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 #define EXTENSIONS_DIR "/lib/ext"
359 #define ENDORSED_DIR "/lib/endorsed"
360 #define REG_DIR "/usr/java/packages"
361
362 // sysclasspath, java_home, dll_dir
363 {
364 char *pslash;
365 char buf[MAXPATHLEN];
366 os::jvm_path(buf, sizeof(buf));
367
368 // Found the full path to libjvm.so.
369 // Now cut the path to <java_home>/jre if we can.
370 *(strrchr(buf, '/')) = '\0'; // Get rid of /libjvm.so.
371 pslash = strrchr(buf, '/');
372 if (pslash != NULL) {
373 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
374 }
375 Arguments::set_dll_dir(buf);
376
377 if (pslash != NULL) {
378 pslash = strrchr(buf, '/');
379 if (pslash != NULL) {
380 *pslash = '\0'; // Get rid of /<arch>.
381 pslash = strrchr(buf, '/');
382 if (pslash != NULL) {
383 *pslash = '\0'; // Get rid of /lib.
384 }
385 }
386 }
387 Arguments::set_java_home(buf);
388
389 if (!set_boot_path('/', ':')) {
390 return;
391 }
392 }
393
394 // Where to look for native libraries.
395 //
396 // Note: Due to a legacy implementation, most of the library path
397 // is set in the launcher. This was to accomodate linking restrictions
398 // on legacy Linux implementations (which are no longer supported).
399 // Eventually, all the library path setting will be done here.
400 //
401 // However, to prevent the proliferation of improperly built native
402 // libraries, the new path component /usr/java/packages is added here.
403 // Eventually, all the library path setting will be done here.
404 {
405 char *ld_library_path;
406
407 // Construct the invariant part of ld_library_path. Note that the
408 // space for the colon and the trailing null are provided by the
409 // nulls included by the sizeof operator (so actually we allocate
410 // a byte more than necessary).
411 ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, sizeof(REG_DIR) + sizeof("/lib/") +
412 strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH), mtInternal);
413 sprintf(ld_library_path, REG_DIR "/lib/%s:" DEFAULT_LIBPATH, cpu_arch);
414
415 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
416 // should always exist (until the legacy problem cited above is
417 // addressed).
418 char *v = ::getenv("LD_LIBRARY_PATH");
419 if (v != NULL) {
420 char *t = ld_library_path;
421 // That's +1 for the colon and +1 for the trailing '\0'.
422 ld_library_path = (char *)NEW_C_HEAP_ARRAY(char, strlen(v) + 1 + strlen(t) + 1, mtInternal);
423 sprintf(ld_library_path, "%s:%s", v, t);
424 FREE_C_HEAP_ARRAY(char, t, mtInternal);
425 }
426 Arguments::set_library_path(ld_library_path);
427 FREE_C_HEAP_ARRAY(char, ld_library_path, mtInternal);
428 }
429
430 // Buffer that fits both sprintfs.
431 // Note that the space for the colon and the trailing null are provided
432 // by the nulls included by the sizeof operator (so actually one byte more
433 // than necessary is allocated).
434 char buf[MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(REG_DIR) + sizeof(EXTENSIONS_DIR) + sizeof(ENDORSED_DIR)];
435
436 // Extensions directories.
437 sprintf(buf, "%s" EXTENSIONS_DIR ":" REG_DIR EXTENSIONS_DIR, Arguments::get_java_home());
438 Arguments::set_ext_dirs(buf);
439
440 // Endorsed standards default directory.
441 sprintf(buf, "%s" ENDORSED_DIR, Arguments::get_java_home());
442 Arguments::set_endorsed_dirs(buf);
443
444 #undef DEFAULT_LIBPATH
445 #undef EXTENSIONS_DIR
446 #undef ENDORSED_DIR
447 #undef REG_DIR
448 }
449
450 ////////////////////////////////////////////////////////////////////////////////
451 // breakpoint support
452
453 void os::breakpoint() {
454 BREAKPOINT;
455 }
456
457 extern "C" void breakpoint() {
458 // use debugger to set breakpoint here
459 }
460
461 ////////////////////////////////////////////////////////////////////////////////
462 // signal support
463
464 debug_only(static bool signal_sets_initialized = false);
465 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
466
467 bool os::Linux::is_sig_ignored(int sig) {
468 struct sigaction oact;
469 sigaction(sig, (struct sigaction*)NULL, &oact);
470 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
471 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
472 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN))
473 return true;
474 else
475 return false;
476 }
477
478 void os::Linux::signal_sets_init() {
479 // Should also have an assertion stating we are still single-threaded.
480 assert(!signal_sets_initialized, "Already initialized");
481 // Fill in signals that are necessarily unblocked for all threads in
482 // the VM. Currently, we unblock the following signals:
483 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
484 // by -Xrs (=ReduceSignalUsage));
485 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
486 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
487 // the dispositions or masks wrt these signals.
488 // Programs embedding the VM that want to use the above signals for their
489 // own purposes must, at this time, use the "-Xrs" option to prevent
490 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
491 // (See bug 4345157, and other related bugs).
492 // In reality, though, unblocking these signals is really a nop, since
493 // these signals are not blocked by default.
494 sigemptyset(&unblocked_sigs);
495 sigemptyset(&allowdebug_blocked_sigs);
496 sigaddset(&unblocked_sigs, SIGILL);
497 sigaddset(&unblocked_sigs, SIGSEGV);
498 sigaddset(&unblocked_sigs, SIGBUS);
499 sigaddset(&unblocked_sigs, SIGFPE);
500 #if defined(PPC64)
501 sigaddset(&unblocked_sigs, SIGTRAP);
502 #endif
503 sigaddset(&unblocked_sigs, SR_signum);
504
505 if (!ReduceSignalUsage) {
506 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
507 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
508 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
509 }
510 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
511 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
512 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
513 }
514 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
515 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
516 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
517 }
518 }
519 // Fill in signals that are blocked by all but the VM thread.
520 sigemptyset(&vm_sigs);
521 if (!ReduceSignalUsage)
522 sigaddset(&vm_sigs, BREAK_SIGNAL);
523 debug_only(signal_sets_initialized = true);
524
525 }
526
527 // These are signals that are unblocked while a thread is running Java.
528 // (For some reason, they get blocked by default.)
529 sigset_t* os::Linux::unblocked_signals() {
530 assert(signal_sets_initialized, "Not initialized");
531 return &unblocked_sigs;
532 }
533
534 // These are the signals that are blocked while a (non-VM) thread is
535 // running Java. Only the VM thread handles these signals.
536 sigset_t* os::Linux::vm_signals() {
537 assert(signal_sets_initialized, "Not initialized");
538 return &vm_sigs;
539 }
540
541 // These are signals that are blocked during cond_wait to allow debugger in
542 sigset_t* os::Linux::allowdebug_blocked_signals() {
543 assert(signal_sets_initialized, "Not initialized");
544 return &allowdebug_blocked_sigs;
545 }
546
547 void os::Linux::hotspot_sigmask(Thread* thread) {
548
549 //Save caller's signal mask before setting VM signal mask
550 sigset_t caller_sigmask;
551 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
552
553 OSThread* osthread = thread->osthread();
554 osthread->set_caller_sigmask(caller_sigmask);
555
556 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
557
558 if (!ReduceSignalUsage) {
559 if (thread->is_VM_thread()) {
560 // Only the VM thread handles BREAK_SIGNAL ...
561 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
562 } else {
563 // ... all other threads block BREAK_SIGNAL
564 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
565 }
566 }
567 }
568
569 //////////////////////////////////////////////////////////////////////////////
570 // detecting pthread library
571
572 void os::Linux::libpthread_init() {
573 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
574 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
575 // generic name for earlier versions.
576 // Define macros here so we can build HotSpot on old systems.
577 # ifndef _CS_GNU_LIBC_VERSION
578 # define _CS_GNU_LIBC_VERSION 2
579 # endif
580 # ifndef _CS_GNU_LIBPTHREAD_VERSION
581 # define _CS_GNU_LIBPTHREAD_VERSION 3
582 # endif
583
584 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
585 if (n > 0) {
586 char *str = (char *)malloc(n, mtInternal);
587 confstr(_CS_GNU_LIBC_VERSION, str, n);
588 os::Linux::set_glibc_version(str);
589 } else {
590 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
591 static char _gnu_libc_version[32];
592 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
593 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
594 os::Linux::set_glibc_version(_gnu_libc_version);
595 }
596
597 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
598 if (n > 0) {
599 char *str = (char *)malloc(n, mtInternal);
600 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
601 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
602 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
603 // is the case. LinuxThreads has a hard limit on max number of threads.
604 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
605 // On the other hand, NPTL does not have such a limit, sysconf()
606 // will return -1 and errno is not changed. Check if it is really NPTL.
607 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
608 strstr(str, "NPTL") &&
609 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
610 free(str);
611 os::Linux::set_libpthread_version("linuxthreads");
612 } else {
613 os::Linux::set_libpthread_version(str);
614 }
615 } else {
616 // glibc before 2.3.2 only has LinuxThreads.
617 os::Linux::set_libpthread_version("linuxthreads");
618 }
619
620 if (strstr(libpthread_version(), "NPTL")) {
621 os::Linux::set_is_NPTL();
622 } else {
623 os::Linux::set_is_LinuxThreads();
624 }
625
626 // LinuxThreads have two flavors: floating-stack mode, which allows variable
627 // stack size; and fixed-stack mode. NPTL is always floating-stack.
628 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
629 os::Linux::set_is_floating_stack();
630 }
631 }
632
633 /////////////////////////////////////////////////////////////////////////////
634 // thread stack
635
636 // Force Linux kernel to expand current thread stack. If "bottom" is close
637 // to the stack guard, caller should block all signals.
638 //
639 // MAP_GROWSDOWN:
640 // A special mmap() flag that is used to implement thread stacks. It tells
641 // kernel that the memory region should extend downwards when needed. This
642 // allows early versions of LinuxThreads to only mmap the first few pages
643 // when creating a new thread. Linux kernel will automatically expand thread
644 // stack as needed (on page faults).
645 //
646 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
647 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
648 // region, it's hard to tell if the fault is due to a legitimate stack
649 // access or because of reading/writing non-exist memory (e.g. buffer
650 // overrun). As a rule, if the fault happens below current stack pointer,
651 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
652 // application (see Linux kernel fault.c).
653 //
654 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
655 // stack overflow detection.
656 //
657 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
658 // not use this flag. However, the stack of initial thread is not created
659 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
660 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
661 // and then attach the thread to JVM.
662 //
663 // To get around the problem and allow stack banging on Linux, we need to
664 // manually expand thread stack after receiving the SIGSEGV.
665 //
666 // There are two ways to expand thread stack to address "bottom", we used
667 // both of them in JVM before 1.5:
668 // 1. adjust stack pointer first so that it is below "bottom", and then
669 // touch "bottom"
670 // 2. mmap() the page in question
671 //
672 // Now alternate signal stack is gone, it's harder to use 2. For instance,
673 // if current sp is already near the lower end of page 101, and we need to
674 // call mmap() to map page 100, it is possible that part of the mmap() frame
675 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
676 // That will destroy the mmap() frame and cause VM to crash.
677 //
678 // The following code works by adjusting sp first, then accessing the "bottom"
679 // page to force a page fault. Linux kernel will then automatically expand the
680 // stack mapping.
681 //
682 // _expand_stack_to() assumes its frame size is less than page size, which
683 // should always be true if the function is not inlined.
684
685 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
686 #define NOINLINE
687 #else
688 #define NOINLINE __attribute__ ((noinline))
689 #endif
690
691 static void _expand_stack_to(address bottom) NOINLINE;
692
693 static void _expand_stack_to(address bottom) {
694 address sp;
695 size_t size;
696 volatile char *p;
697
698 // Adjust bottom to point to the largest address within the same page, it
699 // gives us a one-page buffer if alloca() allocates slightly more memory.
700 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
701 bottom += os::Linux::page_size() - 1;
702
703 // sp might be slightly above current stack pointer; if that's the case, we
704 // will alloca() a little more space than necessary, which is OK. Don't use
705 // os::current_stack_pointer(), as its result can be slightly below current
706 // stack pointer, causing us to not alloca enough to reach "bottom".
707 sp = (address)&sp;
708
709 if (sp > bottom) {
710 size = sp - bottom;
711 p = (volatile char *)alloca(size);
712 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
713 p[0] = '\0';
714 }
715 }
716
717 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
718 assert(t!=NULL, "just checking");
719 assert(t->osthread()->expanding_stack(), "expand should be set");
720 assert(t->stack_base() != NULL, "stack_base was not initialized");
721
722 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
723 sigset_t mask_all, old_sigset;
724 sigfillset(&mask_all);
725 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
726 _expand_stack_to(addr);
727 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
728 return true;
729 }
730 return false;
731 }
732
733 //////////////////////////////////////////////////////////////////////////////
734 // create new thread
735
736 static address highest_vm_reserved_address();
737
738 // check if it's safe to start a new thread
739 static bool _thread_safety_check(Thread* thread) {
740 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
741 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
742 // Heap is mmap'ed at lower end of memory space. Thread stacks are
743 // allocated (MAP_FIXED) from high address space. Every thread stack
744 // occupies a fixed size slot (usually 2Mbytes, but user can change
745 // it to other values if they rebuild LinuxThreads).
746 //
747 // Problem with MAP_FIXED is that mmap() can still succeed even part of
748 // the memory region has already been mmap'ed. That means if we have too
749 // many threads and/or very large heap, eventually thread stack will
750 // collide with heap.
751 //
752 // Here we try to prevent heap/stack collision by comparing current
753 // stack bottom with the highest address that has been mmap'ed by JVM
754 // plus a safety margin for memory maps created by native code.
