2476 // a counter for each possible signal value
2477 static volatile jint pending_signals[NSIG+1] = { 0 };
2478
2479 // Linux(POSIX) specific hand shaking semaphore.
2480 static sem_t sig_sem;
2481 static Semaphore sr_semaphore;
2482
2483 void os::signal_init_pd() {
2484 // Initialize signal structures
2485 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2486
2487 // Initialize signal semaphore
2488 ::sem_init(&sig_sem, 0, 0);
2489 }
2490
2491 void os::signal_notify(int sig) {
2492 Atomic::inc(&pending_signals[sig]);
2493 ::sem_post(&sig_sem);
2494 }
2495
2496 static int check_pending_signals(bool wait) {
2497 Atomic::store(0, &sigint_count);
2498 for (;;) {
2499 for (int i = 0; i < NSIG + 1; i++) {
2500 jint n = pending_signals[i];
2501 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2502 return i;
2503 }
2504 }
2505 if (!wait) {
2506 return -1;
2507 }
2508 JavaThread *thread = JavaThread::current();
2509 ThreadBlockInVM tbivm(thread);
2510
2511 bool threadIsSuspended;
2512 do {
2513 thread->set_suspend_equivalent();
2514 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2515 ::sem_wait(&sig_sem);
2516
2517 // were we externally suspended while we were waiting?
2518 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2519 if (threadIsSuspended) {
2520 //
2521 // The semaphore has been incremented, but while we were waiting
2522 // another thread suspended us. We don't want to continue running
2523 // while suspended because that would surprise the thread that
2524 // suspended us.
2525 //
2526 ::sem_post(&sig_sem);
2527
2528 thread->java_suspend_self();
2529 }
2530 } while (threadIsSuspended);
2531 }
2532 }
2533
2534 int os::signal_lookup() {
2535 return check_pending_signals(false);
2536 }
2537
2538 int os::signal_wait() {
2539 return check_pending_signals(true);
2540 }
2541
2542 ////////////////////////////////////////////////////////////////////////////////
2543 // Virtual Memory
2544
2545 int os::vm_page_size() {
2546 // Seems redundant as all get out
2547 assert(os::Linux::page_size() != -1, "must call os::init");
2548 return os::Linux::page_size();
2549 }
2550
2551 // Solaris allocates memory by pages.
2552 int os::vm_allocation_granularity() {
2553 assert(os::Linux::page_size() != -1, "must call os::init");
2554 return os::Linux::page_size();
2555 }
2556
2557 // Rationale behind this function:
2558 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2559 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2560 // samples for JITted code. Here we create private executable mapping over the code cache
2561 // and then we can use standard (well, almost, as mapping can change) way to provide
2562 // info for the reporting script by storing timestamp and location of symbol
2563 void linux_wrap_code(char* base, size_t size) {
2564 static volatile jint cnt = 0;
2565
2566 if (!UseOprofile) {
2567 return;
2568 }
2569
2570 char buf[PATH_MAX+1];
2571 int num = Atomic::add(1, &cnt);
2572
2573 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2574 os::get_temp_directory(), os::current_process_id(), num);
2575 unlink(buf);
2576
2577 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2578
2579 if (fd != -1) {
2580 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2581 if (rv != (off_t)-1) {
2582 if (::write(fd, "", 1) == 1) {
2583 mmap(base, size,
2584 PROT_READ|PROT_WRITE|PROT_EXEC,
2585 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2586 }
2587 }
2588 ::close(fd);
2589 unlink(buf);
2590 }
2591 }
2592
2593 static bool recoverable_mmap_error(int err) {
2594 // See if the error is one we can let the caller handle. This
2595 // list of errno values comes from JBS-6843484. I can't find a
2596 // Linux man page that documents this specific set of errno
2597 // values so while this list currently matches Solaris, it may
2598 // change as we gain experience with this failure mode.
2599 switch (err) {
2600 case EBADF:
2601 case EINVAL:
2602 case ENOTSUP:
2603 // let the caller deal with these errors
2604 return true;
2605
2606 default:
2607 // Any remaining errors on this OS can cause our reserved mapping
2608 // to be lost. That can cause confusion where different data
2609 // structures think they have the same memory mapped. The worst
2610 // scenario is if both the VM and a library think they have the
2611 // same memory mapped.
2612 return false;
2613 }
2614 }
2615
2616 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2617 int err) {
2618 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2619 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2620 strerror(err), err);
2621 }
2622
2623 static void warn_fail_commit_memory(char* addr, size_t size,
2624 size_t alignment_hint, bool exec,
2625 int err) {
2626 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2627 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2628 alignment_hint, exec, strerror(err), err);
2629 }
2630
2631 // NOTE: Linux kernel does not really reserve the pages for us.
2632 // All it does is to check if there are enough free pages
2633 // left at the time of mmap(). This could be a potential
2634 // problem.
2635 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2636 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2637 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2638 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2639 if (res != (uintptr_t) MAP_FAILED) {
2640 if (UseNUMAInterleaving) {
2641 numa_make_global(addr, size);
2642 }
2643 return 0;
2644 }
2645
2646 int err = errno; // save errno from mmap() call above
2647
2648 if (!recoverable_mmap_error(err)) {
2649 warn_fail_commit_memory(addr, size, exec, err);
2650 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2651 }
2652
2653 return err;
2654 }
2655
2656 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2657 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2658 }
2659
2660 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2661 const char* mesg) {
2662 assert(mesg != NULL, "mesg must be specified");
2663 int err = os::Linux::commit_memory_impl(addr, size, exec);
2664 if (err != 0) {
2665 // the caller wants all commit errors to exit with the specified mesg:
2666 warn_fail_commit_memory(addr, size, exec, err);
2667 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2668 }
2669 }
2670
2671 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2672 #ifndef MAP_HUGETLB
2673 #define MAP_HUGETLB 0x40000
2674 #endif
2675
2676 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2677 #ifndef MADV_HUGEPAGE
2678 #define MADV_HUGEPAGE 14
2679 #endif
2680
2681 int os::Linux::commit_memory_impl(char* addr, size_t size,
2682 size_t alignment_hint, bool exec) {
2683 int err = os::Linux::commit_memory_impl(addr, size, exec);
2684 if (err == 0) {
2685 realign_memory(addr, size, alignment_hint);
2686 }
2687 return err;
2688 }
2689
2690 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2691 bool exec) {
2692 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2693 }
2694
2695 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2696 size_t alignment_hint, bool exec,
2697 const char* mesg) {
2698 assert(mesg != NULL, "mesg must be specified");
2699 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2700 if (err != 0) {
2701 // the caller wants all commit errors to exit with the specified mesg:
2702 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2703 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2704 }
2705 }
2706
2707 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2708 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2709 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2710 // be supported or the memory may already be backed by huge pages.
2711 ::madvise(addr, bytes, MADV_HUGEPAGE);
2712 }
2713 }
2714
2715 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2716 // This method works by doing an mmap over an existing mmaping and effectively discarding
2717 // the existing pages. However it won't work for SHM-based large pages that cannot be
2718 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2719 // small pages on top of the SHM segment. This method always works for small pages, so we
2720 // allow that in any case.
2721 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2722 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2723 }
2724 }
2725
2726 void os::numa_make_global(char *addr, size_t bytes) {
2727 Linux::numa_interleave_memory(addr, bytes);
2728 }
2729
2730 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2731 // bind policy to MPOL_PREFERRED for the current thread.
2732 #define USE_MPOL_PREFERRED 0
2733
2734 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2735 // To make NUMA and large pages more robust when both enabled, we need to ease
2736 // the requirements on where the memory should be allocated. MPOL_BIND is the
2737 // default policy and it will force memory to be allocated on the specified
2738 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2739 // the specified node, but will not force it. Using this policy will prevent
2740 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2741 // free large pages.
2742 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2743 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2744 }
2745
2746 bool os::numa_topology_changed() { return false; }
2747
2748 size_t os::numa_get_groups_num() {
2749 int max_node = Linux::numa_max_node();
2750 return max_node > 0 ? max_node + 1 : 1;
2751 }
2752
2753 int os::numa_get_group_id() {
2754 int cpu_id = Linux::sched_getcpu();
2755 if (cpu_id != -1) {
2756 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2757 if (lgrp_id != -1) {
2758 return lgrp_id;
2759 }
2760 }
2761 return 0;
2762 }
2763
2764 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2765 for (size_t i = 0; i < size; i++) {
2766 ids[i] = i;
2767 }
2768 return size;
2769 }
2770
2771 bool os::get_page_info(char *start, page_info* info) {
2772 return false;
2773 }
2774
2775 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2776 return end;
2777 }
2778
2779
2780 int os::Linux::sched_getcpu_syscall(void) {
2781 unsigned int cpu;
2782 int retval = -1;
2783
2784 #if defined(IA32)
2785 # ifndef SYS_getcpu
2786 # define SYS_getcpu 318
2787 # endif
2788 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2789 #elif defined(AMD64)
2790 // Unfortunately we have to bring all these macros here from vsyscall.h
2791 // to be able to compile on old linuxes.
2792 # define __NR_vgetcpu 2
2793 # define VSYSCALL_START (-10UL << 20)
2794 # define VSYSCALL_SIZE 1024
2795 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2796 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2797 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2798 retval = vgetcpu(&cpu, NULL, NULL);
2799 #endif
2800
2801 return (retval == -1) ? retval : cpu;
2802 }
2803
2804 // Something to do with the numa-aware allocator needs these symbols
2805 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2806 extern "C" JNIEXPORT void numa_error(char *where) { }
2807 extern "C" JNIEXPORT int fork1() { return fork(); }
2808
2809
2810 // If we are running with libnuma version > 2, then we should
2811 // be trying to use symbols with versions 1.1
2812 // If we are running with earlier version, which did not have symbol versions,
2813 // we should use the base version.
2814 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2815 void *f = dlvsym(handle, name, "libnuma_1.1");
2816 if (f == NULL) {
2817 f = dlsym(handle, name);
2818 }
2819 return f;
2820 }
2821
2822 bool os::Linux::libnuma_init() {
2823 // sched_getcpu() should be in libc.
2824 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2825 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2826
2827 // If it's not, try a direct syscall.
2828 if (sched_getcpu() == -1)
2829 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2830
2831 if (sched_getcpu() != -1) { // Does it work?
2832 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2833 if (handle != NULL) {
2834 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2835 libnuma_dlsym(handle, "numa_node_to_cpus")));
2836 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2837 libnuma_dlsym(handle, "numa_max_node")));
2838 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2839 libnuma_dlsym(handle, "numa_available")));
2840 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2841 libnuma_dlsym(handle, "numa_tonode_memory")));
2842 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2843 libnuma_dlsym(handle, "numa_interleave_memory")));
2844 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2845 libnuma_dlsym(handle, "numa_set_bind_policy")));
2846
2847
2848 if (numa_available() != -1) {
2849 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2850 // Create a cpu -> node mapping
2851 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2852 rebuild_cpu_to_node_map();
2853 return true;
2854 }
2855 }
2856 }
2857 return false;
2858 }
2859
2860 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2861 // The table is later used in get_node_by_cpu().
2862 void os::Linux::rebuild_cpu_to_node_map() {
2863 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2864 // in libnuma (possible values are starting from 16,
2865 // and continuing up with every other power of 2, but less
2866 // than the maximum number of CPUs supported by kernel), and
2867 // is a subject to change (in libnuma version 2 the requirements
2868 // are more reasonable) we'll just hardcode the number they use
2869 // in the library.
2870 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2871
2872 size_t cpu_num = os::active_processor_count();
2873 size_t cpu_map_size = NCPUS / BitsPerCLong;
2874 size_t cpu_map_valid_size =
2875 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2876
2877 cpu_to_node()->clear();
2878 cpu_to_node()->at_grow(cpu_num - 1);
2879 size_t node_num = numa_get_groups_num();
2880
2881 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2882 for (size_t i = 0; i < node_num; i++) {
2883 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2884 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2885 if (cpu_map[j] != 0) {
2886 for (size_t k = 0; k < BitsPerCLong; k++) {
2887 if (cpu_map[j] & (1UL << k)) {
2888 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2889 }
2890 }
2891 }
2892 }
2893 }
2894 }
2895 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2896 }
2897
2898 int os::Linux::get_node_by_cpu(int cpu_id) {
2899 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2900 return cpu_to_node()->at(cpu_id);
2901 }
2902 return -1;
2903 }
2904
2905 GrowableArray<int>* os::Linux::_cpu_to_node;
2906 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2907 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2908 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2909 os::Linux::numa_available_func_t os::Linux::_numa_available;
2910 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2911 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2912 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2913 unsigned long* os::Linux::_numa_all_nodes;
2914
2915 bool os::pd_uncommit_memory(char* addr, size_t size) {
2916 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2917 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2918 return res != (uintptr_t) MAP_FAILED;
2919 }
2920
2921 static
2922 address get_stack_commited_bottom(address bottom, size_t size) {
2923 address nbot = bottom;
2924 address ntop = bottom + size;
2925
2926 size_t page_sz = os::vm_page_size();
2927 unsigned pages = size / page_sz;
2928
2929 unsigned char vec[1];
2930 unsigned imin = 1, imax = pages + 1, imid;
2931 int mincore_return_value = 0;
2932
2933 assert(imin <= imax, "Unexpected page size");
2934
2935 while (imin < imax) {
2936 imid = (imax + imin) / 2;
2937 nbot = ntop - (imid * page_sz);
2938
2939 // Use a trick with mincore to check whether the page is mapped or not.