755 //
756 // This feature can be disabled by setting ThreadSafetyMargin to 0
757 //
758 if (ThreadSafetyMargin > 0) {
759 address stack_bottom = os::current_stack_base() - os::current_stack_size();
760
761 // not safe if our stack extends below the safety margin
762 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
763 } else {
764 return true;
765 }
766 } else {
767 // Floating stack LinuxThreads or NPTL:
768 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
769 // there's not enough space left, pthread_create() will fail. If we come
770 // here, that means enough space has been reserved for stack.
771 return true;
772 }
773 }
774
775 // Thread start routine for all newly created threads
776 static void *java_start(Thread *thread) {
777 // Try to randomize the cache line index of hot stack frames.
778 // This helps when threads of the same stack traces evict each other's
779 // cache lines. The threads can be either from the same JVM instance, or
780 // from different JVM instances. The benefit is especially true for
781 // processors with hyperthreading technology.
782 static int counter = 0;
783 int pid = os::current_process_id();
784 alloca(((pid ^ counter++) & 7) * 128);
785
786 ThreadLocalStorage::set_thread(thread);
787
788 OSThread* osthread = thread->osthread();
789 Monitor* sync = osthread->startThread_lock();
790
791 // non floating stack LinuxThreads needs extra check, see above
792 if (!_thread_safety_check(thread)) {
793 // notify parent thread
794 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
795 osthread->set_state(ZOMBIE);
796 sync->notify_all();
797 return NULL;
798 }
799
800 // thread_id is kernel thread id (similar to Solaris LWP id)
801 osthread->set_thread_id(os::Linux::gettid());
802
803 if (UseNUMA) {
804 int lgrp_id = os::numa_get_group_id();
805 if (lgrp_id != -1) {
806 thread->set_lgrp_id(lgrp_id);
807 }
808 }
809 // initialize signal mask for this thread
810 os::Linux::hotspot_sigmask(thread);
811
812 // initialize floating point control register
813 os::Linux::init_thread_fpu_state();
814
815 // handshaking with parent thread
816 {
817 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
818
819 // notify parent thread
820 osthread->set_state(INITIALIZED);
821 sync->notify_all();
822
823 // wait until os::start_thread()
824 while (osthread->get_state() == INITIALIZED) {
825 sync->wait(Mutex::_no_safepoint_check_flag);
826 }
827 }
828
829 // call one more level start routine
830 thread->run();
831
832 return 0;
833 }
834
835 bool os::create_thread(Thread* thread, ThreadType thr_type, size_t stack_size) {
836 assert(thread->osthread() == NULL, "caller responsible");
837
838 // Allocate the OSThread object
839 OSThread* osthread = new OSThread(NULL, NULL);
840 if (osthread == NULL) {
841 return false;
842 }
843
844 // set the correct thread state
845 osthread->set_thread_type(thr_type);
846
847 // Initial state is ALLOCATED but not INITIALIZED
848 osthread->set_state(ALLOCATED);
849
850 thread->set_osthread(osthread);
851
852 // init thread attributes
853 pthread_attr_t attr;
854 pthread_attr_init(&attr);
855 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
856
857 // stack size
858 if (os::Linux::supports_variable_stack_size()) {
859 // calculate stack size if it's not specified by caller
860 if (stack_size == 0) {
861 stack_size = os::Linux::default_stack_size(thr_type);
862
863 switch (thr_type) {
864 case os::java_thread:
865 // Java threads use ThreadStackSize which default value can be
866 // changed with the flag -Xss
867 assert (JavaThread::stack_size_at_create() > 0, "this should be set");
868 stack_size = JavaThread::stack_size_at_create();
869 break;
870 case os::compiler_thread:
871 if (CompilerThreadStackSize > 0) {
872 stack_size = (size_t)(CompilerThreadStackSize * K);
873 break;
874 } // else fall through:
875 // use VMThreadStackSize if CompilerThreadStackSize is not defined
876 case os::vm_thread:
877 case os::pgc_thread:
878 case os::cgc_thread:
879 case os::watcher_thread:
880 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
881 break;
882 }
883 }
884
885 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
886 pthread_attr_setstacksize(&attr, stack_size);
887 } else {
888 // let pthread_create() pick the default value.
889 }
890
891 // glibc guard page
892 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
893
894 ThreadState state;
895
896 {
897 // Serialize thread creation if we are running with fixed stack LinuxThreads
898 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
899 if (lock) {
900 os::Linux::createThread_lock()->lock_without_safepoint_check();
901 }
902
903 pthread_t tid;
904 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
905
906 pthread_attr_destroy(&attr);
907
908 if (ret != 0) {
909 if (PrintMiscellaneous && (Verbose || WizardMode)) {
910 perror("pthread_create()");
911 }
912 // Need to clean up stuff we've allocated so far
913 thread->set_osthread(NULL);
914 delete osthread;
915 if (lock) os::Linux::createThread_lock()->unlock();
916 return false;
917 }
918
919 // Store pthread info into the OSThread
920 osthread->set_pthread_id(tid);
921
922 // Wait until child thread is either initialized or aborted
923 {
924 Monitor* sync_with_child = osthread->startThread_lock();
925 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
926 while ((state = osthread->get_state()) == ALLOCATED) {
927 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
928 }
929 }
930
931 if (lock) {
932 os::Linux::createThread_lock()->unlock();
933 }
934 }
935
936 // Aborted due to thread limit being reached
937 if (state == ZOMBIE) {
938 thread->set_osthread(NULL);
939 delete osthread;
940 return false;
941 }
942
943 // The thread is returned suspended (in state INITIALIZED),
944 // and is started higher up in the call chain
945 assert(state == INITIALIZED, "race condition");
946 return true;
947 }
948
949 /////////////////////////////////////////////////////////////////////////////
950 // attach existing thread
951
952 // bootstrap the main thread
953 bool os::create_main_thread(JavaThread* thread) {
954 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
955 return create_attached_thread(thread);
956 }
957
958 bool os::create_attached_thread(JavaThread* thread) {
959 #ifdef ASSERT
960 thread->verify_not_published();
961 #endif
962
963 // Allocate the OSThread object
964 OSThread* osthread = new OSThread(NULL, NULL);
965
966 if (osthread == NULL) {
967 return false;
968 }
969
970 // Store pthread info into the OSThread
971 osthread->set_thread_id(os::Linux::gettid());
972 osthread->set_pthread_id(::pthread_self());
973
974 // initialize floating point control register
975 os::Linux::init_thread_fpu_state();
976
977 // Initial thread state is RUNNABLE
978 osthread->set_state(RUNNABLE);
979
980 thread->set_osthread(osthread);
981
982 if (UseNUMA) {
983 int lgrp_id = os::numa_get_group_id();
984 if (lgrp_id != -1) {
985 thread->set_lgrp_id(lgrp_id);
986 }
987 }
988
989 if (os::Linux::is_initial_thread()) {
990 // If current thread is initial thread, its stack is mapped on demand,
991 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
992 // the entire stack region to avoid SEGV in stack banging.
993 // It is also useful to get around the heap-stack-gap problem on SuSE
994 // kernel (see 4821821 for details). We first expand stack to the top
995 // of yellow zone, then enable stack yellow zone (order is significant,
996 // enabling yellow zone first will crash JVM on SuSE Linux), so there
997 // is no gap between the last two virtual memory regions.
998
999 JavaThread *jt = (JavaThread *)thread;
1000 address addr = jt->stack_yellow_zone_base();
1001 assert(addr != NULL, "initialization problem?");
1002 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
1003
1004 osthread->set_expanding_stack();
1005 os::Linux::manually_expand_stack(jt, addr);
1006 osthread->clear_expanding_stack();
1007 }
1008
1009 // initialize signal mask for this thread
1010 // and save the caller's signal mask
1011 os::Linux::hotspot_sigmask(thread);
1012
1013 return true;
1014 }
1015
1016 void os::pd_start_thread(Thread* thread) {
1017 OSThread * osthread = thread->osthread();
1018 assert(osthread->get_state() != INITIALIZED, "just checking");
1019 Monitor* sync_with_child = osthread->startThread_lock();
1020 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
1021 sync_with_child->notify();
1022 }
1023
1024 // Free Linux resources related to the OSThread
1025 void os::free_thread(OSThread* osthread) {
1026 assert(osthread != NULL, "osthread not set");
1027
1028 if (Thread::current()->osthread() == osthread) {
1029 // Restore caller's signal mask
1030 sigset_t sigmask = osthread->caller_sigmask();
1031 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
1032 }
1033
1034 delete osthread;
1035 }
1036
1037 //////////////////////////////////////////////////////////////////////////////
1038 // thread local storage
1039
1040 int os::allocate_thread_local_storage() {
1041 pthread_key_t key;
1042 int rslt = pthread_key_create(&key, NULL);
1043 assert(rslt == 0, "cannot allocate thread local storage");
1044 return (int)key;
1045 }
1046
1047 // Note: This is currently not used by VM, as we don't destroy TLS key
1048 // on VM exit.
1049 void os::free_thread_local_storage(int index) {
1050 int rslt = pthread_key_delete((pthread_key_t)index);
1051 assert(rslt == 0, "invalid index");
1052 }
1053
1054 void os::thread_local_storage_at_put(int index, void* value) {
1055 int rslt = pthread_setspecific((pthread_key_t)index, value);
1056 assert(rslt == 0, "pthread_setspecific failed");
1057 }
1058
1059 extern "C" Thread* get_thread() {
1060 return ThreadLocalStorage::thread();
1061 }
1062
1063 //////////////////////////////////////////////////////////////////////////////
1064 // initial thread
1065
1066 // Check if current thread is the initial thread, similar to Solaris thr_main.
1067 bool os::Linux::is_initial_thread(void) {
1068 char dummy;
1069 // If called before init complete, thread stack bottom will be null.
1070 // Can be called if fatal error occurs before initialization.
1071 if (initial_thread_stack_bottom() == NULL) return false;
1072 assert(initial_thread_stack_bottom() != NULL &&
1073 initial_thread_stack_size() != 0,
1074 "os::init did not locate initial thread's stack region");
1075 if ((address)&dummy >= initial_thread_stack_bottom() &&
1076 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size())
1077 return true;
1078 else return false;
1079 }
1080
1081 // Find the virtual memory area that contains addr
1082 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1083 FILE *fp = fopen("/proc/self/maps", "r");
1084 if (fp) {
1085 address low, high;
1086 while (!feof(fp)) {
1087 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1088 if (low <= addr && addr < high) {
1089 if (vma_low) *vma_low = low;
1090 if (vma_high) *vma_high = high;
1091 fclose (fp);
1092 return true;
1093 }
1094 }
1095 for (;;) {
1096 int ch = fgetc(fp);
1097 if (ch == EOF || ch == (int)'\n') break;
1098 }
1099 }
1100 fclose(fp);
1101 }
1102 return false;
1103 }
1104
1105 // Locate initial thread stack. This special handling of initial thread stack
1106 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1107 // bogus value for initial thread.
1108 void os::Linux::capture_initial_stack(size_t max_size) {
1109 // stack size is the easy part, get it from RLIMIT_STACK
1110 size_t stack_size;
1111 struct rlimit rlim;
1112 getrlimit(RLIMIT_STACK, &rlim);
1113 stack_size = rlim.rlim_cur;
1114
1115 // 6308388: a bug in ld.so will relocate its own .data section to the
1116 // lower end of primordial stack; reduce ulimit -s value a little bit
1117 // so we won't install guard page on ld.so's data section.
1118 stack_size -= 2 * page_size();
1119
1120 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1121 // 7.1, in both cases we will get 2G in return value.
1122 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1123 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1124 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1125 // in case other parts in glibc still assumes 2M max stack size.
1126 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1127 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1128 if (stack_size > 2 * K * K IA64_ONLY(*2))
1129 stack_size = 2 * K * K IA64_ONLY(*2);
1130 // Try to figure out where the stack base (top) is. This is harder.
1131 //
1132 // When an application is started, glibc saves the initial stack pointer in
1133 // a global variable "__libc_stack_end", which is then used by system
1134 // libraries. __libc_stack_end should be pretty close to stack top. The
1135 // variable is available since the very early days. However, because it is
1136 // a private interface, it could disappear in the future.
1137 //
1138 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1139 // to __libc_stack_end, it is very close to stack top, but isn't the real
1140 // stack top. Note that /proc may not exist if VM is running as a chroot
1141 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1142 // /proc/<pid>/stat could change in the future (though unlikely).
1143 //
1144 // We try __libc_stack_end first. If that doesn't work, look for
1145 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1146 // as a hint, which should work well in most cases.
1147
1148 uintptr_t stack_start;
1149
1150 // try __libc_stack_end first
1151 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1152 if (p && *p) {
1153 stack_start = *p;
1154 } else {
1155 // see if we can get the start_stack field from /proc/self/stat
1156 FILE *fp;
1157 int pid;
1158 char state;
1159 int ppid;
1160 int pgrp;
1161 int session;
1162 int nr;
1163 int tpgrp;
1164 unsigned long flags;
1165 unsigned long minflt;
1166 unsigned long cminflt;
1167 unsigned long majflt;
1168 unsigned long cmajflt;
1169 unsigned long utime;
1170 unsigned long stime;
1171 long cutime;
1172 long cstime;
1173 long prio;
1174 long nice;
1175 long junk;
1176 long it_real;
1177 uintptr_t start;
1178 uintptr_t vsize;
1179 intptr_t rss;
1180 uintptr_t rsslim;
1181 uintptr_t scodes;
1182 uintptr_t ecode;
1183 int i;
1184
1185 // Figure what the primordial thread stack base is. Code is inspired
1186 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1187 // followed by command name surrounded by parentheses, state, etc.