2940 // mincore sets vec to 1 if page resides in memory and to 0 if page
2941 // is swapped output but if page we are asking for is unmapped
2942 // it returns -1,ENOMEM
2943 mincore_return_value = mincore(nbot, page_sz, vec);
2944
2945 if (mincore_return_value == -1) {
2946 // Page is not mapped go up
2947 // to find first mapped page
2948 if (errno != EAGAIN) {
2949 assert(errno == ENOMEM, "Unexpected mincore errno");
2950 imax = imid;
2951 }
2952 } else {
2953 // Page is mapped go down
2954 // to find first not mapped page
2955 imin = imid + 1;
2956 }
2957 }
2958
2959 nbot = nbot + page_sz;
2960
2961 // Adjust stack bottom one page up if last checked page is not mapped
2962 if (mincore_return_value == -1) {
2963 nbot = nbot + page_sz;
2964 }
2965
2966 return nbot;
2967 }
2968
2969
2970 // Linux uses a growable mapping for the stack, and if the mapping for
2971 // the stack guard pages is not removed when we detach a thread the
2972 // stack cannot grow beyond the pages where the stack guard was
2973 // mapped. If at some point later in the process the stack expands to
2974 // that point, the Linux kernel cannot expand the stack any further
2975 // because the guard pages are in the way, and a segfault occurs.
2976 //
2977 // However, it's essential not to split the stack region by unmapping
2978 // a region (leaving a hole) that's already part of the stack mapping,
2979 // so if the stack mapping has already grown beyond the guard pages at
2980 // the time we create them, we have to truncate the stack mapping.
2981 // So, we need to know the extent of the stack mapping when
2982 // create_stack_guard_pages() is called.
2983
2984 // We only need this for stacks that are growable: at the time of
2985 // writing thread stacks don't use growable mappings (i.e. those
2986 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2987 // only applies to the main thread.
2988
2989 // If the (growable) stack mapping already extends beyond the point
2990 // where we're going to put our guard pages, truncate the mapping at
2991 // that point by munmap()ping it. This ensures that when we later
2992 // munmap() the guard pages we don't leave a hole in the stack
2993 // mapping. This only affects the main/initial thread
2994
2995 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2996
2997 if (os::Linux::is_initial_thread()) {
2998 // As we manually grow stack up to bottom inside create_attached_thread(),
2999 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3000 // we don't need to do anything special.
3001 // Check it first, before calling heavy function.
3002 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3003 unsigned char vec[1];
3004
3005 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3006 // Fallback to slow path on all errors, including EAGAIN
3007 stack_extent = (uintptr_t) get_stack_commited_bottom(
3008 os::Linux::initial_thread_stack_bottom(),
3009 (size_t)addr - stack_extent);
3010 }
3011
3012 if (stack_extent < (uintptr_t)addr) {
3013 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3014 }
3015 }
3016
3017 return os::commit_memory(addr, size, !ExecMem);
3018 }
3019
3020 // If this is a growable mapping, remove the guard pages entirely by
3021 // munmap()ping them. If not, just call uncommit_memory(). This only
3022 // affects the main/initial thread, but guard against future OS changes
3023 // It's safe to always unmap guard pages for initial thread because we
3024 // always place it right after end of the mapped region
3025
3026 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3027 uintptr_t stack_extent, stack_base;
3028
3029 if (os::Linux::is_initial_thread()) {
3030 return ::munmap(addr, size) == 0;
3031 }
3032
3033 return os::uncommit_memory(addr, size);
3034 }
3035
3036 static address _highest_vm_reserved_address = NULL;
3037
3038 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3039 // at 'requested_addr'. If there are existing memory mappings at the same
3040 // location, however, they will be overwritten. If 'fixed' is false,
3041 // 'requested_addr' is only treated as a hint, the return value may or
3042 // may not start from the requested address. Unlike Linux mmap(), this
3043 // function returns NULL to indicate failure.
3044 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3045 char * addr;
3046 int flags;
3047
3048 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3049 if (fixed) {
3050 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3051 flags |= MAP_FIXED;
3052 }
3053
3054 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3055 // touch an uncommitted page. Otherwise, the read/write might
3056 // succeed if we have enough swap space to back the physical page.
3057 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3058 flags, -1, 0);
3059
3060 if (addr != MAP_FAILED) {
3061 // anon_mmap() should only get called during VM initialization,
3062 // don't need lock (actually we can skip locking even it can be called
3063 // from multiple threads, because _highest_vm_reserved_address is just a
3064 // hint about the upper limit of non-stack memory regions.)
3065 if ((address)addr + bytes > _highest_vm_reserved_address) {
3066 _highest_vm_reserved_address = (address)addr + bytes;
3067 }
3068 }
3069
3070 return addr == MAP_FAILED ? NULL : addr;
3071 }
3072
3073 // Don't update _highest_vm_reserved_address, because there might be memory
3074 // regions above addr + size. If so, releasing a memory region only creates
3075 // a hole in the address space, it doesn't help prevent heap-stack collision.
3076 //
3077 static int anon_munmap(char * addr, size_t size) {
3078 return ::munmap(addr, size) == 0;
3079 }
3080
3081 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3082 size_t alignment_hint) {
3083 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3084 }
3085
3086 bool os::pd_release_memory(char* addr, size_t size) {
3087 return anon_munmap(addr, size);
3088 }
3089
3090 static address highest_vm_reserved_address() {
3091 return _highest_vm_reserved_address;
3092 }
3093
3094 static bool linux_mprotect(char* addr, size_t size, int prot) {
3095 // Linux wants the mprotect address argument to be page aligned.
3096 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3097
3098 // According to SUSv3, mprotect() should only be used with mappings
3099 // established by mmap(), and mmap() always maps whole pages. Unaligned
3100 // 'addr' likely indicates problem in the VM (e.g. trying to change
3101 // protection of malloc'ed or statically allocated memory). Check the
3102 // caller if you hit this assert.
3103 assert(addr == bottom, "sanity check");
3104
3105 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3106 return ::mprotect(bottom, size, prot) == 0;
3107 }
3108
3109 // Set protections specified
3110 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3111 bool is_committed) {
3112 unsigned int p = 0;
3113 switch (prot) {
3114 case MEM_PROT_NONE: p = PROT_NONE; break;
3115 case MEM_PROT_READ: p = PROT_READ; break;
3116 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3117 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3118 default:
3119 ShouldNotReachHere();
3120 }
3121 // is_committed is unused.
3122 return linux_mprotect(addr, bytes, p);
3123 }
3124
3125 bool os::guard_memory(char* addr, size_t size) {
3126 return linux_mprotect(addr, size, PROT_NONE);
3127 }
3128
3129 bool os::unguard_memory(char* addr, size_t size) {
3130 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3131 }
3132
3133 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3134 bool result = false;
3135 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3136 MAP_ANONYMOUS|MAP_PRIVATE,
3137 -1, 0);
3138 if (p != MAP_FAILED) {
3139 void *aligned_p = align_ptr_up(p, page_size);
3140
3141 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3142
3143 munmap(p, page_size * 2);
3144 }
3145
3146 if (warn && !result) {
3147 warning("TransparentHugePages is not supported by the operating system.");
3148 }
3149
3150 return result;
3151 }
3152
3153 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3154 bool result = false;
3155 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3156 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3157 -1, 0);
3158
3159 if (p != MAP_FAILED) {
3160 // We don't know if this really is a huge page or not.
3161 FILE *fp = fopen("/proc/self/maps", "r");
3162 if (fp) {
3163 while (!feof(fp)) {
3164 char chars[257];
3165 long x = 0;
3166 if (fgets(chars, sizeof(chars), fp)) {
3167 if (sscanf(chars, "%lx-%*x", &x) == 1
3168 && x == (long)p) {
3169 if (strstr (chars, "hugepage")) {
3170 result = true;
3171 break;
3172 }
3173 }
3174 }
3175 }
3176 fclose(fp);
3177 }
3178 munmap(p, page_size);
3179 }
3180
3181 if (warn && !result) {
3182 warning("HugeTLBFS is not supported by the operating system.");
3183 }
3184
3185 return result;
3186 }
3187
3188 /*
3189 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3190 *
3191 * From the coredump_filter documentation:
3192 *
3193 * - (bit 0) anonymous private memory
3194 * - (bit 1) anonymous shared memory
3195 * - (bit 2) file-backed private memory
3196 * - (bit 3) file-backed shared memory
3197 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3198 * effective only if the bit 2 is cleared)
3199 * - (bit 5) hugetlb private memory
3200 * - (bit 6) hugetlb shared memory
3201 */
3202 static void set_coredump_filter(void) {
3203 FILE *f;
3204 long cdm;
3205
3206 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3207 return;
3208 }
3209
3210 if (fscanf(f, "%lx", &cdm) != 1) {
3211 fclose(f);
3212 return;
3213 }
3214
3215 rewind(f);
3216
3217 if ((cdm & LARGEPAGES_BIT) == 0) {
3218 cdm |= LARGEPAGES_BIT;
3219 fprintf(f, "%#lx", cdm);
3220 }
3221
3222 fclose(f);
3223 }
3224
3225 // Large page support
3226
3227 static size_t _large_page_size = 0;
3228
3229 size_t os::Linux::find_large_page_size() {
3230 size_t large_page_size = 0;
3231
3232 // large_page_size on Linux is used to round up heap size. x86 uses either
3233 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3234 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3235 // page as large as 256M.
3236 //
3237 // Here we try to figure out page size by parsing /proc/meminfo and looking
3238 // for a line with the following format:
3239 // Hugepagesize: 2048 kB
3240 //
3241 // If we can't determine the value (e.g. /proc is not mounted, or the text
3242 // format has been changed), we'll use the largest page size supported by
3243 // the processor.
3244
3245 #ifndef ZERO
3246 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3247 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3248 #endif // ZERO
3249
3250 FILE *fp = fopen("/proc/meminfo", "r");
3251 if (fp) {
3252 while (!feof(fp)) {
3253 int x = 0;
3254 char buf[16];
3255 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3256 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3257 large_page_size = x * K;
3258 break;
3259 }
3260 } else {
3261 // skip to next line
3262 for (;;) {
3263 int ch = fgetc(fp);
3264 if (ch == EOF || ch == (int)'\n') break;
3265 }
3266 }
3267 }
3268 fclose(fp);
3269 }
3270
3271 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3272 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3273 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3274 proper_unit_for_byte_size(large_page_size));
3275 }
3276
3277 return large_page_size;
3278 }
3279
3280 size_t os::Linux::setup_large_page_size() {
3281 _large_page_size = Linux::find_large_page_size();
3282 const size_t default_page_size = (size_t)Linux::page_size();
3283 if (_large_page_size > default_page_size) {
3284 _page_sizes[0] = _large_page_size;
3285 _page_sizes[1] = default_page_size;
3286 _page_sizes[2] = 0;
3287 }
3288
3289 return _large_page_size;
3290 }
3291
3292 bool os::Linux::setup_large_page_type(size_t page_size) {
3293 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3294 FLAG_IS_DEFAULT(UseSHM) &&
3295 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3296
3297 // The type of large pages has not been specified by the user.
3298
3299 // Try UseHugeTLBFS and then UseSHM.
3300 UseHugeTLBFS = UseSHM = true;
3301
3302 // Don't try UseTransparentHugePages since there are known
3303 // performance issues with it turned on. This might change in the future.
3304 UseTransparentHugePages = false;
3305 }
3306
3307 if (UseTransparentHugePages) {
3308 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3309 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3310 UseHugeTLBFS = false;
3311 UseSHM = false;
3312 return true;
3313 }
3314 UseTransparentHugePages = false;
3315 }
3316
3317 if (UseHugeTLBFS) {
3318 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3319 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3320 UseSHM = false;
3321 return true;
3322 }
3323 UseHugeTLBFS = false;
3324 }
3325
3326 return UseSHM;
3327 }
3328
3329 void os::large_page_init() {
3330 if (!UseLargePages &&
3331 !UseTransparentHugePages &&
3332 !UseHugeTLBFS &&
3333 !UseSHM) {
3334 // Not using large pages.
3335 return;
3336 }
3337
3338 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3339 // The user explicitly turned off large pages.
3340 // Ignore the rest of the large pages flags.
3341 UseTransparentHugePages = false;
3342 UseHugeTLBFS = false;
3343 UseSHM = false;
3344 return;
3345 }
3346
3347 size_t large_page_size = Linux::setup_large_page_size();
3348 UseLargePages = Linux::setup_large_page_type(large_page_size);
3349
3350 set_coredump_filter();
3351 }
3352
3353 #ifndef SHM_HUGETLB
3354 #define SHM_HUGETLB 04000
3355 #endif
3356
3357 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3358 // "exec" is passed in but not used. Creating the shared image for
3359 // the code cache doesn't have an SHM_X executable permission to check.
3360 assert(UseLargePages && UseSHM, "only for SHM large pages");
3361 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3362
3363 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3364 return NULL; // Fallback to small pages.
3365 }
3366
3367 key_t key = IPC_PRIVATE;
3368 char *addr;
3369
3370 bool warn_on_failure = UseLargePages &&
3371 (!FLAG_IS_DEFAULT(UseLargePages) ||
3372 !FLAG_IS_DEFAULT(UseSHM) ||
3373 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3374 );
3375 char msg[128];
3376
3377 // Create a large shared memory region to attach to based on size.
3378 // Currently, size is the total size of the heap
3379 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3380 if (shmid == -1) {
3381 // Possible reasons for shmget failure:
3382 // 1. shmmax is too small for Java heap.
3383 // > check shmmax value: cat /proc/sys/kernel/shmmax
3384 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3385 // 2. not enough large page memory.
3386 // > check available large pages: cat /proc/meminfo
3387 // > increase amount of large pages:
3388 // echo new_value > /proc/sys/vm/nr_hugepages
3389 // Note 1: different Linux may use different name for this property,
3390 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3391 // Note 2: it's possible there's enough physical memory available but
3392 // they are so fragmented after a long run that they can't
3393 // coalesce into large pages. Try to reserve large pages when
3394 // the system is still "fresh".