1188 char stat[2048];
1189 int statlen;
1190
1191 fp = fopen("/proc/self/stat", "r");
1192 if (fp) {
1193 statlen = fread(stat, 1, 2047, fp);
1194 stat[statlen] = '\0';
1195 fclose(fp);
1196
1197 // Skip pid and the command string. Note that we could be dealing with
1198 // weird command names, e.g. user could decide to rename java launcher
1199 // to "java 1.4.2 :)", then the stat file would look like
1200 // 1234 (java 1.4.2 :)) R ... ...
1201 // We don't really need to know the command string, just find the last
1202 // occurrence of ")" and then start parsing from there. See bug 4726580.
1203 char * s = strrchr(stat, ')');
1204
1205 i = 0;
1206 if (s) {
1207 // Skip blank chars
1208 do s++; while (isspace(*s));
1209
1210 #define _UFM UINTX_FORMAT
1211 #define _DFM INTX_FORMAT
1212
1213 /* 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 */
1214 /* 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 */
1215 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,
1216 &state, /* 3 %c */
1217 &ppid, /* 4 %d */
1218 &pgrp, /* 5 %d */
1219 &session, /* 6 %d */
1220 &nr, /* 7 %d */
1221 &tpgrp, /* 8 %d */
1222 &flags, /* 9 %lu */
1223 &minflt, /* 10 %lu */
1224 &cminflt, /* 11 %lu */
1225 &majflt, /* 12 %lu */
1226 &cmajflt, /* 13 %lu */
1227 &utime, /* 14 %lu */
1228 &stime, /* 15 %lu */
1229 &cutime, /* 16 %ld */
1230 &cstime, /* 17 %ld */
1231 &prio, /* 18 %ld */
1232 &nice, /* 19 %ld */
1233 &junk, /* 20 %ld */
1234 &it_real, /* 21 %ld */
1235 &start, /* 22 UINTX_FORMAT */
1236 &vsize, /* 23 UINTX_FORMAT */
1237 &rss, /* 24 INTX_FORMAT */
1238 &rsslim, /* 25 UINTX_FORMAT */
1239 &scodes, /* 26 UINTX_FORMAT */
1240 &ecode, /* 27 UINTX_FORMAT */
1241 &stack_start); /* 28 UINTX_FORMAT */
1242 }
1243
1244 #undef _UFM
1245 #undef _DFM
1246
1247 if (i != 28 - 2) {
1248 assert(false, "Bad conversion from /proc/self/stat");
1249 // product mode - assume we are the initial thread, good luck in the
1250 // embedded case.
1251 warning("Can't detect initial thread stack location - bad conversion");
1252 stack_start = (uintptr_t) &rlim;
1253 }
1254 } else {
1255 // For some reason we can't open /proc/self/stat (for example, running on
1256 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1257 // most cases, so don't abort:
1258 warning("Can't detect initial thread stack location - no /proc/self/stat");
1259 stack_start = (uintptr_t) &rlim;
1260 }
1261 }
1262
1263 // Now we have a pointer (stack_start) very close to the stack top, the
1264 // next thing to do is to figure out the exact location of stack top. We
1265 // can find out the virtual memory area that contains stack_start by
1266 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1267 // and its upper limit is the real stack top. (again, this would fail if
1268 // running inside chroot, because /proc may not exist.)
1269
1270 uintptr_t stack_top;
1271 address low, high;
1272 if (find_vma((address)stack_start, &low, &high)) {
1273 // success, "high" is the true stack top. (ignore "low", because initial
1274 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1275 stack_top = (uintptr_t)high;
1276 } else {
1277 // failed, likely because /proc/self/maps does not exist
1278 warning("Can't detect initial thread stack location - find_vma failed");
1279 // best effort: stack_start is normally within a few pages below the real
1280 // stack top, use it as stack top, and reduce stack size so we won't put
1281 // guard page outside stack.
1282 stack_top = stack_start;
1283 stack_size -= 16 * page_size();
1284 }
1285
1286 // stack_top could be partially down the page so align it
1287 stack_top = align_size_up(stack_top, page_size());
1288
1289 if (max_size && stack_size > max_size) {
1290 _initial_thread_stack_size = max_size;
1291 } else {
1292 _initial_thread_stack_size = stack_size;
1293 }
1294
1295 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1296 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1297 }
1298
1299 ////////////////////////////////////////////////////////////////////////////////
1300 // time support
1301
1302 // Time since start-up in seconds to a fine granularity.
1303 // Used by VMSelfDestructTimer and the MemProfiler.
1304 double os::elapsedTime() {
1305
1306 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1307 }
1308
1309 jlong os::elapsed_counter() {
1310 return javaTimeNanos() - initial_time_count;
1311 }
1312
1313 jlong os::elapsed_frequency() {
1314 return NANOSECS_PER_SEC; // nanosecond resolution
1315 }
1316
1317 bool os::supports_vtime() { return true; }
1318 bool os::enable_vtime() { return false; }
1319 bool os::vtime_enabled() { return false; }
1320
1321 double os::elapsedVTime() {
1322 struct rusage usage;
1323 int retval = getrusage(RUSAGE_THREAD, &usage);
1324 if (retval == 0) {
1325 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);
1326 } else {
1327 // better than nothing, but not much
1328 return elapsedTime();
1329 }
1330 }
1331
1332 jlong os::javaTimeMillis() {
1333 timeval time;
1334 int status = gettimeofday(&time, NULL);
1335 assert(status != -1, "linux error");
1336 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1337 }
1338
1339 #ifndef CLOCK_MONOTONIC
1340 #define CLOCK_MONOTONIC (1)
1341 #endif
1342
1343 void os::Linux::clock_init() {
1344 // we do dlopen's in this particular order due to bug in linux
1345 // dynamical loader (see 6348968) leading to crash on exit
1346 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1347 if (handle == NULL) {
1348 handle = dlopen("librt.so", RTLD_LAZY);
1349 }
1350
1351 if (handle) {
1352 int (*clock_getres_func)(clockid_t, struct timespec*) =
1353 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1354 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1355 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1356 if (clock_getres_func && clock_gettime_func) {
1357 // See if monotonic clock is supported by the kernel. Note that some
1358 // early implementations simply return kernel jiffies (updated every
1359 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1360 // for nano time (though the monotonic property is still nice to have).
1361 // It's fixed in newer kernels, however clock_getres() still returns
1362 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1363 // resolution for now. Hopefully as people move to new kernels, this
1364 // won't be a problem.
1365 struct timespec res;
1366 struct timespec tp;
1367 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1368 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1369 // yes, monotonic clock is supported
1370 _clock_gettime = clock_gettime_func;
1371 return;
1372 } else {
1373 // close librt if there is no monotonic clock
1374 dlclose(handle);
1375 }
1376 }
1377 }
1378 warning("No monotonic clock was available - timed services may " \
1379 "be adversely affected if the time-of-day clock changes");
1380 }
1381
1382 #ifndef SYS_clock_getres
1383
1384 #if defined(IA32) || defined(AMD64)
1385 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1386 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1387 #else
1388 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1389 #define sys_clock_getres(x,y) -1
1390 #endif
1391
1392 #else
1393 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1394 #endif
1395
1396 void os::Linux::fast_thread_clock_init() {
1397 if (!UseLinuxPosixThreadCPUClocks) {
1398 return;
1399 }
1400 clockid_t clockid;
1401 struct timespec tp;
1402 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1403 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1404
1405 // Switch to using fast clocks for thread cpu time if
1406 // the sys_clock_getres() returns 0 error code.
1407 // Note, that some kernels may support the current thread
1408 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1409 // returned by the pthread_getcpuclockid().
1410 // If the fast Posix clocks are supported then the sys_clock_getres()
1411 // must return at least tp.tv_sec == 0 which means a resolution
1412 // better than 1 sec. This is extra check for reliability.
1413
1414 if(pthread_getcpuclockid_func &&
1415 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1416 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1417
1418 _supports_fast_thread_cpu_time = true;
1419 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1420 }
1421 }
1422
1423 jlong os::javaTimeNanos() {
1424 if (os::supports_monotonic_clock()) {
1425 struct timespec tp;
1426 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1427 assert(status == 0, "gettime error");
1428 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1429 return result;
1430 } else {
1431 timeval time;
1432 int status = gettimeofday(&time, NULL);
1433 assert(status != -1, "linux error");
1434 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1435 return 1000 * usecs;
1436 }
1437 }
1438
1439 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1440 if (os::supports_monotonic_clock()) {
1441 info_ptr->max_value = ALL_64_BITS;
1442
1443 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1444 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1445 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1446 } else {
1447 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1448 info_ptr->max_value = ALL_64_BITS;
1449
1450 // gettimeofday is a real time clock so it skips
1451 info_ptr->may_skip_backward = true;
1452 info_ptr->may_skip_forward = true;
1453 }
1454
1455 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1456 }
1457
1458 // Return the real, user, and system times in seconds from an
1459 // arbitrary fixed point in the past.
1460 bool os::getTimesSecs(double* process_real_time,
1461 double* process_user_time,
1462 double* process_system_time) {
1463 struct tms ticks;
1464 clock_t real_ticks = times(&ticks);
1465
1466 if (real_ticks == (clock_t) (-1)) {
1467 return false;
1468 } else {
1469 double ticks_per_second = (double) clock_tics_per_sec;
1470 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1471 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1472 *process_real_time = ((double) real_ticks) / ticks_per_second;
1473
1474 return true;
1475 }
1476 }
1477
1478
1479 char * os::local_time_string(char *buf, size_t buflen) {
1480 struct tm t;
1481 time_t long_time;
1482 time(&long_time);
1483 localtime_r(&long_time, &t);
1484 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1485 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1486 t.tm_hour, t.tm_min, t.tm_sec);
1487 return buf;
1488 }
1489
1490 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1491 return localtime_r(clock, res);
1492 }
1493
1494 ////////////////////////////////////////////////////////////////////////////////
1495 // runtime exit support
1496
1497 // Note: os::shutdown() might be called very early during initialization, or
1498 // called from signal handler. Before adding something to os::shutdown(), make
1499 // sure it is async-safe and can handle partially initialized VM.
1500 void os::shutdown() {
1501
1502 // allow PerfMemory to attempt cleanup of any persistent resources
1503 perfMemory_exit();
1504
1505 // needs to remove object in file system
1506 AttachListener::abort();
1507
1508 // flush buffered output, finish log files
1509 ostream_abort();
1510
1511 // Check for abort hook
1512 abort_hook_t abort_hook = Arguments::abort_hook();
1513 if (abort_hook != NULL) {
1514 abort_hook();
1515 }
1516
1517 }
1518
1519 // Note: os::abort() might be called very early during initialization, or
1520 // called from signal handler. Before adding something to os::abort(), make
1521 // sure it is async-safe and can handle partially initialized VM.
1522 void os::abort(bool dump_core) {
1523 os::shutdown();
1524 if (dump_core) {
1525 #ifndef PRODUCT
1526 fdStream out(defaultStream::output_fd());
1527 out.print_raw("Current thread is ");
1528 char buf[16];
1529 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1530 out.print_raw_cr(buf);
1531 out.print_raw_cr("Dumping core ...");
1532 #endif
1533 ::abort(); // dump core
1534 }
1535
1536 ::exit(1);
1537 }
1538
1539 // Die immediately, no exit hook, no abort hook, no cleanup.
1540 void os::die() {
1541 // _exit() on LinuxThreads only kills current thread
1542 ::abort();
1543 }
1544
1545 // unused on linux for now.
1546 void os::set_error_file(const char *logfile) {}
1547
1548
1549 // This method is a copy of JDK's sysGetLastErrorString
1550 // from src/solaris/hpi/src/system_md.c
1551
1552 size_t os::lasterror(char *buf, size_t len) {
1553
1554 if (errno == 0) return 0;
1555
1556 const char *s = ::strerror(errno);
1557 size_t n = ::strlen(s);
1558 if (n >= len) {
1559 n = len - 1;
1560 }
1561 ::strncpy(buf, s, n);
1562 buf[n] = '\0';
1563 return n;
1564 }
1565
1566 intx os::current_thread_id() { return (intx)pthread_self(); }
1567 int os::current_process_id() {
1568
1569 // Under the old linux thread library, linux gives each thread
1570 // its own process id. Because of this each thread will return
1571 // a different pid if this method were to return the result
1572 // of getpid(2). Linux provides no api that returns the pid
1573 // of the launcher thread for the vm. This implementation
1574 // returns a unique pid, the pid of the launcher thread
1575 // that starts the vm 'process'.
1576
1577 // Under the NPTL, getpid() returns the same pid as the
1578 // launcher thread rather than a unique pid per thread.
1579 // Use gettid() if you want the old pre NPTL behaviour.
1580
1581 // if you are looking for the result of a call to getpid() that
1582 // returns a unique pid for the calling thread, then look at the
1583 // OSThread::thread_id() method in osThread_linux.hpp file
1584
1585 return (int)(_initial_pid ? _initial_pid : getpid());
1586 }
1587
1588 // DLL functions
1589
1590 const char* os::dll_file_extension() { return ".so"; }
1591
1592 // This must be hard coded because it's the system's temporary
1593 // directory not the java application's temp directory, ala java.io.tmpdir.
1594 const char* os::get_temp_directory() { return "/tmp"; }
1595
1596 static bool file_exists(const char* filename) {
1597 struct stat statbuf;
1598 if (filename == NULL || strlen(filename) == 0) {
1599 return false;
1600 }
1601 return os::stat(filename, &statbuf) == 0;
1602 }
1603
1604 bool os::dll_build_name(char* buffer, size_t buflen,
1605 const char* pname, const char* fname) {
1606 bool retval = false;
1607 // Copied from libhpi
1608 const size_t pnamelen = pname ? strlen(pname) : 0;
1609
1610 // Return error on buffer overflow.