3395 if (warn_on_failure) {
3396 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3397 warning(msg);
3398 }
3399 return NULL;
3400 }
3401
3402 // attach to the region
3403 addr = (char*)shmat(shmid, req_addr, 0);
3404 int err = errno;
3405
3406 // Remove shmid. If shmat() is successful, the actual shared memory segment
3407 // will be deleted when it's detached by shmdt() or when the process
3408 // terminates. If shmat() is not successful this will remove the shared
3409 // segment immediately.
3410 shmctl(shmid, IPC_RMID, NULL);
3411
3412 if ((intptr_t)addr == -1) {
3413 if (warn_on_failure) {
3414 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3415 warning(msg);
3416 }
3417 return NULL;
3418 }
3419
3420 return addr;
3421 }
3422
3423 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3424 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3425
3426 bool warn_on_failure = UseLargePages &&
3427 (!FLAG_IS_DEFAULT(UseLargePages) ||
3428 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3429 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3430
3431 if (warn_on_failure) {
3432 char msg[128];
3433 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3434 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3435 warning(msg);
3436 }
3437 }
3438
3439 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3440 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3441 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3442 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3443
3444 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3445 char* addr = (char*)::mmap(req_addr, bytes, prot,
3446 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3447 -1, 0);
3448
3449 if (addr == MAP_FAILED) {
3450 warn_on_large_pages_failure(req_addr, bytes, errno);
3451 return NULL;
3452 }
3453
3454 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3455
3456 return addr;
3457 }
3458
3459 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3460 size_t large_page_size = os::large_page_size();
3461
3462 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3463
3464 // Allocate small pages.
3465
3466 char* start;
3467 if (req_addr != NULL) {
3468 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3469 assert(is_size_aligned(bytes, alignment), "Must be");
3470 start = os::reserve_memory(bytes, req_addr);
3471 assert(start == NULL || start == req_addr, "Must be");
3472 } else {
3473 start = os::reserve_memory_aligned(bytes, alignment);
3474 }
3475
3476 if (start == NULL) {
3477 return NULL;
3478 }
3479
3480 assert(is_ptr_aligned(start, alignment), "Must be");
3481
3482 // os::reserve_memory_special will record this memory area.
3483 // Need to release it here to prevent overlapping reservations.
3484 MemTracker::record_virtual_memory_release((address)start, bytes);
3485
3486 char* end = start + bytes;
3487
3488 // Find the regions of the allocated chunk that can be promoted to large pages.
3489 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3490 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3491
3492 size_t lp_bytes = lp_end - lp_start;
3493
3494 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3495
3496 if (lp_bytes == 0) {
3497 // The mapped region doesn't even span the start and the end of a large page.
3498 // Fall back to allocate a non-special area.
3499 ::munmap(start, end - start);
3500 return NULL;
3501 }
3502
3503 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3504
3505
3506 void* result;
3507
3508 if (start != lp_start) {
3509 result = ::mmap(start, lp_start - start, prot,
3510 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3511 -1, 0);
3512 if (result == MAP_FAILED) {
3513 ::munmap(lp_start, end - lp_start);
3514 return NULL;
3515 }
3516 }
3517
3518 result = ::mmap(lp_start, lp_bytes, prot,
3519 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3520 -1, 0);
3521 if (result == MAP_FAILED) {
3522 warn_on_large_pages_failure(req_addr, bytes, errno);
3523 // If the mmap above fails, the large pages region will be unmapped and we
3524 // have regions before and after with small pages. Release these regions.
3525 //
3526 // | mapped | unmapped | mapped |
3527 // ^ ^ ^ ^
3528 // start lp_start lp_end end
3529 //
3530 ::munmap(start, lp_start - start);
3531 ::munmap(lp_end, end - lp_end);
3532 return NULL;
3533 }
3534
3535 if (lp_end != end) {
3536 result = ::mmap(lp_end, end - lp_end, prot,
3537 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3538 -1, 0);
3539 if (result == MAP_FAILED) {
3540 ::munmap(start, lp_end - start);
3541 return NULL;
3542 }
3543 }
3544
3545 return start;
3546 }
3547
3548 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3549 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3550 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3551 assert(is_power_of_2(alignment), "Must be");
3552 assert(is_power_of_2(os::large_page_size()), "Must be");
3553 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3554
3555 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3556 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3557 } else {
3558 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3559 }
3560 }
3561
3562 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3563 assert(UseLargePages, "only for large pages");
3564
3565 char* addr;
3566 if (UseSHM) {
3567 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3568 } else {
3569 assert(UseHugeTLBFS, "must be");
3570 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3571 }
3572
3573 if (addr != NULL) {
3574 if (UseNUMAInterleaving) {
3575 numa_make_global(addr, bytes);
3576 }
3577
3578 // The memory is committed
3579 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3580 }
3581
3582 return addr;
3583 }
3584
3585 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3586 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3587 return shmdt(base) == 0;
3588 }
3589
3590 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3591 return pd_release_memory(base, bytes);
3592 }
3593
3594 bool os::release_memory_special(char* base, size_t bytes) {
3595 assert(UseLargePages, "only for large pages");
3596
3597 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3598
3599 bool res;
3600 if (UseSHM) {
3601 res = os::Linux::release_memory_special_shm(base, bytes);
3602 } else {
3603 assert(UseHugeTLBFS, "must be");
3604 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3605 }
3606
3607 if (res) {
3608 tkr.record((address)base, bytes);
3609 } else {
3610 tkr.discard();
3611 }
3612
3613 return res;
3614 }
3615
3616 size_t os::large_page_size() {
3617 return _large_page_size;
3618 }
3619
3620 // With SysV SHM the entire memory region must be allocated as shared
3621 // memory.
3622 // HugeTLBFS allows application to commit large page memory on demand.
3623 // However, when committing memory with HugeTLBFS fails, the region
3624 // that was supposed to be committed will lose the old reservation
3625 // and allow other threads to steal that memory region. Because of this
3626 // behavior we can't commit HugeTLBFS memory.
3627 bool os::can_commit_large_page_memory() {
3628 return UseTransparentHugePages;
3629 }
3630
3631 bool os::can_execute_large_page_memory() {
3632 return UseTransparentHugePages || UseHugeTLBFS;
3633 }
3634
3635 // Reserve memory at an arbitrary address, only if that area is
3636 // available (and not reserved for something else).
3637
3638 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3639 const int max_tries = 10;
3640 char* base[max_tries];
3641 size_t size[max_tries];
3642 const size_t gap = 0x000000;
3643
3644 // Assert only that the size is a multiple of the page size, since
3645 // that's all that mmap requires, and since that's all we really know
3646 // about at this low abstraction level. If we need higher alignment,
3647 // we can either pass an alignment to this method or verify alignment
3648 // in one of the methods further up the call chain. See bug 5044738.
3649 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3650
3651 // Repeatedly allocate blocks until the block is allocated at the
3652 // right spot. Give up after max_tries. Note that reserve_memory() will
3653 // automatically update _highest_vm_reserved_address if the call is
3654 // successful. The variable tracks the highest memory address every reserved
3655 // by JVM. It is used to detect heap-stack collision if running with
3656 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3657 // space than needed, it could confuse the collision detecting code. To
3658 // solve the problem, save current _highest_vm_reserved_address and
3659 // calculate the correct value before return.
3660 address old_highest = _highest_vm_reserved_address;
3661
3662 // Linux mmap allows caller to pass an address as hint; give it a try first,
3663 // if kernel honors the hint then we can return immediately.
3664 char * addr = anon_mmap(requested_addr, bytes, false);
3665 if (addr == requested_addr) {
3666 return requested_addr;
3667 }
3668
3669 if (addr != NULL) {
3670 // mmap() is successful but it fails to reserve at the requested address
3671 anon_munmap(addr, bytes);
3672 }
3673
3674 int i;
3675 for (i = 0; i < max_tries; ++i) {
3676 base[i] = reserve_memory(bytes);
3677
3678 if (base[i] != NULL) {
3679 // Is this the block we wanted?
3680 if (base[i] == requested_addr) {
3681 size[i] = bytes;
3682 break;
3683 }
3684
3685 // Does this overlap the block we wanted? Give back the overlapped
3686 // parts and try again.
3687
3688 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3689 if (top_overlap >= 0 && top_overlap < bytes) {
3690 unmap_memory(base[i], top_overlap);
3691 base[i] += top_overlap;
3692 size[i] = bytes - top_overlap;
3693 } else {
3694 size_t bottom_overlap = base[i] + bytes - requested_addr;
3695 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3696 unmap_memory(requested_addr, bottom_overlap);
3697 size[i] = bytes - bottom_overlap;
3698 } else {
3699 size[i] = bytes;
3700 }
3701 }
3702 }
3703 }
3704
3705 // Give back the unused reserved pieces.
3706
3707 for (int j = 0; j < i; ++j) {
3708 if (base[j] != NULL) {
3709 unmap_memory(base[j], size[j]);
3710 }
3711 }
3712
3713 if (i < max_tries) {
3714 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3715 return requested_addr;
3716 } else {
3717 _highest_vm_reserved_address = old_highest;
3718 return NULL;
3719 }
3720 }
3721
3722 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3723 return ::read(fd, buf, nBytes);
3724 }
3725
3726 //
3727 // Short sleep, direct OS call.
3728 //
3729 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3730 // sched_yield(2) will actually give up the CPU:
3731 //
3732 // * Alone on this pariticular CPU, keeps running.
3733 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3734 // (pre 2.6.39).
3735 //
3736 // So calling this with 0 is an alternative.
3737 //
3738 void os::naked_short_sleep(jlong ms) {
3739 struct timespec req;
3740
3741 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3742 req.tv_sec = 0;
3743 if (ms > 0) {
3744 req.tv_nsec = (ms % 1000) * 1000000;
3745 }
3746 else {
3747 req.tv_nsec = 1;
3748 }
3749
3750 nanosleep(&req, NULL);
3751
3752 return;
3753 }
3754
3755 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3756 void os::infinite_sleep() {
3757 while (true) { // sleep forever ...
3758 ::sleep(100); // ... 100 seconds at a time
3759 }
3760 }
3761
3762 // Used to convert frequent JVM_Yield() to nops
3763 bool os::dont_yield() {
3764 return DontYieldALot;
3765 }
3766
3767 void os::yield() {
3768 sched_yield();
3769 }
3770
3771 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3772
3773 void os::yield_all(int attempts) {
3774 // Yields to all threads, including threads with lower priorities
3775 // Threads on Linux are all with same priority. The Solaris style
3776 // os::yield_all() with nanosleep(1ms) is not necessary.
3777 sched_yield();
3778 }
3779
3780 // Called from the tight loops to possibly influence time-sharing heuristics
3781 void os::loop_breaker(int attempts) {
3782 os::yield_all(attempts);
3783 }
3784
3785 ////////////////////////////////////////////////////////////////////////////////
3786 // thread priority support
3787
3788 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3789 // only supports dynamic priority, static priority must be zero. For real-time
3790 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3791 // However, for large multi-threaded applications, SCHED_RR is not only slower
3792 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3793 // of 5 runs - Sep 2005).
3794 //
3795 // The following code actually changes the niceness of kernel-thread/LWP. It
3796 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3797 // not the entire user process, and user level threads are 1:1 mapped to kernel
3798 // threads. It has always been the case, but could change in the future. For
3799 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3800 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3801
3802 int os::java_to_os_priority[CriticalPriority + 1] = {
3803 19, // 0 Entry should never be used
3804
3805 4, // 1 MinPriority
3806 3, // 2
3807 2, // 3
3808
3809 1, // 4
3810 0, // 5 NormPriority
3811 -1, // 6
3812
3813 -2, // 7
3814 -3, // 8
3815 -4, // 9 NearMaxPriority
3816
3817 -5, // 10 MaxPriority
3818
3819 -5 // 11 CriticalPriority
3820 };
3821
3822 static int prio_init() {
3823 if (ThreadPriorityPolicy == 1) {
3824 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3825 // if effective uid is not root. Perhaps, a more elegant way of doing
3826 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3827 if (geteuid() != 0) {
3828 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3829 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3830 }
3831 ThreadPriorityPolicy = 0;
3832 }
3833 }
3834 if (UseCriticalJavaThreadPriority) {
3835 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3836 }
3837 return 0;
3838 }
3839
3840 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3841 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3842
3843 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3844 return (ret == 0) ? OS_OK : OS_ERR;
3845 }
3846
3847 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3848 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3849 *priority_ptr = java_to_os_priority[NormPriority];
3850 return OS_OK;
3851 }
3852
3853 errno = 0;
3854 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3855 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3856 }
3857
3858 // Hint to the underlying OS that a task switch would not be good.
3859 // Void return because it's a hint and can fail.
3860 void os::hint_no_preempt() {}
3861
3862 ////////////////////////////////////////////////////////////////////////////////
3863 // suspend/resume support
3864
3865 // the low-level signal-based suspend/resume support is a remnant from the
3866 // old VM-suspension that used to be for java-suspension, safepoints etc,
3867 // within hotspot. Now there is a single use-case for this:
3868 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3869 // that runs in the watcher thread.
3870 // The remaining code is greatly simplified from the more general suspension
3871 // code that used to be used.
3872 //
3873 // The protocol is quite simple:
3874 // - suspend:
3875 // - sends a signal to the target thread
3876 // - polls the suspend state of the osthread using a yield loop
3877 // - target thread signal handler (SR_handler) sets suspend state
3878 // and blocks in sigsuspend until continued
3879 // - resume:
3880 // - sets target osthread state to continue
3881 // - sends signal to end the sigsuspend loop in the SR_handler
3882 //
3883 // Note that the SR_lock plays no role in this suspend/resume protocol.
3884 //
3885
3886 static void resume_clear_context(OSThread *osthread) {
3887 osthread->set_ucontext(NULL);
3888 osthread->set_siginfo(NULL);
3889 }
3890
3891 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3892 osthread->set_ucontext(context);
3893 osthread->set_siginfo(siginfo);
3894 }
3895
3896 //
3897 // Handler function invoked when a thread's execution is suspended or
3898 // resumed. We have to be careful that only async-safe functions are
3899 // called here (Note: most pthread functions are not async safe and
3900 // should be avoided.)
3901 //
3902 // Note: sigwait() is a more natural fit than sigsuspend() from an
3903 // interface point of view, but sigwait() prevents the signal hander
3904 // from being run. libpthread would get very confused by not having
3905 // its signal handlers run and prevents sigwait()'s use with the
3906 // mutex granting granting signal.
3907 //
3908 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3909 //
3910 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3911 // Save and restore errno to avoid confusing native code with EINTR
3912 // after sigsuspend.
3913 int old_errno = errno;
3914
3915 Thread* thread = Thread::current();
3916 OSThread* osthread = thread->osthread();
3917 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3918
3919 os::SuspendResume::State current = osthread->sr.state();
3920 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3921 suspend_save_context(osthread, siginfo, context);
3922
3923 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3924 os::SuspendResume::State state = osthread->sr.suspended();
3925 if (state == os::SuspendResume::SR_SUSPENDED) {
3926 sigset_t suspend_set; // signals for sigsuspend()
3927
3928 // get current set of blocked signals and unblock resume signal
3929 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3930 sigdelset(&suspend_set, SR_signum);
3931
3932 sr_semaphore.signal();
3933 // wait here until we are resumed
3934 while (1) {
3935 sigsuspend(&suspend_set);
3936
3937 os::SuspendResume::State result = osthread->sr.running();
3938 if (result == os::SuspendResume::SR_RUNNING) {
3939 sr_semaphore.signal();
3940 break;
3941 }
3942 }
3943
3944 } else if (state == os::SuspendResume::SR_RUNNING) {
3945 // request was cancelled, continue
3946 } else {
3947 ShouldNotReachHere();
3948 }
3949
3950 resume_clear_context(osthread);
3951 } else if (current == os::SuspendResume::SR_RUNNING) {
3952 // request was cancelled, continue
3953 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3954 // ignore
3955 } else {
3956 // ignore
3957 }
3958
3959 errno = old_errno;
3960 }
3961
3962
3963 static int SR_initialize() {
3964 struct sigaction act;
3965 char *s;
3966 /* Get signal number to use for suspend/resume */
3967 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3968 int sig = ::strtol(s, 0, 10);
3969 if (sig > 0 || sig < _NSIG) {
3970 SR_signum = sig;
3971 }
3972 }
3973
3974 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3975 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3976
3977 sigemptyset(&SR_sigset);
3978 sigaddset(&SR_sigset, SR_signum);
3979
3980 /* Set up signal handler for suspend/resume */
3981 act.sa_flags = SA_RESTART|SA_SIGINFO;
3982 act.sa_handler = (void (*)(int)) SR_handler;
3983
3984 // SR_signum is blocked by default.
3985 // 4528190 - We also need to block pthread restart signal (32 on all
3986 // supported Linux platforms). Note that LinuxThreads need to block
3987 // this signal for all threads to work properly. So we don't have
3988 // to use hard-coded signal number when setting up the mask.
3989 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3990
3991 if (sigaction(SR_signum, &act, 0) == -1) {
3992 return -1;
3993 }
3994
3995 // Save signal flag
3996 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3997 return 0;
3998 }
3999
4000 static int sr_notify(OSThread* osthread) {
4001 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4002 assert_status(status == 0, status, "pthread_kill");
4003 return status;
4004 }
4005
4006 // "Randomly" selected value for how long we want to spin
4007 // before bailing out on suspending a thread, also how often
4008 // we send a signal to a thread we want to resume
4009 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4010 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4011
4012 // returns true on success and false on error - really an error is fatal
4013 // but this seems the normal response to library errors
4014 static bool do_suspend(OSThread* osthread) {
4015 assert(osthread->sr.is_running(), "thread should be running");
4016 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4017
4018 // mark as suspended and send signal
4019 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4020 // failed to switch, state wasn't running?
4021 ShouldNotReachHere();
4022 return false;
4023 }
4024
4025 if (sr_notify(osthread) != 0) {
4026 ShouldNotReachHere();
4027 }
4028
4029 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4030 while (true) {
4031 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4032 break;
4033 } else {
4034 // timeout
4035 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4036 if (cancelled == os::SuspendResume::SR_RUNNING) {
4037 return false;
4038 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4039 // make sure that we consume the signal on the semaphore as well
4040 sr_semaphore.wait();
4041 break;
4042 } else {
4043 ShouldNotReachHere();
4044 return false;
4045 }
4046 }
4047 }
4048
4049 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4050 return true;
4051 }
4052
4053 static void do_resume(OSThread* osthread) {
4054 assert(osthread->sr.is_suspended(), "thread should be suspended");
4055 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4056
4057 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4058 // failed to switch to WAKEUP_REQUEST
4059 ShouldNotReachHere();
4060 return;
4061 }
4062
4063 while (true) {
4064 if (sr_notify(osthread) == 0) {
4065 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4066 if (osthread->sr.is_running()) {
4067 return;
4068 }
4069 }
4070 } else {
4071 ShouldNotReachHere();
4072 }
4073 }
4074
4075 guarantee(osthread->sr.is_running(), "Must be running!");
4076 }
4077
4078 ///////////////////////////////////////////////////////////////////////////////////
4079 // signal handling (except suspend/resume)
4080
4081 // This routine may be used by user applications as a "hook" to catch signals.
4082 // The user-defined signal handler must pass unrecognized signals to this
4083 // routine, and if it returns true (non-zero), then the signal handler must
4084 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4085 // routine will never retun false (zero), but instead will execute a VM panic
4086 // routine kill the process.
4087 //
4088 // If this routine returns false, it is OK to call it again. This allows
4089 // the user-defined signal handler to perform checks either before or after
4090 // the VM performs its own checks. Naturally, the user code would be making
4091 // a serious error if it tried to handle an exception (such as a null check
4092 // or breakpoint) that the VM was generating for its own correct operation.
4093 //
4094 // This routine may recognize any of the following kinds of signals:
4095 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4096 // It should be consulted by handlers for any of those signals.
4097 //
4098 // The caller of this routine must pass in the three arguments supplied
4099 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4100 // field of the structure passed to sigaction(). This routine assumes that
4101 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4102 //
4103 // Note that the VM will print warnings if it detects conflicting signal
4104 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4105 //
4106 extern "C" JNIEXPORT int
4107 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4108 void* ucontext, int abort_if_unrecognized);
4109
4110 void signalHandler(int sig, siginfo_t* info, void* uc) {
4111 assert(info != NULL && uc != NULL, "it must be old kernel");
4112 int orig_errno = errno; // Preserve errno value over signal handler.
4113 JVM_handle_linux_signal(sig, info, uc, true);
4114 errno = orig_errno;
4115 }
4116
4117
4118 // This boolean allows users to forward their own non-matching signals
4119 // to JVM_handle_linux_signal, harmlessly.
4120 bool os::Linux::signal_handlers_are_installed = false;
4121
4122 // For signal-chaining
4123 struct sigaction os::Linux::sigact[MAXSIGNUM];
4124 unsigned int os::Linux::sigs = 0;
4125 bool os::Linux::libjsig_is_loaded = false;
4126 typedef struct sigaction *(*get_signal_t)(int);
4127 get_signal_t os::Linux::get_signal_action = NULL;
4128
4129 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4130 struct sigaction *actp = NULL;
4131
4132 if (libjsig_is_loaded) {
4133 // Retrieve the old signal handler from libjsig
4134 actp = (*get_signal_action)(sig);
4135 }
4136 if (actp == NULL) {
4137 // Retrieve the preinstalled signal handler from jvm
4138 actp = get_preinstalled_handler(sig);
4139 }
4140
4141 return actp;
4142 }
4143
4144 static bool call_chained_handler(struct sigaction *actp, int sig,
4145 siginfo_t *siginfo, void *context) {
4146 // Call the old signal handler
4147 if (actp->sa_handler == SIG_DFL) {
4148 // It's more reasonable to let jvm treat it as an unexpected exception
4149 // instead of taking the default action.
4150 return false;
4151 } else if (actp->sa_handler != SIG_IGN) {
4152 if ((actp->sa_flags & SA_NODEFER) == 0) {
4153 // automaticlly block the signal
4154 sigaddset(&(actp->sa_mask), sig);
4155 }
4156
4157 sa_handler_t hand;
4158 sa_sigaction_t sa;
4159 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4160 // retrieve the chained handler
4161 if (siginfo_flag_set) {
4162 sa = actp->sa_sigaction;
4163 } else {
4164 hand = actp->sa_handler;
4165 }
4166
4167 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4168 actp->sa_handler = SIG_DFL;
4169 }
4170
4171 // try to honor the signal mask
4172 sigset_t oset;
4173 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4174
4175 // call into the chained handler
4176 if (siginfo_flag_set) {
4177 (*sa)(sig, siginfo, context);
4178 } else {
4179 (*hand)(sig);
4180 }
4181
4182 // restore the signal mask
4183 pthread_sigmask(SIG_SETMASK, &oset, 0);
4184 }
4185 // Tell jvm's signal handler the signal is taken care of.
4186 return true;
4187 }
4188
4189 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4190 bool chained = false;
4191 // signal-chaining
4192 if (UseSignalChaining) {
4193 struct sigaction *actp = get_chained_signal_action(sig);
4194 if (actp != NULL) {
4195 chained = call_chained_handler(actp, sig, siginfo, context);
4196 }
4197 }
4198 return chained;
4199 }
4200
4201 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4202 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4203 return &sigact[sig];
4204 }
4205 return NULL;
4206 }
4207
4208 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4209 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4210 sigact[sig] = oldAct;
4211 sigs |= (unsigned int)1 << sig;
4212 }
4213
4214 // for diagnostic
4215 int os::Linux::sigflags[MAXSIGNUM];
4216
4217 int os::Linux::get_our_sigflags(int sig) {
4218 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4219 return sigflags[sig];
4220 }
4221
4222 void os::Linux::set_our_sigflags(int sig, int flags) {
4223 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4224 sigflags[sig] = flags;
4225 }
4226
4227 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4228 // Check for overwrite.
4229 struct sigaction oldAct;
4230 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4231
4232 void* oldhand = oldAct.sa_sigaction
4233 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4234 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4235 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4236 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4237 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4238 if (AllowUserSignalHandlers || !set_installed) {
4239 // Do not overwrite; user takes responsibility to forward to us.
4240 return;
4241 } else if (UseSignalChaining) {
4242 // save the old handler in jvm
4243 save_preinstalled_handler(sig, oldAct);
4244 // libjsig also interposes the sigaction() call below and saves the
4245 // old sigaction on it own.
4246 } else {
4247 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4248 "%#lx for signal %d.", (long)oldhand, sig));
4249 }
4250 }
4251
4252 struct sigaction sigAct;
4253 sigfillset(&(sigAct.sa_mask));
4254 sigAct.sa_handler = SIG_DFL;
4255 if (!set_installed) {
4256 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4257 } else {
4258 sigAct.sa_sigaction = signalHandler;
4259 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4260 }
4261 // Save flags, which are set by ours
4262 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4263 sigflags[sig] = sigAct.sa_flags;
4264
4265 int ret = sigaction(sig, &sigAct, &oldAct);
4266 assert(ret == 0, "check");
4267
4268 void* oldhand2 = oldAct.sa_sigaction
4269 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4270 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4271 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4272 }
4273
4274 // install signal handlers for signals that HotSpot needs to
4275 // handle in order to support Java-level exception handling.
4276
4277 void os::Linux::install_signal_handlers() {
4278 if (!signal_handlers_are_installed) {
4279 signal_handlers_are_installed = true;
4280
4281 // signal-chaining
4282 typedef void (*signal_setting_t)();
4283 signal_setting_t begin_signal_setting = NULL;
4284 signal_setting_t end_signal_setting = NULL;
4285 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4286 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4287 if (begin_signal_setting != NULL) {
4288 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4289 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4290 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4291 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4292 libjsig_is_loaded = true;
4293 assert(UseSignalChaining, "should enable signal-chaining");
4294 }
4295 if (libjsig_is_loaded) {
4296 // Tell libjsig jvm is setting signal handlers
4297 (*begin_signal_setting)();
4298 }
4299
4300 set_signal_handler(SIGSEGV, true);
4301 set_signal_handler(SIGPIPE, true);
4302 set_signal_handler(SIGBUS, true);
4303 set_signal_handler(SIGILL, true);
4304 set_signal_handler(SIGFPE, true);
4305 #if defined(PPC64)
4306 set_signal_handler(SIGTRAP, true);
4307 #endif
4308 set_signal_handler(SIGXFSZ, true);
4309
4310 if (libjsig_is_loaded) {
4311 // Tell libjsig jvm finishes setting signal handlers
4312 (*end_signal_setting)();
4313 }
4314
4315 // We don't activate signal checker if libjsig is in place, we trust ourselves
4316 // and if UserSignalHandler is installed all bets are off.
4317 // Log that signal checking is off only if -verbose:jni is specified.
4318 if (CheckJNICalls) {
4319 if (libjsig_is_loaded) {
4320 if (PrintJNIResolving) {
4321 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4322 }
4323 check_signals = false;
4324 }
4325 if (AllowUserSignalHandlers) {
4326 if (PrintJNIResolving) {
4327 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4328 }
4329 check_signals = false;
4330 }
4331 }
4332 }
4333 }
4334
4335 // This is the fastest way to get thread cpu time on Linux.
4336 // Returns cpu time (user+sys) for any thread, not only for current.