1611 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1612 return retval;
1613 }
1614
1615 if (pnamelen == 0) {
1616 snprintf(buffer, buflen, "lib%s.so", fname);
1617 retval = true;
1618 } else if (strchr(pname, *os::path_separator()) != NULL) {
1619 int n;
1620 char** pelements = split_path(pname, &n);
1621 if (pelements == NULL) {
1622 return false;
1623 }
1624 for (int i = 0 ; i < n ; i++) {
1625 // Really shouldn't be NULL, but check can't hurt
1626 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1627 continue; // skip the empty path values
1628 }
1629 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1630 if (file_exists(buffer)) {
1631 retval = true;
1632 break;
1633 }
1634 }
1635 // release the storage
1636 for (int i = 0 ; i < n ; i++) {
1637 if (pelements[i] != NULL) {
1638 FREE_C_HEAP_ARRAY(char, pelements[i], mtInternal);
1639 }
1640 }
1641 if (pelements != NULL) {
1642 FREE_C_HEAP_ARRAY(char*, pelements, mtInternal);
1643 }
1644 } else {
1645 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1646 retval = true;
1647 }
1648 return retval;
1649 }
1650
1651 // check if addr is inside libjvm.so
1652 bool os::address_is_in_vm(address addr) {
1653 static address libjvm_base_addr;
1654 Dl_info dlinfo;
1655
1656 if (libjvm_base_addr == NULL) {
1657 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1658 libjvm_base_addr = (address)dlinfo.dli_fbase;
1659 }
1660 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1661 }
1662
1663 if (dladdr((void *)addr, &dlinfo) != 0) {
1664 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1665 }
1666
1667 return false;
1668 }
1669
1670 bool os::dll_address_to_function_name(address addr, char *buf,
1671 int buflen, int *offset) {
1672 // buf is not optional, but offset is optional
1673 assert(buf != NULL, "sanity check");
1674
1675 Dl_info dlinfo;
1676
1677 if (dladdr((void*)addr, &dlinfo) != 0) {
1678 // see if we have a matching symbol
1679 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1680 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1681 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1682 }
1683 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1684 return true;
1685 }
1686 // no matching symbol so try for just file info
1687 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1688 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1689 buf, buflen, offset, dlinfo.dli_fname)) {
1690 return true;
1691 }
1692 }
1693 }
1694
1695 buf[0] = '\0';
1696 if (offset != NULL) *offset = -1;
1697 return false;
1698 }
1699
1700 struct _address_to_library_name {
1701 address addr; // input : memory address
1702 size_t buflen; // size of fname
1703 char* fname; // output: library name
1704 address base; // library base addr
1705 };
1706
1707 static int address_to_library_name_callback(struct dl_phdr_info *info,
1708 size_t size, void *data) {
1709 int i;
1710 bool found = false;
1711 address libbase = NULL;
1712 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1713
1714 // iterate through all loadable segments
1715 for (i = 0; i < info->dlpi_phnum; i++) {
1716 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1717 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1718 // base address of a library is the lowest address of its loaded
1719 // segments.
1720 if (libbase == NULL || libbase > segbase) {
1721 libbase = segbase;
1722 }
1723 // see if 'addr' is within current segment
1724 if (segbase <= d->addr &&
1725 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1726 found = true;
1727 }
1728 }
1729 }
1730
1731 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1732 // so dll_address_to_library_name() can fall through to use dladdr() which
1733 // can figure out executable name from argv[0].
1734 if (found && info->dlpi_name && info->dlpi_name[0]) {
1735 d->base = libbase;
1736 if (d->fname) {
1737 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1738 }
1739 return 1;
1740 }
1741 return 0;
1742 }
1743
1744 bool os::dll_address_to_library_name(address addr, char* buf,
1745 int buflen, int* offset) {
1746 // buf is not optional, but offset is optional
1747 assert(buf != NULL, "sanity check");
1748
1749 Dl_info dlinfo;
1750 struct _address_to_library_name data;
1751
1752 // There is a bug in old glibc dladdr() implementation that it could resolve
1753 // to wrong library name if the .so file has a base address != NULL. Here
1754 // we iterate through the program headers of all loaded libraries to find
1755 // out which library 'addr' really belongs to. This workaround can be
1756 // removed once the minimum requirement for glibc is moved to 2.3.x.
1757 data.addr = addr;
1758 data.fname = buf;
1759 data.buflen = buflen;
1760 data.base = NULL;
1761 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1762
1763 if (rslt) {
1764 // buf already contains library name
1765 if (offset) *offset = addr - data.base;
1766 return true;
1767 }
1768 if (dladdr((void*)addr, &dlinfo) != 0) {
1769 if (dlinfo.dli_fname != NULL) {
1770 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1771 }
1772 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1773 *offset = addr - (address)dlinfo.dli_fbase;
1774 }
1775 return true;
1776 }
1777
1778 buf[0] = '\0';
1779 if (offset) *offset = -1;
1780 return false;
1781 }
1782
1783 // Loads .dll/.so and
1784 // in case of error it checks if .dll/.so was built for the
1785 // same architecture as Hotspot is running on
1786
1787
1788 // Remember the stack's state. The Linux dynamic linker will change
1789 // the stack to 'executable' at most once, so we must safepoint only once.
1790 bool os::Linux::_stack_is_executable = false;
1791
1792 // VM operation that loads a library. This is necessary if stack protection
1793 // of the Java stacks can be lost during loading the library. If we
1794 // do not stop the Java threads, they can stack overflow before the stacks
1795 // are protected again.
1796 class VM_LinuxDllLoad: public VM_Operation {
1797 private:
1798 const char *_filename;
1799 char *_ebuf;
1800 int _ebuflen;
1801 void *_lib;
1802 public:
1803 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1804 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1805 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1806 void doit() {
1807 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1808 os::Linux::_stack_is_executable = true;
1809 }
1810 void* loaded_library() { return _lib; }
1811 };
1812
1813 void * os::dll_load(const char *filename, char *ebuf, int ebuflen)
1814 {
1815 void * result = NULL;
1816 bool load_attempted = false;
1817
1818 // Check whether the library to load might change execution rights
1819 // of the stack. If they are changed, the protection of the stack
1820 // guard pages will be lost. We need a safepoint to fix this.
1821 //
1822 // See Linux man page execstack(8) for more info.
1823 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1824 ElfFile ef(filename);
1825 if (!ef.specifies_noexecstack()) {
1826 if (!is_init_completed()) {
1827 os::Linux::_stack_is_executable = true;
1828 // This is OK - No Java threads have been created yet, and hence no
1829 // stack guard pages to fix.
1830 //
1831 // This should happen only when you are building JDK7 using a very
1832 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1833 //
1834 // Dynamic loader will make all stacks executable after
1835 // this function returns, and will not do that again.
1836 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1837 } else {
1838 warning("You have loaded library %s which might have disabled stack guard. "
1839 "The VM will try to fix the stack guard now.\n"
1840 "It's highly recommended that you fix the library with "
1841 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1842 filename);
1843
1844 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1845 JavaThread *jt = JavaThread::current();
1846 if (jt->thread_state() != _thread_in_native) {
1847 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1848 // that requires ExecStack. Cannot enter safe point. Let's give up.
1849 warning("Unable to fix stack guard. Giving up.");
1850 } else {
1851 if (!LoadExecStackDllInVMThread) {
1852 // This is for the case where the DLL has an static
1853 // constructor function that executes JNI code. We cannot
1854 // load such DLLs in the VMThread.
1855 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1856 }
1857
1858 ThreadInVMfromNative tiv(jt);
1859 debug_only(VMNativeEntryWrapper vew;)
1860
1861 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1862 VMThread::execute(&op);
1863 if (LoadExecStackDllInVMThread) {
1864 result = op.loaded_library();
1865 }
1866 load_attempted = true;
1867 }
1868 }
1869 }
1870 }
1871
1872 if (!load_attempted) {
1873 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1874 }
1875
1876 if (result != NULL) {
1877 // Successful loading
1878 return result;
1879 }
1880
1881 Elf32_Ehdr elf_head;
1882 int diag_msg_max_length=ebuflen-strlen(ebuf);
1883 char* diag_msg_buf=ebuf+strlen(ebuf);
1884
1885 if (diag_msg_max_length==0) {
1886 // No more space in ebuf for additional diagnostics message
1887 return NULL;
1888 }
1889
1890
1891 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1892
1893 if (file_descriptor < 0) {
1894 // Can't open library, report dlerror() message
1895 return NULL;
1896 }
1897
1898 bool failed_to_read_elf_head=
1899 (sizeof(elf_head)!=
1900 (::read(file_descriptor, &elf_head,sizeof(elf_head)))) ;
1901
1902 ::close(file_descriptor);
1903 if (failed_to_read_elf_head) {
1904 // file i/o error - report dlerror() msg
1905 return NULL;
1906 }
1907
1908 typedef struct {
1909 Elf32_Half code; // Actual value as defined in elf.h
1910 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1911 char elf_class; // 32 or 64 bit
1912 char endianess; // MSB or LSB
1913 char* name; // String representation
1914 } arch_t;
1915
1916 #ifndef EM_486
1917 #define EM_486 6 /* Intel 80486 */
1918 #endif
1919
1920 static const arch_t arch_array[]={
1921 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1922 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1923 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1924 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1925 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1926 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1927 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1928 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1929 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1930 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1931 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1932 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1933 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1934 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1935 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1936 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"}
1937 };
1938
1939 #if (defined IA32)
1940 static Elf32_Half running_arch_code=EM_386;
1941 #elif (defined AMD64)
1942 static Elf32_Half running_arch_code=EM_X86_64;
1943 #elif (defined IA64)
1944 static Elf32_Half running_arch_code=EM_IA_64;
1945 #elif (defined __sparc) && (defined _LP64)
1946 static Elf32_Half running_arch_code=EM_SPARCV9;
1947 #elif (defined __sparc) && (!defined _LP64)
1948 static Elf32_Half running_arch_code=EM_SPARC;
1949 #elif (defined __powerpc64__)
1950 static Elf32_Half running_arch_code=EM_PPC64;
1951 #elif (defined __powerpc__)
1952 static Elf32_Half running_arch_code=EM_PPC;
1953 #elif (defined ARM)
1954 static Elf32_Half running_arch_code=EM_ARM;
1955 #elif (defined S390)
1956 static Elf32_Half running_arch_code=EM_S390;
1957 #elif (defined ALPHA)
1958 static Elf32_Half running_arch_code=EM_ALPHA;
1959 #elif (defined MIPSEL)
1960 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1961 #elif (defined PARISC)
1962 static Elf32_Half running_arch_code=EM_PARISC;
1963 #elif (defined MIPS)
1964 static Elf32_Half running_arch_code=EM_MIPS;
1965 #elif (defined M68K)
1966 static Elf32_Half running_arch_code=EM_68K;
1967 #else
1968 #error Method os::dll_load requires that one of following is defined:\
1969 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K
1970 #endif
1971
1972 // Identify compatability class for VM's architecture and library's architecture
1973 // Obtain string descriptions for architectures
1974
1975 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1976 int running_arch_index=-1;
1977
1978 for (unsigned int i=0 ; i < ARRAY_SIZE(arch_array) ; i++ ) {
1979 if (running_arch_code == arch_array[i].code) {
1980 running_arch_index = i;
1981 }
1982 if (lib_arch.code == arch_array[i].code) {
1983 lib_arch.compat_class = arch_array[i].compat_class;
1984 lib_arch.name = arch_array[i].name;
1985 }
1986 }
1987
1988 assert(running_arch_index != -1,
1989 "Didn't find running architecture code (running_arch_code) in arch_array");
1990 if (running_arch_index == -1) {
1991 // Even though running architecture detection failed
1992 // we may still continue with reporting dlerror() message
1993 return NULL;
1994 }
1995
1996 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1997 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1998 return NULL;
1999 }
2000
2001 #ifndef S390
2002 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
2003 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
2004 return NULL;
2005 }
2006 #endif // !S390
2007
2008 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
2009 if ( lib_arch.name!=NULL ) {
2010 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2011 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
2012 lib_arch.name, arch_array[running_arch_index].name);
2013 } else {
2014 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2015 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2016 lib_arch.code,
2017 arch_array[running_arch_index].name);
2018 }
2019 }
2020
2021 return NULL;
2022 }
2023
2024 void * os::Linux::dlopen_helper(const char *filename, char *ebuf, int ebuflen) {
2025 void * result = ::dlopen(filename, RTLD_LAZY);
2026 if (result == NULL) {
2027 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2028 ebuf[ebuflen-1] = '\0';
2029 }
2030 return result;
2031 }
2032
2033 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf, int ebuflen) {
2034 void * result = NULL;
2035 if (LoadExecStackDllInVMThread) {
2036 result = dlopen_helper(filename, ebuf, ebuflen);
2037 }
2038
2039 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2040 // library that requires an executable stack, or which does not have this
2041 // stack attribute set, dlopen changes the stack attribute to executable. The
2042 // read protection of the guard pages gets lost.
2043 //
2044 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2045 // may have been queued at the same time.
2046
2047 if (!_stack_is_executable) {
2048 JavaThread *jt = Threads::first();
2049
2050 while (jt) {
2051 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2052 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2053 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2054 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2055 warning("Attempt to reguard stack yellow zone failed.");
2056 }
2057 }
2058 jt = jt->next();
2059 }
2060 }
2061
2062 return result;
2063 }
2064
2065 /*
2066 * glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2067 * chances are you might want to run the generated bits against glibc-2.0
2068 * libdl.so, so always use locking for any version of glibc.