4337 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4338 // It might work on 2.6.10+ with a special kernel/glibc patch.
4339 // For reference, please, see IEEE Std 1003.1-2004:
4340 // http://www.unix.org/single_unix_specification
4341
4342 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4343 struct timespec tp;
4344 int rc = os::Linux::clock_gettime(clockid, &tp);
4345 assert(rc == 0, "clock_gettime is expected to return 0 code");
4346
4347 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4348 }
4349
4350 /////
4351 // glibc on Linux platform uses non-documented flag
4352 // to indicate, that some special sort of signal
4353 // trampoline is used.
4354 // We will never set this flag, and we should
4355 // ignore this flag in our diagnostic
4356 #ifdef SIGNIFICANT_SIGNAL_MASK
4357 #undef SIGNIFICANT_SIGNAL_MASK
4358 #endif
4359 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4360
4361 static const char* get_signal_handler_name(address handler,
4362 char* buf, int buflen) {
4363 int offset;
4364 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4365 if (found) {
4366 // skip directory names
4367 const char *p1, *p2;
4368 p1 = buf;
4369 size_t len = strlen(os::file_separator());
4370 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4371 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4372 } else {
4373 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4374 }
4375 return buf;
4376 }
4377
4378 static void print_signal_handler(outputStream* st, int sig,
4379 char* buf, size_t buflen) {
4380 struct sigaction sa;
4381
4382 sigaction(sig, NULL, &sa);
4383
4384 // See comment for SIGNIFICANT_SIGNAL_MASK define
4385 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4386
4387 st->print("%s: ", os::exception_name(sig, buf, buflen));
4388
4389 address handler = (sa.sa_flags & SA_SIGINFO)
4390 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4391 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4392
4393 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4394 st->print("SIG_DFL");
4395 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4396 st->print("SIG_IGN");
4397 } else {
4398 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4399 }
4400
4401 st->print(", sa_mask[0]=");
4402 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4403
4404 address rh = VMError::get_resetted_sighandler(sig);
4405 // May be, handler was resetted by VMError?
4406 if(rh != NULL) {
4407 handler = rh;
4408 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4409 }
4410
4411 st->print(", sa_flags=");
4412 os::Posix::print_sa_flags(st, sa.sa_flags);
4413
4414 // Check: is it our handler?
4415 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4416 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4417 // It is our signal handler
4418 // check for flags, reset system-used one!
4419 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4420 st->print(
4421 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4422 os::Linux::get_our_sigflags(sig));
4423 }
4424 }
4425 st->cr();
4426 }
4427
4428
4429 #define DO_SIGNAL_CHECK(sig) \
4430 if (!sigismember(&check_signal_done, sig)) \
4431 os::Linux::check_signal_handler(sig)
4432
4433 // This method is a periodic task to check for misbehaving JNI applications
4434 // under CheckJNI, we can add any periodic checks here
4435
4436 void os::run_periodic_checks() {
4437
4438 if (check_signals == false) return;
4439
4440 // SEGV and BUS if overridden could potentially prevent
4441 // generation of hs*.log in the event of a crash, debugging
4442 // such a case can be very challenging, so we absolutely
4443 // check the following for a good measure:
4444 DO_SIGNAL_CHECK(SIGSEGV);
4445 DO_SIGNAL_CHECK(SIGILL);
4446 DO_SIGNAL_CHECK(SIGFPE);
4447 DO_SIGNAL_CHECK(SIGBUS);
4448 DO_SIGNAL_CHECK(SIGPIPE);
4449 DO_SIGNAL_CHECK(SIGXFSZ);
4450 #if defined(PPC64)
4451 DO_SIGNAL_CHECK(SIGTRAP);
4452 #endif
4453
4454 // ReduceSignalUsage allows the user to override these handlers
4455 // see comments at the very top and jvm_solaris.h
4456 if (!ReduceSignalUsage) {
4457 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4458 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4459 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4460 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4461 }
4462
4463 DO_SIGNAL_CHECK(SR_signum);
4464 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4465 }
4466
4467 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4468
4469 static os_sigaction_t os_sigaction = NULL;
4470
4471 void os::Linux::check_signal_handler(int sig) {
4472 char buf[O_BUFLEN];
4473 address jvmHandler = NULL;
4474
4475
4476 struct sigaction act;
4477 if (os_sigaction == NULL) {
4478 // only trust the default sigaction, in case it has been interposed
4479 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4480 if (os_sigaction == NULL) return;
4481 }
4482
4483 os_sigaction(sig, (struct sigaction*)NULL, &act);
4484
4485
4486 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4487
4488 address thisHandler = (act.sa_flags & SA_SIGINFO)
4489 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4490 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4491
4492
4493 switch(sig) {
4494 case SIGSEGV:
4495 case SIGBUS:
4496 case SIGFPE:
4497 case SIGPIPE:
4498 case SIGILL:
4499 case SIGXFSZ:
4500 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4501 break;
4502
4503 case SHUTDOWN1_SIGNAL:
4504 case SHUTDOWN2_SIGNAL:
4505 case SHUTDOWN3_SIGNAL:
4506 case BREAK_SIGNAL:
4507 jvmHandler = (address)user_handler();
4508 break;
4509
4510 case INTERRUPT_SIGNAL:
4511 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4512 break;
4513
4514 default:
4515 if (sig == SR_signum) {
4516 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4517 } else {
4518 return;
4519 }
4520 break;
4521 }
4522
4523 if (thisHandler != jvmHandler) {
4524 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4525 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4526 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4527 // No need to check this sig any longer
4528 sigaddset(&check_signal_done, sig);
4529 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4530 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4531 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4532 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4533 // No need to check this sig any longer
4534 sigaddset(&check_signal_done, sig);
4535 }
4536
4537 // Dump all the signal
4538 if (sigismember(&check_signal_done, sig)) {
4539 print_signal_handlers(tty, buf, O_BUFLEN);
4540 }
4541 }
4542
4543 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4544
4545 extern bool signal_name(int signo, char* buf, size_t len);
4546
4547 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4548 if (0 < exception_code && exception_code <= SIGRTMAX) {
4549 // signal
4550 if (!signal_name(exception_code, buf, size)) {
4551 jio_snprintf(buf, size, "SIG%d", exception_code);
4552 }
4553 return buf;
4554 } else {
4555 return NULL;
4556 }
4557 }
4558
4559 // this is called _before_ the most of global arguments have been parsed
4560 void os::init(void) {
4561 char dummy; /* used to get a guess on initial stack address */
4562 // first_hrtime = gethrtime();
4563
4564 // With LinuxThreads the JavaMain thread pid (primordial thread)
4565 // is different than the pid of the java launcher thread.
4566 // So, on Linux, the launcher thread pid is passed to the VM
4567 // via the sun.java.launcher.pid property.
4568 // Use this property instead of getpid() if it was correctly passed.
4569 // See bug 6351349.
4570 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4571
4572 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4573
4574 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4575
4576 init_random(1234567);
4577
4578 ThreadCritical::initialize();
4579
4580 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4581 if (Linux::page_size() == -1) {
4582 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4583 strerror(errno)));
4584 }
4585 init_page_sizes((size_t) Linux::page_size());
4586
4587 Linux::initialize_system_info();
4588
4589 // main_thread points to the aboriginal thread
4590 Linux::_main_thread = pthread_self();
4591
4592 Linux::clock_init();
4593 initial_time_count = javaTimeNanos();
4594
4595 // pthread_condattr initialization for monotonic clock
4596 int status;
4597 pthread_condattr_t* _condattr = os::Linux::condAttr();
4598 if ((status = pthread_condattr_init(_condattr)) != 0) {
4599 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4600 }
4601 // Only set the clock if CLOCK_MONOTONIC is available
4602 if (os::supports_monotonic_clock()) {
4603 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4604 if (status == EINVAL) {
4605 warning("Unable to use monotonic clock with relative timed-waits" \
4606 " - changes to the time-of-day clock may have adverse affects");
4607 } else {
4608 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4609 }
4610 }
4611 }
4612 // else it defaults to CLOCK_REALTIME
4613
4614 pthread_mutex_init(&dl_mutex, NULL);
4615
4616 // If the pagesize of the VM is greater than 8K determine the appropriate
4617 // number of initial guard pages. The user can change this with the
4618 // command line arguments, if needed.
4619 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4620 StackYellowPages = 1;
4621 StackRedPages = 1;
4622 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4623 }
4624 }
4625
4626 // To install functions for atexit system call
4627 extern "C" {
4628 static void perfMemory_exit_helper() {
4629 perfMemory_exit();
4630 }
4631 }
4632
4633 // this is called _after_ the global arguments have been parsed
4634 jint os::init_2(void)
4635 {
4636 Linux::fast_thread_clock_init();
4637
4638 // Allocate a single page and mark it as readable for safepoint polling
4639 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4640 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4641
4642 os::set_polling_page( polling_page );
4643
4644 #ifndef PRODUCT
4645 if(Verbose && PrintMiscellaneous)
4646 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4647 #endif
4648
4649 if (!UseMembar) {
4650 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4651 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4652 os::set_memory_serialize_page( mem_serialize_page );
4653
4654 #ifndef PRODUCT
4655 if(Verbose && PrintMiscellaneous)
4656 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4657 #endif
4658 }
4659
4660 // initialize suspend/resume support - must do this before signal_sets_init()
4661 if (SR_initialize() != 0) {
4662 perror("SR_initialize failed");
4663 return JNI_ERR;
4664 }
4665
4666 Linux::signal_sets_init();
4667 Linux::install_signal_handlers();
4668
4669 // Check minimum allowable stack size for thread creation and to initialize
4670 // the java system classes, including StackOverflowError - depends on page
4671 // size. Add a page for compiler2 recursion in main thread.
4672 // Add in 2*BytesPerWord times page size to account for VM stack during
4673 // class initialization depending on 32 or 64 bit VM.
4674 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4675 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4676 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4677
4678 size_t threadStackSizeInBytes = ThreadStackSize * K;
4679 if (threadStackSizeInBytes != 0 &&
4680 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4681 tty->print_cr("\nThe stack size specified is too small, "
4682 "Specify at least %dk",
4683 os::Linux::min_stack_allowed/ K);
4684 return JNI_ERR;
4685 }
4686
4687 // Make the stack size a multiple of the page size so that
4688 // the yellow/red zones can be guarded.
4689 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4690 vm_page_size()));
4691
4692 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4693
4694 #if defined(IA32)
4695 workaround_expand_exec_shield_cs_limit();
4696 #endif
4697
4698 Linux::libpthread_init();
4699 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4700 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4701 Linux::glibc_version(), Linux::libpthread_version(),
4702 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4703 }
4704
4705 if (UseNUMA) {
4706 if (!Linux::libnuma_init()) {
4707 UseNUMA = false;
4708 } else {
4709 if ((Linux::numa_max_node() < 1)) {
4710 // There's only one node(they start from 0), disable NUMA.
4711 UseNUMA = false;
4712 }
4713 }
4714 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4715 // we can make the adaptive lgrp chunk resizing work. If the user specified
4716 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4717 // disable adaptive resizing.
4718 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4719 if (FLAG_IS_DEFAULT(UseNUMA)) {
4720 UseNUMA = false;
4721 } else {
4722 if (FLAG_IS_DEFAULT(UseLargePages) &&
4723 FLAG_IS_DEFAULT(UseSHM) &&
4724 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4725 UseLargePages = false;
4726 } else {
4727 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4728 UseAdaptiveSizePolicy = false;
4729 UseAdaptiveNUMAChunkSizing = false;
4730 }
4731 }
4732 }
4733 if (!UseNUMA && ForceNUMA) {
4734 UseNUMA = true;
4735 }
4736 }
4737
4738 if (MaxFDLimit) {
4739 // set the number of file descriptors to max. print out error
4740 // if getrlimit/setrlimit fails but continue regardless.
4741 struct rlimit nbr_files;
4742 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4743 if (status != 0) {
4744 if (PrintMiscellaneous && (Verbose || WizardMode))
4745 perror("os::init_2 getrlimit failed");
4746 } else {
4747 nbr_files.rlim_cur = nbr_files.rlim_max;
4748 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4749 if (status != 0) {
4750 if (PrintMiscellaneous && (Verbose || WizardMode))
4751 perror("os::init_2 setrlimit failed");
4752 }
4753 }
4754 }
4755
4756 // Initialize lock used to serialize thread creation (see os::create_thread)
4757 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4758
4759 // at-exit methods are called in the reverse order of their registration.
4760 // atexit functions are called on return from main or as a result of a
4761 // call to exit(3C). There can be only 32 of these functions registered
4762 // and atexit() does not set errno.
4763
4764 if (PerfAllowAtExitRegistration) {
4765 // only register atexit functions if PerfAllowAtExitRegistration is set.
4766 // atexit functions can be delayed until process exit time, which
4767 // can be problematic for embedded VM situations. Embedded VMs should
4768 // call DestroyJavaVM() to assure that VM resources are released.
4769
4770 // note: perfMemory_exit_helper atexit function may be removed in
4771 // the future if the appropriate cleanup code can be added to the
4772 // VM_Exit VMOperation's doit method.
4773 if (atexit(perfMemory_exit_helper) != 0) {
4774 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4775 }
4776 }
4777
4778 // initialize thread priority policy
4779 prio_init();
4780
4781 return JNI_OK;
4782 }
4783
4784 // this is called at the end of vm_initialization
4785 void os::init_3(void) {
4786 #ifdef JAVASE_EMBEDDED
4787 // Start the MemNotifyThread
4788 if (LowMemoryProtection) {
4789 MemNotifyThread::start();
4790 }
4791 return;
4792 #endif
4793 }
4794
4795 // Mark the polling page as unreadable
4796 void os::make_polling_page_unreadable(void) {
4797 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4798 fatal("Could not disable polling page");
4799 };
4800
4801 // Mark the polling page as readable
4802 void os::make_polling_page_readable(void) {
4803 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4804 fatal("Could not enable polling page");
4805 }
4806 };
4807
4808 int os::active_processor_count() {
4809 // Linux doesn't yet have a (official) notion of processor sets,
4810 // so just return the number of online processors.
4811 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4812 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4813 return online_cpus;
4814 }
4815
4816 void os::set_native_thread_name(const char *name) {
4817 // Not yet implemented.
4818 return;
4819 }
4820
4821 bool os::distribute_processes(uint length, uint* distribution) {
4822 // Not yet implemented.
4823 return false;
4824 }
4825
4826 bool os::bind_to_processor(uint processor_id) {
4827 // Not yet implemented.
4828 return false;
4829 }
4830
4831 ///
4832
4833 void os::SuspendedThreadTask::internal_do_task() {
4834 if (do_suspend(_thread->osthread())) {
4835 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4836 do_task(context);
4837 do_resume(_thread->osthread());
4838 }
4839 }
4840
4841 class PcFetcher : public os::SuspendedThreadTask {
4842 public:
4843 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4844 ExtendedPC result();
4845 protected:
4846 void do_task(const os::SuspendedThreadTaskContext& context);
4847 private:
4848 ExtendedPC _epc;
4849 };
4850
4851 ExtendedPC PcFetcher::result() {
4852 guarantee(is_done(), "task is not done yet.");
4853 return _epc;
4854 }
4855
4856 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4857 Thread* thread = context.thread();
4858 OSThread* osthread = thread->osthread();
4859 if (osthread->ucontext() != NULL) {
4860 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4861 } else {
4862 // NULL context is unexpected, double-check this is the VMThread
4863 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4864 }
4865 }
4866
4867 // Suspends the target using the signal mechanism and then grabs the PC before
4868 // resuming the target. Used by the flat-profiler only
4869 ExtendedPC os::get_thread_pc(Thread* thread) {
4870 // Make sure that it is called by the watcher for the VMThread
4871 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4872 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4873
4874 PcFetcher fetcher(thread);
4875 fetcher.run();
4876 return fetcher.result();
4877 }
4878
4879 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4880 {
4881 if (is_NPTL()) {
4882 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4883 } else {
4884 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4885 // word back to default 64bit precision if condvar is signaled. Java
4886 // wants 53bit precision. Save and restore current value.
4887 int fpu = get_fpu_control_word();
4888 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4889 set_fpu_control_word(fpu);
4890 return status;
4891 }
4892 }
4893
4894 ////////////////////////////////////////////////////////////////////////////////
4895 // debug support
4896
4897 bool os::find(address addr, outputStream* st) {
4898 Dl_info dlinfo;
4899 memset(&dlinfo, 0, sizeof(dlinfo));
4900 if (dladdr(addr, &dlinfo) != 0) {
4901 st->print(PTR_FORMAT ": ", addr);
4902 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4903 st->print("%s+%#x", dlinfo.dli_sname,
4904 addr - (intptr_t)dlinfo.dli_saddr);
4905 } else if (dlinfo.dli_fbase != NULL) {
4906 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4907 } else {
4908 st->print("<absolute address>");
4909 }
4910 if (dlinfo.dli_fname != NULL) {
4911 st->print(" in %s", dlinfo.dli_fname);
4912 }
4913 if (dlinfo.dli_fbase != NULL) {
4914 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4915 }
4916 st->cr();
4917
4918 if (Verbose) {
4919 // decode some bytes around the PC
4920 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4921 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4922 address lowest = (address) dlinfo.dli_sname;
4923 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4924 if (begin < lowest) begin = lowest;
4925 Dl_info dlinfo2;
4926 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4927 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4928 end = (address) dlinfo2.dli_saddr;
4929 Disassembler::decode(begin, end, st);
4930 }
4931 return true;
4932 }
4933 return false;
4934 }
4935
4936 ////////////////////////////////////////////////////////////////////////////////
4937 // misc
4938
4939 // This does not do anything on Linux. This is basically a hook for being
4940 // able to use structured exception handling (thread-local exception filters)
4941 // on, e.g., Win32.
4942 void
4943 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4944 JavaCallArguments* args, Thread* thread) {
4945 f(value, method, args, thread);
4946 }
4947
4948 void os::print_statistics() {
4949 }
4950
4951 int os::message_box(const char* title, const char* message) {
4952 int i;
4953 fdStream err(defaultStream::error_fd());
4954 for (i = 0; i < 78; i++) err.print_raw("=");
4955 err.cr();
4956 err.print_raw_cr(title);
4957 for (i = 0; i < 78; i++) err.print_raw("-");
4958 err.cr();
4959 err.print_raw_cr(message);
4960 for (i = 0; i < 78; i++) err.print_raw("=");
4961 err.cr();
4962
4963 char buf[16];
4964 // Prevent process from exiting upon "read error" without consuming all CPU
4965 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4966
4967 return buf[0] == 'y' || buf[0] == 'Y';
4968 }
4969
4970 int os::stat(const char *path, struct stat *sbuf) {
4971 char pathbuf[MAX_PATH];
4972 if (strlen(path) > MAX_PATH - 1) {
4973 errno = ENAMETOOLONG;
4974 return -1;
4975 }
4976 os::native_path(strcpy(pathbuf, path));
4977 return ::stat(pathbuf, sbuf);
4978 }
4979
4980 bool os::check_heap(bool force) {
4981 return true;
4982 }
4983
4984 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4985 return ::vsnprintf(buf, count, format, args);
4986 }
4987
4988 // Is a (classpath) directory empty?
4989 bool os::dir_is_empty(const char* path) {
4990 DIR *dir = NULL;
4991 struct dirent *ptr;
4992
4993 dir = opendir(path);
4994 if (dir == NULL) return true;
4995
4996 /* Scan the directory */
4997 bool result = true;
4998 char buf[sizeof(struct dirent) + MAX_PATH];
4999 while (result && (ptr = ::readdir(dir)) != NULL) {
5000 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5001 result = false;
5002 }
5003 }
5004 closedir(dir);
5005 return result;
5006 }
5007
5008 // This code originates from JDK's sysOpen and open64_w
5009 // from src/solaris/hpi/src/system_md.c
5010
5011 #ifndef O_DELETE
5012 #define O_DELETE 0x10000
5013 #endif
5014
5015 // Open a file. Unlink the file immediately after open returns
5016 // if the specified oflag has the O_DELETE flag set.
5017 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5018
5019 int os::open(const char *path, int oflag, int mode) {
5020
5021 if (strlen(path) > MAX_PATH - 1) {
5022 errno = ENAMETOOLONG;
5023 return -1;
5024 }
5025 int fd;
5026 int o_delete = (oflag & O_DELETE);
5027 oflag = oflag & ~O_DELETE;
5028
5029 fd = ::open64(path, oflag, mode);
5030 if (fd == -1) return -1;
5031
5032 //If the open succeeded, the file might still be a directory
5033 {
5034 struct stat64 buf64;
5035 int ret = ::fstat64(fd, &buf64);
5036 int st_mode = buf64.st_mode;
5037
5038 if (ret != -1) {
5039 if ((st_mode & S_IFMT) == S_IFDIR) {
5040 errno = EISDIR;
5041 ::close(fd);
5042 return -1;
5043 }
5044 } else {
5045 ::close(fd);
5046 return -1;
5047 }
5048 }
5049
5050 /*
5051 * All file descriptors that are opened in the JVM and not
5052 * specifically destined for a subprocess should have the
5053 * close-on-exec flag set. If we don't set it, then careless 3rd
5054 * party native code might fork and exec without closing all
5055 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5056 * UNIXProcess.c), and this in turn might:
5057 *
5058 * - cause end-of-file to fail to be detected on some file
5059 * descriptors, resulting in mysterious hangs, or
5060 *
5061 * - might cause an fopen in the subprocess to fail on a system
5062 * suffering from bug 1085341.
5063 *
5064 * (Yes, the default setting of the close-on-exec flag is a Unix
5065 * design flaw)
5066 *
5067 * See:
5068 * 1085341: 32-bit stdio routines should support file descriptors >255
5069 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5070 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5071 */
5072 #ifdef FD_CLOEXEC
5073 {
5074 int flags = ::fcntl(fd, F_GETFD);
5075 if (flags != -1)
5076 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5077 }
5078 #endif
5079
5080 if (o_delete != 0) {
5081 ::unlink(path);
5082 }
5083 return fd;
5084 }
5085
5086
5087 // create binary file, rewriting existing file if required
5088 int os::create_binary_file(const char* path, bool rewrite_existing) {
5089 int oflags = O_WRONLY | O_CREAT;
5090 if (!rewrite_existing) {
5091 oflags |= O_EXCL;
5092 }
5093 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5094 }
5095
5096 // return current position of file pointer
5097 jlong os::current_file_offset(int fd) {
5098 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5099 }
5100
5101 // move file pointer to the specified offset
5102 jlong os::seek_to_file_offset(int fd, jlong offset) {
5103 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5104 }
5105
5106 // This code originates from JDK's sysAvailable
5107 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5108
5109 int os::available(int fd, jlong *bytes) {
5110 jlong cur, end;
5111 int mode;
5112 struct stat64 buf64;
5113
5114 if (::fstat64(fd, &buf64) >= 0) {
5115 mode = buf64.st_mode;
5116 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5117 /*
5118 * XXX: is the following call interruptible? If so, this might
5119 * need to go through the INTERRUPT_IO() wrapper as for other
5120 * blocking, interruptible calls in this file.
5121 */
5122 int n;
5123 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5124 *bytes = n;
5125 return 1;
5126 }
5127 }
5128 }
5129 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5130 return 0;
5131 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5132 return 0;
5133 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5134 return 0;
5135 }
5136 *bytes = end - cur;
5137 return 1;
5138 }
5139
5140 int os::socket_available(int fd, jint *pbytes) {
5141 // Linux doc says EINTR not returned, unlike Solaris
5142 int ret = ::ioctl(fd, FIONREAD, pbytes);
5143
5144 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5145 // is expected to return 0 on failure and 1 on success to the jdk.
5146 return (ret < 0) ? 0 : 1;
5147 }
5148
5149 // Map a block of memory.
5150 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5151 char *addr, size_t bytes, bool read_only,
5152 bool allow_exec) {
5153 int prot;
5154 int flags = MAP_PRIVATE;
5155
5156 if (read_only) {
5157 prot = PROT_READ;
5158 } else {
5159 prot = PROT_READ | PROT_WRITE;
5160 }
5161
5162 if (allow_exec) {
5163 prot |= PROT_EXEC;
5164 }
5165
5166 if (addr != NULL) {
5167 flags |= MAP_FIXED;
5168 }
5169
5170 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5171 fd, file_offset);
5172 if (mapped_address == MAP_FAILED) {
5173 return NULL;
5174 }
5175 return mapped_address;
5176 }
5177
5178
5179 // Remap a block of memory.
5180 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5181 char *addr, size_t bytes, bool read_only,
5182 bool allow_exec) {
5183 // same as map_memory() on this OS
5184 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5185 allow_exec);
5186 }
5187
5188
5189 // Unmap a block of memory.
5190 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5191 return munmap(addr, bytes) == 0;
5192 }
5193
5194 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5195
5196 static clockid_t thread_cpu_clockid(Thread* thread) {
5197 pthread_t tid = thread->osthread()->pthread_id();
5198 clockid_t clockid;
5199
5200 // Get thread clockid
5201 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5202 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5203 return clockid;
5204 }
5205
5206 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5207 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5208 // of a thread.
5209 //
5210 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5211 // the fast estimate available on the platform.
5212
5213 jlong os::current_thread_cpu_time() {
5214 if (os::Linux::supports_fast_thread_cpu_time()) {
5215 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5216 } else {
5217 // return user + sys since the cost is the same
5218 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5219 }
5220 }
5221
5222 jlong os::thread_cpu_time(Thread* thread) {
5223 // consistent with what current_thread_cpu_time() returns
5224 if (os::Linux::supports_fast_thread_cpu_time()) {
5225 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5226 } else {
5227 return slow_thread_cpu_time(thread, true /* user + sys */);
5228 }
5229 }
5230
5231 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5232 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5233 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5234 } else {
5235 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5236 }
5237 }
5238
5239 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5240 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5241 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5242 } else {
5243 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5244 }
5245 }
5246
5247 //
5248 // -1 on error.
5249 //
5250
5251 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5252 static bool proc_task_unchecked = true;
5253 pid_t tid = thread->osthread()->thread_id();
5254 char *s;
5255 char stat[2048];
5256 int statlen;
5257 char proc_name[64];
5258 int count;
5259 long sys_time, user_time;
5260 char cdummy;
5261 int idummy;
5262 long ldummy;
5263 FILE *fp;
5264
5265 snprintf(proc_name, 64, "/proc/%d/stat", tid);
5266
5267 // The /proc/<tid>/stat aggregates per-process usage on
5268 // new Linux kernels 2.6+ where NPTL is supported.
5269 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5270 // See bug 6328462.
5271 // There possibly can be cases where there is no directory
5272 // /proc/self/task, so we check its availability.
5273 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5274 // This is executed only once
5275 proc_task_unchecked = false;
5276 fp = fopen("/proc/self/task", "r");
5277 if (fp != NULL) {
5278 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5279 fclose(fp);
5280 }
5281 }
5282
5283 fp = fopen(proc_name, "r");
5284 if ( fp == NULL ) return -1;
5285 statlen = fread(stat, 1, 2047, fp);
5286 stat[statlen] = '\0';
5287 fclose(fp);
5288
5289 // Skip pid and the command string. Note that we could be dealing with
5290 // weird command names, e.g. user could decide to rename java launcher
5291 // to "java 1.4.2 :)", then the stat file would look like
5292 // 1234 (java 1.4.2 :)) R ... ...