2069 */
2070 void* os::dll_lookup(void* handle, const char* name) {
2071 pthread_mutex_lock(&dl_mutex);
2072 void* res = dlsym(handle, name);
2073 pthread_mutex_unlock(&dl_mutex);
2074 return res;
2075 }
2076
2077 void* os::get_default_process_handle() {
2078 return (void*)::dlopen(NULL, RTLD_LAZY);
2079 }
2080
2081 static bool _print_ascii_file(const char* filename, outputStream* st) {
2082 int fd = ::open(filename, O_RDONLY);
2083 if (fd == -1) {
2084 return false;
2085 }
2086
2087 char buf[32];
2088 int bytes;
2089 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2090 st->print_raw(buf, bytes);
2091 }
2092
2093 ::close(fd);
2094
2095 return true;
2096 }
2097
2098 void os::print_dll_info(outputStream *st) {
2099 st->print_cr("Dynamic libraries:");
2100
2101 char fname[32];
2102 pid_t pid = os::Linux::gettid();
2103
2104 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2105
2106 if (!_print_ascii_file(fname, st)) {
2107 st->print("Can not get library information for pid = %d\n", pid);
2108 }
2109 }
2110
2111 void os::print_os_info_brief(outputStream* st) {
2112 os::Linux::print_distro_info(st);
2113
2114 os::Posix::print_uname_info(st);
2115
2116 os::Linux::print_libversion_info(st);
2117
2118 }
2119
2120 void os::print_os_info(outputStream* st) {
2121 st->print("OS:");
2122
2123 os::Linux::print_distro_info(st);
2124
2125 os::Posix::print_uname_info(st);
2126
2127 // Print warning if unsafe chroot environment detected
2128 if (unsafe_chroot_detected) {
2129 st->print("WARNING!! ");
2130 st->print_cr(unstable_chroot_error);
2131 }
2132
2133 os::Linux::print_libversion_info(st);
2134
2135 os::Posix::print_rlimit_info(st);
2136
2137 os::Posix::print_load_average(st);
2138
2139 os::Linux::print_full_memory_info(st);
2140 }
2141
2142 // Try to identify popular distros.
2143 // Most Linux distributions have a /etc/XXX-release file, which contains
2144 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2145 // file that also contains the OS version string. Some have more than one
2146 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2147 // /etc/redhat-release.), so the order is important.
2148 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2149 // their own specific XXX-release file as well as a redhat-release file.
2150 // Because of this the XXX-release file needs to be searched for before the
2151 // redhat-release file.
2152 // Since Red Hat has a lsb-release file that is not very descriptive the
2153 // search for redhat-release needs to be before lsb-release.
2154 // Since the lsb-release file is the new standard it needs to be searched
2155 // before the older style release files.
2156 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2157 // next to last resort. The os-release file is a new standard that contains
2158 // distribution information and the system-release file seems to be an old
2159 // standard that has been replaced by the lsb-release and os-release files.
2160 // Searching for the debian_version file is the last resort. It contains
2161 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2162 // "Debian " is printed before the contents of the debian_version file.
2163 void os::Linux::print_distro_info(outputStream* st) {
2164 if (!_print_ascii_file("/etc/oracle-release", st) &&
2165 !_print_ascii_file("/etc/mandriva-release", st) &&
2166 !_print_ascii_file("/etc/mandrake-release", st) &&
2167 !_print_ascii_file("/etc/sun-release", st) &&
2168 !_print_ascii_file("/etc/redhat-release", st) &&
2169 !_print_ascii_file("/etc/lsb-release", st) &&
2170 !_print_ascii_file("/etc/SuSE-release", st) &&
2171 !_print_ascii_file("/etc/turbolinux-release", st) &&
2172 !_print_ascii_file("/etc/gentoo-release", st) &&
2173 !_print_ascii_file("/etc/ltib-release", st) &&
2174 !_print_ascii_file("/etc/angstrom-version", st) &&
2175 !_print_ascii_file("/etc/system-release", st) &&
2176 !_print_ascii_file("/etc/os-release", st)) {
2177
2178 if (file_exists("/etc/debian_version")) {
2179 st->print("Debian ");
2180 _print_ascii_file("/etc/debian_version", st);
2181 } else {
2182 st->print("Linux");
2183 }
2184 }
2185 st->cr();
2186 }
2187
2188 void os::Linux::print_libversion_info(outputStream* st) {
2189 // libc, pthread
2190 st->print("libc:");
2191 st->print(os::Linux::glibc_version()); st->print(" ");
2192 st->print(os::Linux::libpthread_version()); st->print(" ");
2193 if (os::Linux::is_LinuxThreads()) {
2194 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2195 }
2196 st->cr();
2197 }
2198
2199 void os::Linux::print_full_memory_info(outputStream* st) {
2200 st->print("\n/proc/meminfo:\n");
2201 _print_ascii_file("/proc/meminfo", st);
2202 st->cr();
2203 }
2204
2205 void os::print_memory_info(outputStream* st) {
2206
2207 st->print("Memory:");
2208 st->print(" %dk page", os::vm_page_size()>>10);
2209
2210 // values in struct sysinfo are "unsigned long"
2211 struct sysinfo si;
2212 sysinfo(&si);
2213
2214 st->print(", physical " UINT64_FORMAT "k",
2215 os::physical_memory() >> 10);
2216 st->print("(" UINT64_FORMAT "k free)",
2217 os::available_memory() >> 10);
2218 st->print(", swap " UINT64_FORMAT "k",
2219 ((jlong)si.totalswap * si.mem_unit) >> 10);
2220 st->print("(" UINT64_FORMAT "k free)",
2221 ((jlong)si.freeswap * si.mem_unit) >> 10);
2222 st->cr();
2223 }
2224
2225 void os::pd_print_cpu_info(outputStream* st) {
2226 st->print("\n/proc/cpuinfo:\n");
2227 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2228 st->print(" <Not Available>");
2229 }
2230 st->cr();
2231 }
2232
2233 void os::print_siginfo(outputStream* st, void* siginfo) {
2234 const siginfo_t* si = (const siginfo_t*)siginfo;
2235
2236 os::Posix::print_siginfo_brief(st, si);
2237
2238 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2239 UseSharedSpaces) {
2240 FileMapInfo* mapinfo = FileMapInfo::current_info();
2241 if (mapinfo->is_in_shared_space(si->si_addr)) {
2242 st->print("\n\nError accessing class data sharing archive." \
2243 " Mapped file inaccessible during execution, " \
2244 " possible disk/network problem.");
2245 }
2246 }
2247 st->cr();
2248 }
2249
2250
2251 static void print_signal_handler(outputStream* st, int sig,
2252 char* buf, size_t buflen);
2253
2254 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2255 st->print_cr("Signal Handlers:");
2256 print_signal_handler(st, SIGSEGV, buf, buflen);
2257 print_signal_handler(st, SIGBUS , buf, buflen);
2258 print_signal_handler(st, SIGFPE , buf, buflen);
2259 print_signal_handler(st, SIGPIPE, buf, buflen);
2260 print_signal_handler(st, SIGXFSZ, buf, buflen);
2261 print_signal_handler(st, SIGILL , buf, buflen);
2262 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2263 print_signal_handler(st, SR_signum, buf, buflen);
2264 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2265 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2266 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2267 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2268 #if defined(PPC64)
2269 print_signal_handler(st, SIGTRAP, buf, buflen);
2270 #endif
2271 }
2272
2273 static char saved_jvm_path[MAXPATHLEN] = {0};
2274
2275 // Find the full path to the current module, libjvm.so
2276 void os::jvm_path(char *buf, jint buflen) {
2277 // Error checking.
2278 if (buflen < MAXPATHLEN) {
2279 assert(false, "must use a large-enough buffer");
2280 buf[0] = '\0';
2281 return;
2282 }
2283 // Lazy resolve the path to current module.
2284 if (saved_jvm_path[0] != 0) {
2285 strcpy(buf, saved_jvm_path);
2286 return;
2287 }
2288
2289 char dli_fname[MAXPATHLEN];
2290 bool ret = dll_address_to_library_name(
2291 CAST_FROM_FN_PTR(address, os::jvm_path),
2292 dli_fname, sizeof(dli_fname), NULL);
2293 assert(ret, "cannot locate libjvm");
2294 char *rp = NULL;
2295 if (ret && dli_fname[0] != '\0') {
2296 rp = realpath(dli_fname, buf);
2297 }
2298 if (rp == NULL)
2299 return;
2300
2301 if (Arguments::sun_java_launcher_is_altjvm()) {
2302 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2303 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2304 // If "/jre/lib/" appears at the right place in the string, then
2305 // assume we are installed in a JDK and we're done. Otherwise, check
2306 // for a JAVA_HOME environment variable and fix up the path so it
2307 // looks like libjvm.so is installed there (append a fake suffix
2308 // hotspot/libjvm.so).
2309 const char *p = buf + strlen(buf) - 1;
2310 for (int count = 0; p > buf && count < 5; ++count) {
2311 for (--p; p > buf && *p != '/'; --p)
2312 /* empty */ ;
2313 }
2314
2315 if (strncmp(p, "/jre/lib/", 9) != 0) {
2316 // Look for JAVA_HOME in the environment.
2317 char* java_home_var = ::getenv("JAVA_HOME");
2318 if (java_home_var != NULL && java_home_var[0] != 0) {
2319 char* jrelib_p;
2320 int len;
2321
2322 // Check the current module name "libjvm.so".
2323 p = strrchr(buf, '/');
2324 assert(strstr(p, "/libjvm") == p, "invalid library name");
2325
2326 rp = realpath(java_home_var, buf);
2327 if (rp == NULL)
2328 return;
2329
2330 // determine if this is a legacy image or modules image
2331 // modules image doesn't have "jre" subdirectory
2332 len = strlen(buf);
2333 jrelib_p = buf + len;
2334 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2335 if (0 != access(buf, F_OK)) {
2336 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2337 }
2338
2339 if (0 == access(buf, F_OK)) {
2340 // Use current module name "libjvm.so"
2341 len = strlen(buf);
2342 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2343 } else {
2344 // Go back to path of .so
2345 rp = realpath(dli_fname, buf);
2346 if (rp == NULL)
2347 return;
2348 }
2349 }
2350 }
2351 }
2352
2353 strcpy(saved_jvm_path, buf);
2354 }
2355
2356 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2357 // no prefix required, not even "_"
2358 }
2359
2360 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2361 // no suffix required
2362 }
2363
2364 ////////////////////////////////////////////////////////////////////////////////
2365 // sun.misc.Signal support
2366
2367 static volatile jint sigint_count = 0;
2368
2369 static void
2370 UserHandler(int sig, void *siginfo, void *context) {
2371 // 4511530 - sem_post is serialized and handled by the manager thread. When
2372 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2373 // don't want to flood the manager thread with sem_post requests.
2374 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1)
2375 return;
2376
2377 // Ctrl-C is pressed during error reporting, likely because the error
2378 // handler fails to abort. Let VM die immediately.
2379 if (sig == SIGINT && is_error_reported()) {
2380 os::die();
2381 }
2382
2383 os::signal_notify(sig);
2384 }
2385
2386 void* os::user_handler() {
2387 return CAST_FROM_FN_PTR(void*, UserHandler);
2388 }
2389
2390 class Semaphore : public StackObj {
2391 public:
2392 Semaphore();
2393 ~Semaphore();
2394 void signal();
2395 void wait();
2396 bool trywait();
2397 bool timedwait(unsigned int sec, int nsec);
2398 private:
2399 sem_t _semaphore;
2400 };
2401
2402
2403 Semaphore::Semaphore() {
2404 sem_init(&_semaphore, 0, 0);
2405 }
2406
2407 Semaphore::~Semaphore() {
2408 sem_destroy(&_semaphore);
2409 }
2410
2411 void Semaphore::signal() {
2412 sem_post(&_semaphore);
2413 }
2414
2415 void Semaphore::wait() {
2416 sem_wait(&_semaphore);
2417 }
2418
2419 bool Semaphore::trywait() {
2420 return sem_trywait(&_semaphore) == 0;
2421 }
2422
2423 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2424 struct timespec ts;
2425 unpackTime(&ts, false, (sec * NANOSECS_PER_SEC) + nsec);
2426
2427 while (1) {
2428 int result = sem_timedwait(&_semaphore, &ts);
2429 if (result == 0) {
2430 return true;
2431 } else if (errno == EINTR) {
2432 continue;
2433 } else if (errno == ETIMEDOUT) {
2434 return false;
2435 } else {
2436 return false;
2437 }
2438 }
2439 }
2440
2441 extern "C" {
2442 typedef void (*sa_handler_t)(int);
2443 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2444 }
2445
2446 void* os::signal(int signal_number, void* handler) {
2447 struct sigaction sigAct, oldSigAct;
2448
2449 sigfillset(&(sigAct.sa_mask));
2450 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2451 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2452
2453 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2454 // -1 means registration failed
2455 return (void *)-1;
2456 }
2457
2458 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2459 }
2460
2461 void os::signal_raise(int signal_number) {
2462 ::raise(signal_number);
2463 }
2464
2465 /*
2466 * The following code is moved from os.cpp for making this
2467 * code platform specific, which it is by its very nature.
2468 */
2469
2470 // Will be modified when max signal is changed to be dynamic
2471 int os::sigexitnum_pd() {
2472 return NSIG;
2473 }
2474
2475 // a counter for each possible signal value
2476 static volatile jint pending_signals[NSIG+1] = { 0 };
2477
2478 // Linux(POSIX) specific hand shaking semaphore.