5293 // We don't really need to know the command string, just find the last
5294 // occurrence of ")" and then start parsing from there. See bug 4726580.
5295 s = strrchr(stat, ')');
5296 if (s == NULL ) return -1;
5297
5298 // Skip blank chars
5299 do s++; while (isspace(*s));
5300
5301 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5302 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5303 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5304 &user_time, &sys_time);
5305 if ( count != 13 ) return -1;
5306 if (user_sys_cpu_time) {
5307 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5308 } else {
5309 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5310 }
5311 }
5312
5313 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5314 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5315 info_ptr->may_skip_backward = false; // elapsed time not wall time
5316 info_ptr->may_skip_forward = false; // elapsed time not wall time
5317 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5318 }
5319
5320 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5321 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5322 info_ptr->may_skip_backward = false; // elapsed time not wall time
5323 info_ptr->may_skip_forward = false; // elapsed time not wall time
5324 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5325 }
5326
5327 bool os::is_thread_cpu_time_supported() {
5328 return true;
5329 }
5330
5331 // System loadavg support. Returns -1 if load average cannot be obtained.
5332 // Linux doesn't yet have a (official) notion of processor sets,
5333 // so just return the system wide load average.
5334 int os::loadavg(double loadavg[], int nelem) {
5335 return ::getloadavg(loadavg, nelem);
5336 }
5337
5338 void os::pause() {
5339 char filename[MAX_PATH];
5340 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5341 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5342 } else {
5343 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5344 }
5345
5346 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5347 if (fd != -1) {
5348 struct stat buf;
5349 ::close(fd);
5350 while (::stat(filename, &buf) == 0) {
5351 (void)::poll(NULL, 0, 100);
5352 }
5353 } else {
5354 jio_fprintf(stderr,
5355 "Could not open pause file '%s', continuing immediately.\n", filename);
5356 }
5357 }
5358
5359
5360 // Refer to the comments in os_solaris.cpp park-unpark.
5361 //
5362 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5363 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5364 // For specifics regarding the bug see GLIBC BUGID 261237 :
5365 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5366 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5367 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5368 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5369 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5370 // and monitorenter when we're using 1-0 locking. All those operations may result in
5371 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5372 // of libpthread avoids the problem, but isn't practical.
5373 //
5374 // Possible remedies:
5375 //
5376 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5377 // This is palliative and probabilistic, however. If the thread is preempted
5378 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5379 // than the minimum period may have passed, and the abstime may be stale (in the
5380 // past) resultin in a hang. Using this technique reduces the odds of a hang
5381 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5382 //
5383 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5384 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5385 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5386 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5387 // thread.
5388 //
5389 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5390 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5391 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5392 // This also works well. In fact it avoids kernel-level scalability impediments
5393 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5394 // timers in a graceful fashion.
5395 //
5396 // 4. When the abstime value is in the past it appears that control returns
5397 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5398 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5399 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5400 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5401 // It may be possible to avoid reinitialization by checking the return
5402 // value from pthread_cond_timedwait(). In addition to reinitializing the
5403 // condvar we must establish the invariant that cond_signal() is only called
5404 // within critical sections protected by the adjunct mutex. This prevents
5405 // cond_signal() from "seeing" a condvar that's in the midst of being
5406 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5407 // desirable signal-after-unlock optimization that avoids futile context switching.
5408 //
5409 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5410 // structure when a condvar is used or initialized. cond_destroy() would
5411 // release the helper structure. Our reinitialize-after-timedwait fix
5412 // put excessive stress on malloc/free and locks protecting the c-heap.
5413 //
5414 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5415 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5416 // and only enabling the work-around for vulnerable environments.
5417
5418 // utility to compute the abstime argument to timedwait:
5419 // millis is the relative timeout time
5420 // abstime will be the absolute timeout time
5421 // TODO: replace compute_abstime() with unpackTime()
5422
5423 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5424 if (millis < 0) millis = 0;
5425
5426 jlong seconds = millis / 1000;
5427 millis %= 1000;
5428 if (seconds > 50000000) { // see man cond_timedwait(3T)
5429 seconds = 50000000;
5430 }
5431
5432 if (os::supports_monotonic_clock()) {
5433 struct timespec now;
5434 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5435 assert_status(status == 0, status, "clock_gettime");
5436 abstime->tv_sec = now.tv_sec + seconds;
5437 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5438 if (nanos >= NANOSECS_PER_SEC) {
5439 abstime->tv_sec += 1;
5440 nanos -= NANOSECS_PER_SEC;
5441 }
5442 abstime->tv_nsec = nanos;
5443 } else {
5444 struct timeval now;
5445 int status = gettimeofday(&now, NULL);
5446 assert(status == 0, "gettimeofday");
5447 abstime->tv_sec = now.tv_sec + seconds;
5448 long usec = now.tv_usec + millis * 1000;
5449 if (usec >= 1000000) {
5450 abstime->tv_sec += 1;
5451 usec -= 1000000;
5452 }
5453 abstime->tv_nsec = usec * 1000;
5454 }
5455 return abstime;
5456 }
5457
5458
5459 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5460 // Conceptually TryPark() should be equivalent to park(0).
5461
5462 int os::PlatformEvent::TryPark() {
5463 for (;;) {
5464 const int v = _Event ;
5465 guarantee ((v == 0) || (v == 1), "invariant") ;
5466 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5467 }
5468 }
5469
5470 void os::PlatformEvent::park() { // AKA "down()"
5471 // Invariant: Only the thread associated with the Event/PlatformEvent
5472 // may call park().
5473 // TODO: assert that _Assoc != NULL or _Assoc == Self
5474 int v ;
5475 for (;;) {
5476 v = _Event ;
5477 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5478 }
5479 guarantee (v >= 0, "invariant") ;
5480 if (v == 0) {
5481 // Do this the hard way by blocking ...
5482 int status = pthread_mutex_lock(_mutex);
5483 assert_status(status == 0, status, "mutex_lock");
5484 guarantee (_nParked == 0, "invariant") ;
5485 ++ _nParked ;
5486 while (_Event < 0) {
5487 status = pthread_cond_wait(_cond, _mutex);
5488 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5489 // Treat this the same as if the wait was interrupted
5490 if (status == ETIME) { status = EINTR; }
5491 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5492 }
5493 -- _nParked ;
5494
5495 _Event = 0 ;
5496 status = pthread_mutex_unlock(_mutex);
5497 assert_status(status == 0, status, "mutex_unlock");
5498 // Paranoia to ensure our locked and lock-free paths interact
5499 // correctly with each other.
5500 OrderAccess::fence();
5501 }
5502 guarantee (_Event >= 0, "invariant") ;
5503 }
5504
5505 int os::PlatformEvent::park(jlong millis) {
5506 guarantee (_nParked == 0, "invariant") ;
5507
5508 int v ;
5509 for (;;) {
5510 v = _Event ;
5511 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5512 }
5513 guarantee (v >= 0, "invariant") ;
5514 if (v != 0) return OS_OK ;
5515
5516 // We do this the hard way, by blocking the thread.
5517 // Consider enforcing a minimum timeout value.
5518 struct timespec abst;
5519 compute_abstime(&abst, millis);
5520
5521 int ret = OS_TIMEOUT;
5522 int status = pthread_mutex_lock(_mutex);
5523 assert_status(status == 0, status, "mutex_lock");
5524 guarantee (_nParked == 0, "invariant") ;
5525 ++_nParked ;
5526
5527 // Object.wait(timo) will return because of
5528 // (a) notification
5529 // (b) timeout
5530 // (c) thread.interrupt
5531 //
5532 // Thread.interrupt and object.notify{All} both call Event::set.
5533 // That is, we treat thread.interrupt as a special case of notification.
5534 // The underlying Solaris implementation, cond_timedwait, admits
5535 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5536 // JVM from making those visible to Java code. As such, we must
5537 // filter out spurious wakeups. We assume all ETIME returns are valid.
5538 //
5539 // TODO: properly differentiate simultaneous notify+interrupt.
5540 // In that case, we should propagate the notify to another waiter.
5541
5542 while (_Event < 0) {
5543 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5544 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5545 pthread_cond_destroy (_cond);
5546 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5547 }
5548 assert_status(status == 0 || status == EINTR ||
5549 status == ETIME || status == ETIMEDOUT,
5550 status, "cond_timedwait");
5551 if (!FilterSpuriousWakeups) break ; // previous semantics
5552 if (status == ETIME || status == ETIMEDOUT) break ;
5553 // We consume and ignore EINTR and spurious wakeups.
5554 }
5555 --_nParked ;
5556 if (_Event >= 0) {
5557 ret = OS_OK;
5558 }
5559 _Event = 0 ;
5560 status = pthread_mutex_unlock(_mutex);
5561 assert_status(status == 0, status, "mutex_unlock");
5562 assert (_nParked == 0, "invariant") ;
5563 // Paranoia to ensure our locked and lock-free paths interact
5564 // correctly with each other.
5565 OrderAccess::fence();
5566 return ret;
5567 }
5568
5569 void os::PlatformEvent::unpark() {
5570 // Transitions for _Event:
5571 // 0 :=> 1
5572 // 1 :=> 1
5573 // -1 :=> either 0 or 1; must signal target thread
5574 // That is, we can safely transition _Event from -1 to either
5575 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5576 // unpark() calls.
5577 // See also: "Semaphores in Plan 9" by Mullender & Cox
5578 //
5579 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5580 // that it will take two back-to-back park() calls for the owning
5581 // thread to block. This has the benefit of forcing a spurious return
5582 // from the first park() call after an unpark() call which will help
5583 // shake out uses of park() and unpark() without condition variables.
5584
5585 if (Atomic::xchg(1, &_Event) >= 0) return;
5586
5587 // Wait for the thread associated with the event to vacate
5588 int status = pthread_mutex_lock(_mutex);
5589 assert_status(status == 0, status, "mutex_lock");
5590 int AnyWaiters = _nParked;
5591 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5592 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5593 AnyWaiters = 0;
5594 pthread_cond_signal(_cond);
5595 }
5596 status = pthread_mutex_unlock(_mutex);
5597 assert_status(status == 0, status, "mutex_unlock");
5598 if (AnyWaiters != 0) {
5599 status = pthread_cond_signal(_cond);
5600 assert_status(status == 0, status, "cond_signal");
5601 }
5602
5603 // Note that we signal() _after dropping the lock for "immortal" Events.
5604 // This is safe and avoids a common class of futile wakeups. In rare
5605 // circumstances this can cause a thread to return prematurely from
5606 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5607 // simply re-test the condition and re-park itself.
5608 }
5609
5610
5611 // JSR166
5612 // -------------------------------------------------------
5613
5614 /*
5615 * The solaris and linux implementations of park/unpark are fairly
5616 * conservative for now, but can be improved. They currently use a
5617 * mutex/condvar pair, plus a a count.
5618 * Park decrements count if > 0, else does a condvar wait. Unpark
5619 * sets count to 1 and signals condvar. Only one thread ever waits
5620 * on the condvar. Contention seen when trying to park implies that someone
5621 * is unparking you, so don't wait. And spurious returns are fine, so there
5622 * is no need to track notifications.
5623 */
5624
5625 #define MAX_SECS 100000000
5626 /*
5627 * This code is common to linux and solaris and will be moved to a
5628 * common place in dolphin.
5629 *
5630 * The passed in time value is either a relative time in nanoseconds
5631 * or an absolute time in milliseconds. Either way it has to be unpacked
5632 * into suitable seconds and nanoseconds components and stored in the
5633 * given timespec structure.
5634 * Given time is a 64-bit value and the time_t used in the timespec is only
5635 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5636 * overflow if times way in the future are given. Further on Solaris versions
5637 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5638 * number of seconds, in abstime, is less than current_time + 100,000,000.
5639 * As it will be 28 years before "now + 100000000" will overflow we can
5640 * ignore overflow and just impose a hard-limit on seconds using the value
5641 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5642 * years from "now".
5643 */
5644
5645 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5646 assert (time > 0, "convertTime");
5647 time_t max_secs = 0;
5648
5649 if (!os::supports_monotonic_clock() || isAbsolute) {
5650 struct timeval now;
5651 int status = gettimeofday(&now, NULL);
5652 assert(status == 0, "gettimeofday");
5653
5654 max_secs = now.tv_sec + MAX_SECS;
5655
5656 if (isAbsolute) {
5657 jlong secs = time / 1000;
5658 if (secs > max_secs) {
5659 absTime->tv_sec = max_secs;
5660 } else {
5661 absTime->tv_sec = secs;
5662 }
5663 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5664 } else {
5665 jlong secs = time / NANOSECS_PER_SEC;
5666 if (secs >= MAX_SECS) {
5667 absTime->tv_sec = max_secs;
5668 absTime->tv_nsec = 0;
5669 } else {
5670 absTime->tv_sec = now.tv_sec + secs;
5671 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5672 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5673 absTime->tv_nsec -= NANOSECS_PER_SEC;
5674 ++absTime->tv_sec; // note: this must be <= max_secs
5675 }
5676 }
5677 }
5678 } else {
5679 // must be relative using monotonic clock
5680 struct timespec now;
5681 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5682 assert_status(status == 0, status, "clock_gettime");
5683 max_secs = now.tv_sec + MAX_SECS;
5684 jlong secs = time / NANOSECS_PER_SEC;
5685 if (secs >= MAX_SECS) {
5686 absTime->tv_sec = max_secs;
5687 absTime->tv_nsec = 0;
5688 } else {
5689 absTime->tv_sec = now.tv_sec + secs;
5690 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5691 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5692 absTime->tv_nsec -= NANOSECS_PER_SEC;
5693 ++absTime->tv_sec; // note: this must be <= max_secs
5694 }
5695 }
5696 }
5697 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5698 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5699 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5700 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5701 }
5702
5703 void Parker::park(bool isAbsolute, jlong time) {
5704 // Ideally we'd do something useful while spinning, such
5705 // as calling unpackTime().
5706
5707 // Optional fast-path check:
5708 // Return immediately if a permit is available.
5709 // We depend on Atomic::xchg() having full barrier semantics
5710 // since we are doing a lock-free update to _counter.
5711 if (Atomic::xchg(0, &_counter) > 0) return;
5712
5713 Thread* thread = Thread::current();
5714 assert(thread->is_Java_thread(), "Must be JavaThread");
5715 JavaThread *jt = (JavaThread *)thread;
5716
5717 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5718 // Check interrupt before trying to wait
5719 if (Thread::is_interrupted(thread, false)) {
5720 return;
5721 }
5722
5723 // Next, demultiplex/decode time arguments
5724 timespec absTime;
5725 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5726 return;
5727 }
5728 if (time > 0) {
5729 unpackTime(&absTime, isAbsolute, time);
5730 }
5731
5732
5733 // Enter safepoint region
5734 // Beware of deadlocks such as 6317397.
5735 // The per-thread Parker:: mutex is a classic leaf-lock.
5736 // In particular a thread must never block on the Threads_lock while
5737 // holding the Parker:: mutex. If safepoints are pending both the
5738 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5739 ThreadBlockInVM tbivm(jt);
5740
5741 // Don't wait if cannot get lock since interference arises from
5742 // unblocking. Also. check interrupt before trying wait
5743 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5744 return;
5745 }
5746
5747 int status ;
5748 if (_counter > 0) { // no wait needed
5749 _counter = 0;
5750 status = pthread_mutex_unlock(_mutex);
5751 assert (status == 0, "invariant") ;
5752 // Paranoia to ensure our locked and lock-free paths interact
5753 // correctly with each other and Java-level accesses.
5754 OrderAccess::fence();
5755 return;
5756 }
5757
5758 #ifdef ASSERT
5759 // Don't catch signals while blocked; let the running threads have the signals.
5760 // (This allows a debugger to break into the running thread.)
5761 sigset_t oldsigs;
5762 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5763 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5764 #endif
5765
5766 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5767 jt->set_suspend_equivalent();
5768 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5769
5770 assert(_cur_index == -1, "invariant");
5771 if (time == 0) {
5772 _cur_index = REL_INDEX; // arbitrary choice when not timed
5773 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5774 } else {
5775 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5776 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5777 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5778 pthread_cond_destroy (&_cond[_cur_index]) ;
5779 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5780 }
5781 }
5782 _cur_index = -1;
5783 assert_status(status == 0 || status == EINTR ||
5784 status == ETIME || status == ETIMEDOUT,
5785 status, "cond_timedwait");
5786
5787 #ifdef ASSERT
5788 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5789 #endif
5790
5791 _counter = 0 ;
5792 status = pthread_mutex_unlock(_mutex) ;
5793 assert_status(status == 0, status, "invariant") ;
5794 // Paranoia to ensure our locked and lock-free paths interact
5795 // correctly with each other and Java-level accesses.
5796 OrderAccess::fence();
5797
5798 // If externally suspended while waiting, re-suspend
5799 if (jt->handle_special_suspend_equivalent_condition()) {
5800 jt->java_suspend_self();
5801 }
5802 }
5803
5804 void Parker::unpark() {
5805 int s, status ;
5806 status = pthread_mutex_lock(_mutex);
5807 assert (status == 0, "invariant") ;
5808 s = _counter;
5809 _counter = 1;
5810 if (s < 1) {
5811 // thread might be parked
5812 if (_cur_index != -1) {
5813 // thread is definitely parked
5814 if (WorkAroundNPTLTimedWaitHang) {
5815 status = pthread_cond_signal (&_cond[_cur_index]);
5816 assert (status == 0, "invariant");
5817 status = pthread_mutex_unlock(_mutex);
5818 assert (status == 0, "invariant");
5819 } else {
5820 status = pthread_mutex_unlock(_mutex);
5821 assert (status == 0, "invariant");
5822 status = pthread_cond_signal (&_cond[_cur_index]);
5823 assert (status == 0, "invariant");
5824 }
5825 } else {
5826 pthread_mutex_unlock(_mutex);
5827 assert (status == 0, "invariant") ;
5828 }
5829 } else {
5830 pthread_mutex_unlock(_mutex);
5831 assert (status == 0, "invariant") ;
5832 }
5833 }
5834
5835
5836 extern char** environ;
5837
5838 #ifndef __NR_fork
5839 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5840 #endif
5841
5842 #ifndef __NR_execve
5843 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5844 #endif
5845
5846 // Run the specified command in a separate process. Return its exit value,
5847 // or -1 on failure (e.g. can't fork a new process).
5848 // Unlike system(), this function can be called from signal handler. It
5849 // doesn't block SIGINT et al.
5850 int os::fork_and_exec(char* cmd) {
5851 const char * argv[4] = {"sh", "-c", cmd, NULL};
5852
5853 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5854 // pthread_atfork handlers and reset pthread library. All we need is a
5855 // separate process to execve. Make a direct syscall to fork process.
5856 // On IA64 there's no fork syscall, we have to use fork() and hope for
5857 // the best...
5858 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5859 IA64_ONLY(fork();)
5860
5861 if (pid < 0) {
5862 // fork failed
5863 return -1;
5864
5865 } else if (pid == 0) {
5866 // child process
5867
5868 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5869 // first to kill every thread on the thread list. Because this list is
5870 // not reset by fork() (see notes above), execve() will instead kill
5871 // every thread in the parent process. We know this is the only thread
5872 // in the new process, so make a system call directly.
5873 // IA64 should use normal execve() from glibc to match the glibc fork()
5874 // above.
5875 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5876 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5877
5878 // execve failed
5879 _exit(-1);
5880
5881 } else {
5882 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5883 // care about the actual exit code, for now.
5884
5885 int status;
5886
5887 // Wait for the child process to exit. This returns immediately if
5888 // the child has already exited. */
5889 while (waitpid(pid, &status, 0) < 0) {
5890 switch (errno) {
5891 case ECHILD: return 0;
5892 case EINTR: break;
5893 default: return -1;
5894 }
5895 }
5896
5897 if (WIFEXITED(status)) {
5898 // The child exited normally; get its exit code.
5899 return WEXITSTATUS(status);
5900 } else if (WIFSIGNALED(status)) {
5901 // The child exited because of a signal
5902 // The best value to return is 0x80 + signal number,
5903 // because that is what all Unix shells do, and because
5904 // it allows callers to distinguish between process exit and
5905 // process death by signal.
5906 return 0x80 + WTERMSIG(status);
5907 } else {
5908 // Unknown exit code; pass it through
5909 return status;
5910 }
5911 }
5912 }
5913
5914 // is_headless_jre()
5915 //
5916 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5917 // in order to report if we are running in a headless jre
5918 //
5919 // Since JDK8 xawt/libmawt.so was moved into the same directory
5920 // as libawt.so, and renamed libawt_xawt.so
5921 //
5922 bool os::is_headless_jre() {
5923 struct stat statbuf;
5924 char buf[MAXPATHLEN];
5925 char libmawtpath[MAXPATHLEN];
5926 const char *xawtstr = "/xawt/libmawt.so";
5927 const char *new_xawtstr = "/libawt_xawt.so";
5928 char *p;
5929
5930 // Get path to libjvm.so
5931 os::jvm_path(buf, sizeof(buf));
5932
5933 // Get rid of libjvm.so
5934 p = strrchr(buf, '/');
5935 if (p == NULL) return false;
5936 else *p = '\0';
5937
5938 // Get rid of client or server
5939 p = strrchr(buf, '/');
5940 if (p == NULL) return false;
5941 else *p = '\0';
5942
5943 // check xawt/libmawt.so
5944 strcpy(libmawtpath, buf);
5945 strcat(libmawtpath, xawtstr);
5946 if (::stat(libmawtpath, &statbuf) == 0) return false;
5947
5948 // check libawt_xawt.so
5949 strcpy(libmawtpath, buf);
5950 strcat(libmawtpath, new_xawtstr);
5951 if (::stat(libmawtpath, &statbuf) == 0) return false;
5952
5953 return true;
5954 }
5955
5956 // Get the default path to the core file
5957 // Returns the length of the string
5958 int os::get_core_path(char* buffer, size_t bufferSize) {
5959 const char* p = get_current_directory(buffer, bufferSize);
5960
5961 if (p == NULL) {
5962 assert(p != NULL, "failed to get current directory");
5963 return 0;
5964 }
5965
5966 return strlen(buffer);
5967 }
5968
5969 #ifdef JAVASE_EMBEDDED
5970 //
5971 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5972 //
5973 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5974
5975 // ctor
5976 //
5977 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5978 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5979 _fd = fd;
5980
5981 if (os::create_thread(this, os::os_thread)) {
5982 _memnotify_thread = this;
5983 os::set_priority(this, NearMaxPriority);
5984 os::start_thread(this);
5985 }
5986 }
5987
5988 // Where all the work gets done
5989 //
5990 void MemNotifyThread::run() {
5991 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5992
5993 // Set up the select arguments
5994 fd_set rfds;
5995 if (_fd != -1) {
5996 FD_ZERO(&rfds);
5997 FD_SET(_fd, &rfds);
5998 }
5999
6000 // Now wait for the mem_notify device to wake up
6001 while (1) {
6002 // Wait for the mem_notify device to signal us..
6003 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6004 if (rc == -1) {
6005 perror("select!\n");
6006 break;
6007 } else if (rc) {
6008 //ssize_t free_before = os::available_memory();
6009 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6010
6011 // The kernel is telling us there is not much memory left...
6012 // try to do something about that
6013
6014 // If we are not already in a GC, try one.
6015 if (!Universe::heap()->is_gc_active()) {
6016 Universe::heap()->collect(GCCause::_allocation_failure);
6017
6018 //ssize_t free_after = os::available_memory();
6019 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6020 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6021 }
6022 // We might want to do something like the following if we find the GC's are not helping...
6023 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6024 }
6025 }
6026 }
6027
6028 //
6029 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6030 //
6031 void MemNotifyThread::start() {
6032 int fd;
6033 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6034 if (fd < 0) {
6035 return;
6036 }
6037
6038 if (memnotify_thread() == NULL) {
6039 new MemNotifyThread(fd);
6040 }
6041 }
6042
6043 #endif // JAVASE_EMBEDDED
6044
6045
6046 /////////////// Unit tests ///////////////
6047
6048 #ifndef PRODUCT
6049
6050 #define test_log(...) \
6051 do {\
6052 if (VerboseInternalVMTests) { \
6053 tty->print_cr(__VA_ARGS__); \
6054 tty->flush(); \
6055 }\
6056 } while (false)
6057
6058 class TestReserveMemorySpecial : AllStatic {
6059 public:
6060 static void small_page_write(void* addr, size_t size) {
6061 size_t page_size = os::vm_page_size();
6062
6063 char* end = (char*)addr + size;
6064 for (char* p = (char*)addr; p < end; p += page_size) {
6065 *p = 1;
6066 }
6067 }
6068
6069 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6070 if (!UseHugeTLBFS) {
6071 return;
6072 }
6073
6074 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6075
6076 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6077
6078 if (addr != NULL) {
6079 small_page_write(addr, size);
6080
6081 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6082 }
6083 }
6084
6085 static void test_reserve_memory_special_huge_tlbfs_only() {
6086 if (!UseHugeTLBFS) {
6087 return;
6088 }
6089
6090 size_t lp = os::large_page_size();
6091
6092 for (size_t size = lp; size <= lp * 10; size += lp) {
6093 test_reserve_memory_special_huge_tlbfs_only(size);
6094 }
6095 }
6096
6097 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6098 if (!UseHugeTLBFS) {
6099 return;
6100 }
6101
6102 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6103 size, alignment);
6104
6105 assert(size >= os::large_page_size(), "Incorrect input to test");
6106
6107 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6108
6109 if (addr != NULL) {
6110 small_page_write(addr, size);
6111
6112 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6113 }
6114 }
6115
6116 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6117 size_t lp = os::large_page_size();
6118 size_t ag = os::vm_allocation_granularity();
6119
6120 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6121 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6122 }
6123 }
6124
6125 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6126 size_t lp = os::large_page_size();
6127 size_t ag = os::vm_allocation_granularity();
6128
6129 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6130 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6131 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6132 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6133 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6134 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6135 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6136 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6137 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6138 }
6139
6140 static void test_reserve_memory_special_huge_tlbfs() {
6141 if (!UseHugeTLBFS) {
6142 return;
6143 }
6144
6145 test_reserve_memory_special_huge_tlbfs_only();
6146 test_reserve_memory_special_huge_tlbfs_mixed();
6147 }
6148
6149 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6150 if (!UseSHM) {
6151 return;
6152 }
6153
6154 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6155
6156 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6157
6158 if (addr != NULL) {
6159 assert(is_ptr_aligned(addr, alignment), "Check");
6160 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6161
6162 small_page_write(addr, size);
6163
6164 os::Linux::release_memory_special_shm(addr, size);
6165 }
6166 }
6167
6168 static void test_reserve_memory_special_shm() {
6169 size_t lp = os::large_page_size();
6170 size_t ag = os::vm_allocation_granularity();
6171
6172 for (size_t size = ag; size < lp * 3; size += ag) {
6173 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6174 test_reserve_memory_special_shm(size, alignment);
6175 }
6176 }
6177 }
6178
6179 static void test() {
6180 test_reserve_memory_special_huge_tlbfs();
6181 test_reserve_memory_special_shm();
6182 }
6183 };
6184
6185 void TestReserveMemorySpecial_test() {
6186 TestReserveMemorySpecial::test();
6187 }
6188
6189 #endif