2479 static sem_t sig_sem;
2480 static Semaphore sr_semaphore;
2481
2482 void os::signal_init_pd() {
2483 // Initialize signal structures
2484 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2485
2486 // Initialize signal semaphore
2487 ::sem_init(&sig_sem, 0, 0);
2488 }
2489
2490 void os::signal_notify(int sig) {
2491 Atomic::inc(&pending_signals[sig]);
2492 ::sem_post(&sig_sem);
2493 }
2494
2495 static int check_pending_signals(bool wait) {
2496 Atomic::store(0, &sigint_count);
2497 for (;;) {
2498 for (int i = 0; i < NSIG + 1; i++) {
2499 jint n = pending_signals[i];
2500 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2501 return i;
2502 }
2503 }
2504 if (!wait) {
2505 return -1;
2506 }
2507 JavaThread *thread = JavaThread::current();
2508 ThreadBlockInVM tbivm(thread);
2509
2510 bool threadIsSuspended;
2511 do {
2512 thread->set_suspend_equivalent();
2513 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2514 ::sem_wait(&sig_sem);
2515
2516 // were we externally suspended while we were waiting?
2517 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2518 if (threadIsSuspended) {
2519 //
2520 // The semaphore has been incremented, but while we were waiting
2521 // another thread suspended us. We don't want to continue running
2522 // while suspended because that would surprise the thread that
2523 // suspended us.
2524 //
2525 ::sem_post(&sig_sem);
2526
2527 thread->java_suspend_self();
2528 }
2529 } while (threadIsSuspended);
2530 }
2531 }
2532
2533 int os::signal_lookup() {
2534 return check_pending_signals(false);
2535 }
2536
2537 int os::signal_wait() {
2538 return check_pending_signals(true);
2539 }
2540
2541 ////////////////////////////////////////////////////////////////////////////////
2542 // Virtual Memory
2543
2544 int os::vm_page_size() {
2545 // Seems redundant as all get out
2546 assert(os::Linux::page_size() != -1, "must call os::init");
2547 return os::Linux::page_size();
2548 }
2549
2550 // Solaris allocates memory by pages.
2551 int os::vm_allocation_granularity() {
2552 assert(os::Linux::page_size() != -1, "must call os::init");
2553 return os::Linux::page_size();
2554 }
2555
2556 // Rationale behind this function:
2557 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2558 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2559 // samples for JITted code. Here we create private executable mapping over the code cache
2560 // and then we can use standard (well, almost, as mapping can change) way to provide
2561 // info for the reporting script by storing timestamp and location of symbol
2562 void linux_wrap_code(char* base, size_t size) {
2563 static volatile jint cnt = 0;
2564
2565 if (!UseOprofile) {
2566 return;
2567 }
2568
2569 char buf[PATH_MAX+1];
2570 int num = Atomic::add(1, &cnt);
2571
2572 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2573 os::get_temp_directory(), os::current_process_id(), num);
2574 unlink(buf);
2575
2576 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2577
2578 if (fd != -1) {
2579 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2580 if (rv != (off_t)-1) {
2581 if (::write(fd, "", 1) == 1) {
2582 mmap(base, size,
2583 PROT_READ|PROT_WRITE|PROT_EXEC,
2584 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2585 }
2586 }
2587 ::close(fd);
2588 unlink(buf);
2589 }
2590 }
2591
2592 static bool recoverable_mmap_error(int err) {
2593 // See if the error is one we can let the caller handle. This
2594 // list of errno values comes from JBS-6843484. I can't find a
2595 // Linux man page that documents this specific set of errno
2596 // values so while this list currently matches Solaris, it may
2597 // change as we gain experience with this failure mode.
2598 switch (err) {
2599 case EBADF:
2600 case EINVAL:
2601 case ENOTSUP:
2602 // let the caller deal with these errors
2603 return true;
2604
2605 default:
2606 // Any remaining errors on this OS can cause our reserved mapping
2607 // to be lost. That can cause confusion where different data
2608 // structures think they have the same memory mapped. The worst
2609 // scenario is if both the VM and a library think they have the
2610 // same memory mapped.
2611 return false;
2612 }
2613 }
2614
2615 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2616 int err) {
2617 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2618 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2619 strerror(err), err);
2620 }
2621
2622 static void warn_fail_commit_memory(char* addr, size_t size,
2623 size_t alignment_hint, bool exec,
2624 int err) {
2625 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2626 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2627 alignment_hint, exec, strerror(err), err);
2628 }
2629
2630 // NOTE: Linux kernel does not really reserve the pages for us.
2631 // All it does is to check if there are enough free pages
2632 // left at the time of mmap(). This could be a potential
2633 // problem.
2634 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2635 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2636 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2637 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2638 if (res != (uintptr_t) MAP_FAILED) {
2639 if (UseNUMAInterleaving) {
2640 numa_make_global(addr, size);
2641 }
2642 return 0;
2643 }
2644
2645 int err = errno; // save errno from mmap() call above
2646
2647 if (!recoverable_mmap_error(err)) {
2648 warn_fail_commit_memory(addr, size, exec, err);
2649 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2650 }
2651
2652 return err;
2653 }
2654
2655 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2656 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2657 }
2658
2659 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2660 const char* mesg) {
2661 assert(mesg != NULL, "mesg must be specified");
2662 int err = os::Linux::commit_memory_impl(addr, size, exec);
2663 if (err != 0) {
2664 // the caller wants all commit errors to exit with the specified mesg:
2665 warn_fail_commit_memory(addr, size, exec, err);
2666 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2667 }
2668 }
2669
2670 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2671 #ifndef MAP_HUGETLB
2672 #define MAP_HUGETLB 0x40000
2673 #endif
2674
2675 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2676 #ifndef MADV_HUGEPAGE
2677 #define MADV_HUGEPAGE 14
2678 #endif
2679
2680 int os::Linux::commit_memory_impl(char* addr, size_t size,
2681 size_t alignment_hint, bool exec) {
2682 int err = os::Linux::commit_memory_impl(addr, size, exec);
2683 if (err == 0) {
2684 realign_memory(addr, size, alignment_hint);
2685 }
2686 return err;
2687 }
2688
2689 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2690 bool exec) {
2691 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2692 }
2693
2694 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2695 size_t alignment_hint, bool exec,
2696 const char* mesg) {
2697 assert(mesg != NULL, "mesg must be specified");
2698 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2699 if (err != 0) {
2700 // the caller wants all commit errors to exit with the specified mesg:
2701 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2702 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2703 }
2704 }
2705
2706 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2707 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2708 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2709 // be supported or the memory may already be backed by huge pages.
2710 ::madvise(addr, bytes, MADV_HUGEPAGE);
2711 }
2712 }
2713
2714 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2715 // This method works by doing an mmap over an existing mmaping and effectively discarding
2716 // the existing pages. However it won't work for SHM-based large pages that cannot be
2717 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2718 // small pages on top of the SHM segment. This method always works for small pages, so we
2719 // allow that in any case.
2720 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2721 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2722 }
2723 }
2724
2725 void os::numa_make_global(char *addr, size_t bytes) {
2726 Linux::numa_interleave_memory(addr, bytes);
2727 }
2728
2729 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2730 // bind policy to MPOL_PREFERRED for the current thread.
2731 #define USE_MPOL_PREFERRED 0
2732
2733 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2734 // To make NUMA and large pages more robust when both enabled, we need to ease
2735 // the requirements on where the memory should be allocated. MPOL_BIND is the
2736 // default policy and it will force memory to be allocated on the specified
2737 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2738 // the specified node, but will not force it. Using this policy will prevent
2739 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2740 // free large pages.
2741 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2742 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2743 }
2744
2745 bool os::numa_topology_changed() { return false; }
2746
2747 size_t os::numa_get_groups_num() {
2748 int max_node = Linux::numa_max_node();
2749 return max_node > 0 ? max_node + 1 : 1;
2750 }
2751
2752 int os::numa_get_group_id() {
2753 int cpu_id = Linux::sched_getcpu();
2754 if (cpu_id != -1) {
2755 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2756 if (lgrp_id != -1) {
2757 return lgrp_id;
2758 }
2759 }
2760 return 0;
2761 }
2762
2763 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2764 for (size_t i = 0; i < size; i++) {
2765 ids[i] = i;
2766 }
2767 return size;
2768 }
2769
2770 bool os::get_page_info(char *start, page_info* info) {
2771 return false;
2772 }
2773
2774 char *os::scan_pages(char *start, char* end, page_info* page_expected, page_info* page_found) {
2775 return end;
2776 }
2777
2778
2779 int os::Linux::sched_getcpu_syscall(void) {
2780 unsigned int cpu;
2781 int retval = -1;
2782
2783 #if defined(IA32)
2784 # ifndef SYS_getcpu
2785 # define SYS_getcpu 318
2786 # endif
2787 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2788 #elif defined(AMD64)
2789 // Unfortunately we have to bring all these macros here from vsyscall.h
2790 // to be able to compile on old linuxes.
2791 # define __NR_vgetcpu 2
2792 # define VSYSCALL_START (-10UL << 20)
2793 # define VSYSCALL_SIZE 1024
2794 # define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2795 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2796 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2797 retval = vgetcpu(&cpu, NULL, NULL);
2798 #endif
2799
2800 return (retval == -1) ? retval : cpu;
2801 }
2802
2803 // Something to do with the numa-aware allocator needs these symbols
2804 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2805 extern "C" JNIEXPORT void numa_error(char *where) { }
2806 extern "C" JNIEXPORT int fork1() { return fork(); }
2807
2808
2809 // If we are running with libnuma version > 2, then we should
2810 // be trying to use symbols with versions 1.1
2811 // If we are running with earlier version, which did not have symbol versions,
2812 // we should use the base version.
2813 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2814 void *f = dlvsym(handle, name, "libnuma_1.1");
2815 if (f == NULL) {
2816 f = dlsym(handle, name);
2817 }
2818 return f;
2819 }
2820
2821 bool os::Linux::libnuma_init() {
2822 // sched_getcpu() should be in libc.
2823 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2824 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2825
2826 // If it's not, try a direct syscall.
2827 if (sched_getcpu() == -1)
2828 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t, (void*)&sched_getcpu_syscall));
2829
2830 if (sched_getcpu() != -1) { // Does it work?
2831 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2832 if (handle != NULL) {
2833 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2834 libnuma_dlsym(handle, "numa_node_to_cpus")));
2835 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2836 libnuma_dlsym(handle, "numa_max_node")));
2837 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2838 libnuma_dlsym(handle, "numa_available")));
2839 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2840 libnuma_dlsym(handle, "numa_tonode_memory")));
2841 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2842 libnuma_dlsym(handle, "numa_interleave_memory")));
2843 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2844 libnuma_dlsym(handle, "numa_set_bind_policy")));
2845
2846
2847 if (numa_available() != -1) {
2848 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2849 // Create a cpu -> node mapping
2850 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2851 rebuild_cpu_to_node_map();
2852 return true;
2853 }
2854 }
2855 }
2856 return false;
2857 }
2858
2859 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2860 // The table is later used in get_node_by_cpu().
2861 void os::Linux::rebuild_cpu_to_node_map() {
2862 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2863 // in libnuma (possible values are starting from 16,
2864 // and continuing up with every other power of 2, but less
2865 // than the maximum number of CPUs supported by kernel), and
2866 // is a subject to change (in libnuma version 2 the requirements
2867 // are more reasonable) we'll just hardcode the number they use
2868 // in the library.
2869 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2870
2871 size_t cpu_num = os::active_processor_count();
2872 size_t cpu_map_size = NCPUS / BitsPerCLong;
2873 size_t cpu_map_valid_size =
2874 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2875
2876 cpu_to_node()->clear();
2877 cpu_to_node()->at_grow(cpu_num - 1);
2878 size_t node_num = numa_get_groups_num();
2879
2880 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2881 for (size_t i = 0; i < node_num; i++) {
2882 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2883 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2884 if (cpu_map[j] != 0) {
2885 for (size_t k = 0; k < BitsPerCLong; k++) {
2886 if (cpu_map[j] & (1UL << k)) {
2887 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2888 }
2889 }
2890 }
2891 }
2892 }
2893 }
2894 FREE_C_HEAP_ARRAY(unsigned long, cpu_map, mtInternal);
2895 }
2896
2897 int os::Linux::get_node_by_cpu(int cpu_id) {
2898 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2899 return cpu_to_node()->at(cpu_id);
2900 }
2901 return -1;
2902 }
2903
2904 GrowableArray<int>* os::Linux::_cpu_to_node;
2905 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2906 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2907 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2908 os::Linux::numa_available_func_t os::Linux::_numa_available;
2909 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2910 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2911 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2912 unsigned long* os::Linux::_numa_all_nodes;
2913
2914 bool os::pd_uncommit_memory(char* addr, size_t size) {
2915 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2916 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2917 return res != (uintptr_t) MAP_FAILED;
2918 }
2919
2920 static
2921 address get_stack_commited_bottom(address bottom, size_t size) {
2922 address nbot = bottom;
2923 address ntop = bottom + size;
2924
2925 size_t page_sz = os::vm_page_size();
2926 unsigned pages = size / page_sz;
2927
2928 unsigned char vec[1];
2929 unsigned imin = 1, imax = pages + 1, imid;
2930 int mincore_return_value = 0;
2931
2932 assert(imin <= imax, "Unexpected page size");
2933
2934 while (imin < imax) {
2935 imid = (imax + imin) / 2;
2936 nbot = ntop - (imid * page_sz);
2937
2938 // Use a trick with mincore to check whether the page is mapped or not.
2939 // mincore sets vec to 1 if page resides in memory and to 0 if page
2940 // is swapped output but if page we are asking for is unmapped
2941 // it returns -1,ENOMEM
2942 mincore_return_value = mincore(nbot, page_sz, vec);
2943
2944 if (mincore_return_value == -1) {
2945 // Page is not mapped go up
2946 // to find first mapped page
2947 if (errno != EAGAIN) {
2948 assert(errno == ENOMEM, "Unexpected mincore errno");
2949 imax = imid;
2950 }
2951 } else {
2952 // Page is mapped go down
2953 // to find first not mapped page
2954 imin = imid + 1;
2955 }
2956 }
2957
2958 nbot = nbot + page_sz;
2959
2960 // Adjust stack bottom one page up if last checked page is not mapped
2961 if (mincore_return_value == -1) {
2962 nbot = nbot + page_sz;
2963 }
2964
2965 return nbot;
2966 }
2967
2968
2969 // Linux uses a growable mapping for the stack, and if the mapping for
2970 // the stack guard pages is not removed when we detach a thread the
2971 // stack cannot grow beyond the pages where the stack guard was
2972 // mapped. If at some point later in the process the stack expands to
2973 // that point, the Linux kernel cannot expand the stack any further
2974 // because the guard pages are in the way, and a segfault occurs.
2975 //
2976 // However, it's essential not to split the stack region by unmapping
2977 // a region (leaving a hole) that's already part of the stack mapping,
2978 // so if the stack mapping has already grown beyond the guard pages at
2979 // the time we create them, we have to truncate the stack mapping.
2980 // So, we need to know the extent of the stack mapping when
2981 // create_stack_guard_pages() is called.
2982
2983 // We only need this for stacks that are growable: at the time of
2984 // writing thread stacks don't use growable mappings (i.e. those
2985 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
2986 // only applies to the main thread.
2987
2988 // If the (growable) stack mapping already extends beyond the point
2989 // where we're going to put our guard pages, truncate the mapping at
2990 // that point by munmap()ping it. This ensures that when we later
2991 // munmap() the guard pages we don't leave a hole in the stack
2992 // mapping. This only affects the main/initial thread
2993
2994 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
2995
2996 if (os::Linux::is_initial_thread()) {
2997 // As we manually grow stack up to bottom inside create_attached_thread(),
2998 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
2999 // we don't need to do anything special.
3000 // Check it first, before calling heavy function.
3001 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3002 unsigned char vec[1];
3003
3004 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3005 // Fallback to slow path on all errors, including EAGAIN
3006 stack_extent = (uintptr_t) get_stack_commited_bottom(
3007 os::Linux::initial_thread_stack_bottom(),
3008 (size_t)addr - stack_extent);
3009 }
3010
3011 if (stack_extent < (uintptr_t)addr) {
3012 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3013 }
3014 }
3015
3016 return os::commit_memory(addr, size, !ExecMem);
3017 }
3018
3019 // If this is a growable mapping, remove the guard pages entirely by
3020 // munmap()ping them. If not, just call uncommit_memory(). This only
3021 // affects the main/initial thread, but guard against future OS changes
3022 // It's safe to always unmap guard pages for initial thread because we
3023 // always place it right after end of the mapped region
3024
3025 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3026 uintptr_t stack_extent, stack_base;
3027
3028 if (os::Linux::is_initial_thread()) {
3029 return ::munmap(addr, size) == 0;
3030 }
3031
3032 return os::uncommit_memory(addr, size);
3033 }
3034
3035 static address _highest_vm_reserved_address = NULL;
3036
3037 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3038 // at 'requested_addr'. If there are existing memory mappings at the same
3039 // location, however, they will be overwritten. If 'fixed' is false,
3040 // 'requested_addr' is only treated as a hint, the return value may or
3041 // may not start from the requested address. Unlike Linux mmap(), this
3042 // function returns NULL to indicate failure.
3043 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3044 char * addr;
3045 int flags;
3046
3047 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3048 if (fixed) {
3049 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3050 flags |= MAP_FIXED;
3051 }
3052
3053 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3054 // touch an uncommitted page. Otherwise, the read/write might
3055 // succeed if we have enough swap space to back the physical page.
3056 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3057 flags, -1, 0);
3058
3059 if (addr != MAP_FAILED) {
3060 // anon_mmap() should only get called during VM initialization,
3061 // don't need lock (actually we can skip locking even it can be called
3062 // from multiple threads, because _highest_vm_reserved_address is just a
3063 // hint about the upper limit of non-stack memory regions.)
3064 if ((address)addr + bytes > _highest_vm_reserved_address) {
3065 _highest_vm_reserved_address = (address)addr + bytes;
3066 }
3067 }
3068
3069 return addr == MAP_FAILED ? NULL : addr;
3070 }
3071
3072 // Don't update _highest_vm_reserved_address, because there might be memory
3073 // regions above addr + size. If so, releasing a memory region only creates
3074 // a hole in the address space, it doesn't help prevent heap-stack collision.
3075 //
3076 static int anon_munmap(char * addr, size_t size) {
3077 return ::munmap(addr, size) == 0;
3078 }
3079
3080 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3081 size_t alignment_hint) {
3082 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3083 }
3084
3085 bool os::pd_release_memory(char* addr, size_t size) {
3086 return anon_munmap(addr, size);
3087 }
3088
3089 static address highest_vm_reserved_address() {
3090 return _highest_vm_reserved_address;
3091 }
3092
3093 static bool linux_mprotect(char* addr, size_t size, int prot) {
3094 // Linux wants the mprotect address argument to be page aligned.
3095 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3096
3097 // According to SUSv3, mprotect() should only be used with mappings
3098 // established by mmap(), and mmap() always maps whole pages. Unaligned
3099 // 'addr' likely indicates problem in the VM (e.g. trying to change
3100 // protection of malloc'ed or statically allocated memory). Check the
3101 // caller if you hit this assert.
3102 assert(addr == bottom, "sanity check");
3103
3104 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3105 return ::mprotect(bottom, size, prot) == 0;
3106 }
3107
3108 // Set protections specified
3109 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3110 bool is_committed) {
3111 unsigned int p = 0;
3112 switch (prot) {
3113 case MEM_PROT_NONE: p = PROT_NONE; break;
3114 case MEM_PROT_READ: p = PROT_READ; break;
3115 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3116 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3117 default:
3118 ShouldNotReachHere();
3119 }
3120 // is_committed is unused.
3121 return linux_mprotect(addr, bytes, p);
3122 }
3123
3124 bool os::guard_memory(char* addr, size_t size) {
3125 return linux_mprotect(addr, size, PROT_NONE);
3126 }
3127
3128 bool os::unguard_memory(char* addr, size_t size) {
3129 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3130 }
3131
3132 bool os::Linux::transparent_huge_pages_sanity_check(bool warn, size_t page_size) {
3133 bool result = false;
3134 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3135 MAP_ANONYMOUS|MAP_PRIVATE,
3136 -1, 0);
3137 if (p != MAP_FAILED) {
3138 void *aligned_p = align_ptr_up(p, page_size);
3139
3140 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3141
3142 munmap(p, page_size * 2);
3143 }
3144
3145 if (warn && !result) {
3146 warning("TransparentHugePages is not supported by the operating system.");
3147 }
3148
3149 return result;
3150 }
3151
3152 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3153 bool result = false;
3154 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3155 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3156 -1, 0);
3157
3158 if (p != MAP_FAILED) {
3159 // We don't know if this really is a huge page or not.
3160 FILE *fp = fopen("/proc/self/maps", "r");
3161 if (fp) {
3162 while (!feof(fp)) {
3163 char chars[257];
3164 long x = 0;
3165 if (fgets(chars, sizeof(chars), fp)) {
3166 if (sscanf(chars, "%lx-%*x", &x) == 1
3167 && x == (long)p) {
3168 if (strstr (chars, "hugepage")) {
3169 result = true;
3170 break;
3171 }
3172 }
3173 }
3174 }
3175 fclose(fp);
3176 }
3177 munmap(p, page_size);
3178 }
3179
3180 if (warn && !result) {
3181 warning("HugeTLBFS is not supported by the operating system.");
3182 }
3183
3184 return result;
3185 }
3186
3187 /*
3188 * Set the coredump_filter bits to include largepages in core dump (bit 6)
3189 *
3190 * From the coredump_filter documentation:
3191 *
3192 * - (bit 0) anonymous private memory
3193 * - (bit 1) anonymous shared memory
3194 * - (bit 2) file-backed private memory
3195 * - (bit 3) file-backed shared memory
3196 * - (bit 4) ELF header pages in file-backed private memory areas (it is
3197 * effective only if the bit 2 is cleared)
3198 * - (bit 5) hugetlb private memory
3199 * - (bit 6) hugetlb shared memory
3200 */
3201 static void set_coredump_filter(void) {
3202 FILE *f;
3203 long cdm;
3204
3205 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3206 return;
3207 }
3208
3209 if (fscanf(f, "%lx", &cdm) != 1) {
3210 fclose(f);
3211 return;
3212 }
3213
3214 rewind(f);
3215
3216 if ((cdm & LARGEPAGES_BIT) == 0) {
3217 cdm |= LARGEPAGES_BIT;
3218 fprintf(f, "%#lx", cdm);
3219 }
3220
3221 fclose(f);
3222 }
3223
3224 // Large page support
3225
3226 static size_t _large_page_size = 0;
3227
3228 size_t os::Linux::find_large_page_size() {
3229 size_t large_page_size = 0;
3230
3231 // large_page_size on Linux is used to round up heap size. x86 uses either
3232 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3233 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3234 // page as large as 256M.
3235 //
3236 // Here we try to figure out page size by parsing /proc/meminfo and looking
3237 // for a line with the following format:
3238 // Hugepagesize: 2048 kB
3239 //
3240 // If we can't determine the value (e.g. /proc is not mounted, or the text
3241 // format has been changed), we'll use the largest page size supported by
3242 // the processor.
3243
3244 #ifndef ZERO
3245 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3246 ARM_ONLY(2 * M) PPC_ONLY(4 * M);
3247 #endif // ZERO
3248
3249 FILE *fp = fopen("/proc/meminfo", "r");
3250 if (fp) {
3251 while (!feof(fp)) {
3252 int x = 0;
3253 char buf[16];
3254 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3255 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3256 large_page_size = x * K;
3257 break;
3258 }
3259 } else {
3260 // skip to next line
3261 for (;;) {
3262 int ch = fgetc(fp);
3263 if (ch == EOF || ch == (int)'\n') break;
3264 }
3265 }
3266 }
3267 fclose(fp);
3268 }
3269
3270 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3271 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3272 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3273 proper_unit_for_byte_size(large_page_size));
3274 }
3275
3276 return large_page_size;
3277 }
3278
3279 size_t os::Linux::setup_large_page_size() {
3280 _large_page_size = Linux::find_large_page_size();
3281 const size_t default_page_size = (size_t)Linux::page_size();
3282 if (_large_page_size > default_page_size) {
3283 _page_sizes[0] = _large_page_size;
3284 _page_sizes[1] = default_page_size;
3285 _page_sizes[2] = 0;
3286 }
3287
3288 return _large_page_size;
3289 }
3290
3291 bool os::Linux::setup_large_page_type(size_t page_size) {
3292 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3293 FLAG_IS_DEFAULT(UseSHM) &&
3294 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3295
3296 // The type of large pages has not been specified by the user.
3297
3298 // Try UseHugeTLBFS and then UseSHM.
3299 UseHugeTLBFS = UseSHM = true;
3300
3301 // Don't try UseTransparentHugePages since there are known
3302 // performance issues with it turned on. This might change in the future.
3303 UseTransparentHugePages = false;
3304 }
3305
3306 if (UseTransparentHugePages) {
3307 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3308 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3309 UseHugeTLBFS = false;
3310 UseSHM = false;
3311 return true;
3312 }
3313 UseTransparentHugePages = false;
3314 }
3315
3316 if (UseHugeTLBFS) {
3317 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3318 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3319 UseSHM = false;
3320 return true;
3321 }
3322 UseHugeTLBFS = false;
3323 }
3324
3325 return UseSHM;
3326 }
3327
3328 void os::large_page_init() {
3329 if (!UseLargePages &&
3330 !UseTransparentHugePages &&
3331 !UseHugeTLBFS &&
3332 !UseSHM) {
3333 // Not using large pages.
3334 return;
3335 }
3336
3337 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3338 // The user explicitly turned off large pages.
3339 // Ignore the rest of the large pages flags.
3340 UseTransparentHugePages = false;
3341 UseHugeTLBFS = false;
3342 UseSHM = false;
3343 return;
3344 }
3345
3346 size_t large_page_size = Linux::setup_large_page_size();
3347 UseLargePages = Linux::setup_large_page_type(large_page_size);
3348
3349 set_coredump_filter();
3350 }
3351
3352 #ifndef SHM_HUGETLB
3353 #define SHM_HUGETLB 04000
3354 #endif
3355
3356 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3357 // "exec" is passed in but not used. Creating the shared image for
3358 // the code cache doesn't have an SHM_X executable permission to check.
3359 assert(UseLargePages && UseSHM, "only for SHM large pages");
3360 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3361
3362 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3363 return NULL; // Fallback to small pages.
3364 }
3365
3366 key_t key = IPC_PRIVATE;
3367 char *addr;
3368
3369 bool warn_on_failure = UseLargePages &&
3370 (!FLAG_IS_DEFAULT(UseLargePages) ||
3371 !FLAG_IS_DEFAULT(UseSHM) ||
3372 !FLAG_IS_DEFAULT(LargePageSizeInBytes)
3373 );
3374 char msg[128];
3375
3376 // Create a large shared memory region to attach to based on size.
3377 // Currently, size is the total size of the heap
3378 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3379 if (shmid == -1) {
3380 // Possible reasons for shmget failure:
3381 // 1. shmmax is too small for Java heap.
3382 // > check shmmax value: cat /proc/sys/kernel/shmmax
3383 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3384 // 2. not enough large page memory.
3385 // > check available large pages: cat /proc/meminfo
3386 // > increase amount of large pages:
3387 // echo new_value > /proc/sys/vm/nr_hugepages
3388 // Note 1: different Linux may use different name for this property,
3389 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3390 // Note 2: it's possible there's enough physical memory available but
3391 // they are so fragmented after a long run that they can't
3392 // coalesce into large pages. Try to reserve large pages when
3393 // the system is still "fresh".
3394 if (warn_on_failure) {
3395 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3396 warning(msg);
3397 }
3398 return NULL;
3399 }
3400
3401 // attach to the region
3402 addr = (char*)shmat(shmid, req_addr, 0);
3403 int err = errno;
3404
3405 // Remove shmid. If shmat() is successful, the actual shared memory segment
3406 // will be deleted when it's detached by shmdt() or when the process
3407 // terminates. If shmat() is not successful this will remove the shared
3408 // segment immediately.
3409 shmctl(shmid, IPC_RMID, NULL);
3410
3411 if ((intptr_t)addr == -1) {
3412 if (warn_on_failure) {
3413 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3414 warning(msg);
3415 }
3416 return NULL;
3417 }
3418
3419 return addr;
3420 }
3421
3422 static void warn_on_large_pages_failure(char* req_addr, size_t bytes, int error) {
3423 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3424
3425 bool warn_on_failure = UseLargePages &&
3426 (!FLAG_IS_DEFAULT(UseLargePages) ||
3427 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3428 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3429
3430 if (warn_on_failure) {
3431 char msg[128];
3432 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3433 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3434 warning(msg);
3435 }
3436 }
3437
3438 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes, char* req_addr, bool exec) {
3439 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3440 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3441 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3442
3443 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3444 char* addr = (char*)::mmap(req_addr, bytes, prot,
3445 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3446 -1, 0);
3447
3448 if (addr == MAP_FAILED) {
3449 warn_on_large_pages_failure(req_addr, bytes, errno);
3450 return NULL;
3451 }
3452
3453 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3454
3455 return addr;
3456 }
3457
3458 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3459 size_t large_page_size = os::large_page_size();
3460
3461 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3462
3463 // Allocate small pages.
3464
3465 char* start;
3466 if (req_addr != NULL) {
3467 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3468 assert(is_size_aligned(bytes, alignment), "Must be");
3469 start = os::reserve_memory(bytes, req_addr);
3470 assert(start == NULL || start == req_addr, "Must be");
3471 } else {
3472 start = os::reserve_memory_aligned(bytes, alignment);
3473 }
3474
3475 if (start == NULL) {
3476 return NULL;
3477 }
3478
3479 assert(is_ptr_aligned(start, alignment), "Must be");
3480
3481 // os::reserve_memory_special will record this memory area.
3482 // Need to release it here to prevent overlapping reservations.
3483 MemTracker::record_virtual_memory_release((address)start, bytes);
3484
3485 char* end = start + bytes;
3486
3487 // Find the regions of the allocated chunk that can be promoted to large pages.
3488 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3489 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3490
3491 size_t lp_bytes = lp_end - lp_start;
3492
3493 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3494
3495 if (lp_bytes == 0) {
3496 // The mapped region doesn't even span the start and the end of a large page.
3497 // Fall back to allocate a non-special area.
3498 ::munmap(start, end - start);
3499 return NULL;
3500 }
3501
3502 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3503
3504
3505 void* result;
3506
3507 if (start != lp_start) {
3508 result = ::mmap(start, lp_start - start, prot,
3509 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3510 -1, 0);
3511 if (result == MAP_FAILED) {
3512 ::munmap(lp_start, end - lp_start);
3513 return NULL;
3514 }
3515 }
3516
3517 result = ::mmap(lp_start, lp_bytes, prot,
3518 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3519 -1, 0);
3520 if (result == MAP_FAILED) {
3521 warn_on_large_pages_failure(req_addr, bytes, errno);
3522 // If the mmap above fails, the large pages region will be unmapped and we
3523 // have regions before and after with small pages. Release these regions.
3524 //
3525 // | mapped | unmapped | mapped |
3526 // ^ ^ ^ ^
3527 // start lp_start lp_end end
3528 //
3529 ::munmap(start, lp_start - start);
3530 ::munmap(lp_end, end - lp_end);
3531 return NULL;
3532 }
3533
3534 if (lp_end != end) {
3535 result = ::mmap(lp_end, end - lp_end, prot,
3536 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3537 -1, 0);
3538 if (result == MAP_FAILED) {
3539 ::munmap(start, lp_end - start);
3540 return NULL;
3541 }
3542 }
3543
3544 return start;
3545 }
3546
3547 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3548 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3549 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3550 assert(is_power_of_2(alignment), "Must be");
3551 assert(is_power_of_2(os::large_page_size()), "Must be");
3552 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3553
3554 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3555 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3556 } else {
3557 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3558 }
3559 }
3560
3561 char* os::reserve_memory_special(size_t bytes, size_t alignment, char* req_addr, bool exec) {
3562 assert(UseLargePages, "only for large pages");
3563
3564 char* addr;
3565 if (UseSHM) {
3566 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3567 } else {
3568 assert(UseHugeTLBFS, "must be");
3569 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3570 }
3571
3572 if (addr != NULL) {
3573 if (UseNUMAInterleaving) {
3574 numa_make_global(addr, bytes);
3575 }
3576
3577 // The memory is committed
3578 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, mtNone, CALLER_PC);
3579 }
3580
3581 return addr;
3582 }
3583
3584 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3585 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3586 return shmdt(base) == 0;
3587 }
3588
3589 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3590 return pd_release_memory(base, bytes);
3591 }
3592
3593 bool os::release_memory_special(char* base, size_t bytes) {
3594 assert(UseLargePages, "only for large pages");
3595
3596 MemTracker::Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3597
3598 bool res;
3599 if (UseSHM) {
3600 res = os::Linux::release_memory_special_shm(base, bytes);
3601 } else {
3602 assert(UseHugeTLBFS, "must be");
3603 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3604 }
3605
3606 if (res) {
3607 tkr.record((address)base, bytes);
3608 } else {
3609 tkr.discard();
3610 }
3611
3612 return res;
3613 }
3614
3615 size_t os::large_page_size() {
3616 return _large_page_size;
3617 }
3618
3619 // With SysV SHM the entire memory region must be allocated as shared
3620 // memory.
3621 // HugeTLBFS allows application to commit large page memory on demand.
3622 // However, when committing memory with HugeTLBFS fails, the region
3623 // that was supposed to be committed will lose the old reservation
3624 // and allow other threads to steal that memory region. Because of this
3625 // behavior we can't commit HugeTLBFS memory.
3626 bool os::can_commit_large_page_memory() {
3627 return UseTransparentHugePages;
3628 }
3629
3630 bool os::can_execute_large_page_memory() {
3631 return UseTransparentHugePages || UseHugeTLBFS;
3632 }
3633
3634 // Reserve memory at an arbitrary address, only if that area is
3635 // available (and not reserved for something else).
3636
3637 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3638 const int max_tries = 10;
3639 char* base[max_tries];
3640 size_t size[max_tries];
3641 const size_t gap = 0x000000;
3642
3643 // Assert only that the size is a multiple of the page size, since
3644 // that's all that mmap requires, and since that's all we really know
3645 // about at this low abstraction level. If we need higher alignment,
3646 // we can either pass an alignment to this method or verify alignment
3647 // in one of the methods further up the call chain. See bug 5044738.
3648 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3649
3650 // Repeatedly allocate blocks until the block is allocated at the
3651 // right spot. Give up after max_tries. Note that reserve_memory() will
3652 // automatically update _highest_vm_reserved_address if the call is
3653 // successful. The variable tracks the highest memory address every reserved
3654 // by JVM. It is used to detect heap-stack collision if running with
3655 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3656 // space than needed, it could confuse the collision detecting code. To
3657 // solve the problem, save current _highest_vm_reserved_address and
3658 // calculate the correct value before return.
3659 address old_highest = _highest_vm_reserved_address;
3660
3661 // Linux mmap allows caller to pass an address as hint; give it a try first,
3662 // if kernel honors the hint then we can return immediately.
3663 char * addr = anon_mmap(requested_addr, bytes, false);
3664 if (addr == requested_addr) {
3665 return requested_addr;
3666 }
3667
3668 if (addr != NULL) {
3669 // mmap() is successful but it fails to reserve at the requested address
3670 anon_munmap(addr, bytes);
3671 }
3672
3673 int i;
3674 for (i = 0; i < max_tries; ++i) {
3675 base[i] = reserve_memory(bytes);
3676
3677 if (base[i] != NULL) {
3678 // Is this the block we wanted?
3679 if (base[i] == requested_addr) {
3680 size[i] = bytes;
3681 break;
3682 }
3683
3684 // Does this overlap the block we wanted? Give back the overlapped
3685 // parts and try again.
3686
3687 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3688 if (top_overlap >= 0 && top_overlap < bytes) {
3689 unmap_memory(base[i], top_overlap);
3690 base[i] += top_overlap;
3691 size[i] = bytes - top_overlap;
3692 } else {
3693 size_t bottom_overlap = base[i] + bytes - requested_addr;
3694 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3695 unmap_memory(requested_addr, bottom_overlap);
3696 size[i] = bytes - bottom_overlap;
3697 } else {
3698 size[i] = bytes;
3699 }
3700 }
3701 }
3702 }
3703
3704 // Give back the unused reserved pieces.
3705
3706 for (int j = 0; j < i; ++j) {
3707 if (base[j] != NULL) {
3708 unmap_memory(base[j], size[j]);
3709 }
3710 }
3711
3712 if (i < max_tries) {
3713 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3714 return requested_addr;
3715 } else {
3716 _highest_vm_reserved_address = old_highest;
3717 return NULL;
3718 }
3719 }
3720
3721 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3722 return ::read(fd, buf, nBytes);
3723 }
3724
3725 //
3726 // Short sleep, direct OS call.
3727 //
3728 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3729 // sched_yield(2) will actually give up the CPU:
3730 //
3731 // * Alone on this pariticular CPU, keeps running.
3732 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3733 // (pre 2.6.39).
3734 //
3735 // So calling this with 0 is an alternative.
3736 //
3737 void os::naked_short_sleep(jlong ms) {
3738 struct timespec req;
3739
3740 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3741 req.tv_sec = 0;
3742 if (ms > 0) {
3743 req.tv_nsec = (ms % 1000) * 1000000;
3744 }
3745 else {
3746 req.tv_nsec = 1;
3747 }
3748
3749 nanosleep(&req, NULL);
3750
3751 return;
3752 }
3753
3754 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3755 void os::infinite_sleep() {
3756 while (true) { // sleep forever ...
3757 ::sleep(100); // ... 100 seconds at a time
3758 }
3759 }
3760
3761 // Used to convert frequent JVM_Yield() to nops
3762 bool os::dont_yield() {
3763 return DontYieldALot;
3764 }
3765
3766 void os::yield() {
3767 sched_yield();
3768 }
3769
3770 os::YieldResult os::NakedYield() { sched_yield(); return os::YIELD_UNKNOWN ;}
3771
3772 void os::yield_all(int attempts) {
3773 // Yields to all threads, including threads with lower priorities
3774 // Threads on Linux are all with same priority. The Solaris style
3775 // os::yield_all() with nanosleep(1ms) is not necessary.
3776 sched_yield();
3777 }
3778
3779 // Called from the tight loops to possibly influence time-sharing heuristics
3780 void os::loop_breaker(int attempts) {
3781 os::yield_all(attempts);
3782 }
3783
3784 ////////////////////////////////////////////////////////////////////////////////
3785 // thread priority support
3786
3787 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3788 // only supports dynamic priority, static priority must be zero. For real-time
3789 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3790 // However, for large multi-threaded applications, SCHED_RR is not only slower
3791 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3792 // of 5 runs - Sep 2005).
3793 //
3794 // The following code actually changes the niceness of kernel-thread/LWP. It
3795 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3796 // not the entire user process, and user level threads are 1:1 mapped to kernel
3797 // threads. It has always been the case, but could change in the future. For
3798 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3799 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3800
3801 int os::java_to_os_priority[CriticalPriority + 1] = {
3802 19, // 0 Entry should never be used
3803
3804 4, // 1 MinPriority
3805 3, // 2
3806 2, // 3
3807
3808 1, // 4
3809 0, // 5 NormPriority
3810 -1, // 6
3811
3812 -2, // 7
3813 -3, // 8
3814 -4, // 9 NearMaxPriority
3815
3816 -5, // 10 MaxPriority
3817
3818 -5 // 11 CriticalPriority
3819 };
3820
3821 static int prio_init() {
3822 if (ThreadPriorityPolicy == 1) {
3823 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3824 // if effective uid is not root. Perhaps, a more elegant way of doing
3825 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3826 if (geteuid() != 0) {
3827 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3828 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3829 }
3830 ThreadPriorityPolicy = 0;
3831 }
3832 }
3833 if (UseCriticalJavaThreadPriority) {
3834 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3835 }
3836 return 0;
3837 }
3838
3839 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3840 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) return OS_OK;
3841
3842 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3843 return (ret == 0) ? OS_OK : OS_ERR;
3844 }
3845
3846 OSReturn os::get_native_priority(const Thread* const thread, int *priority_ptr) {
3847 if ( !UseThreadPriorities || ThreadPriorityPolicy == 0 ) {
3848 *priority_ptr = java_to_os_priority[NormPriority];
3849 return OS_OK;
3850 }
3851
3852 errno = 0;
3853 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3854 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3855 }
3856
3857 // Hint to the underlying OS that a task switch would not be good.
3858 // Void return because it's a hint and can fail.
3859 void os::hint_no_preempt() {}
3860
3861 ////////////////////////////////////////////////////////////////////////////////
3862 // suspend/resume support
3863
3864 // the low-level signal-based suspend/resume support is a remnant from the
3865 // old VM-suspension that used to be for java-suspension, safepoints etc,
3866 // within hotspot. Now there is a single use-case for this:
3867 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3868 // that runs in the watcher thread.
3869 // The remaining code is greatly simplified from the more general suspension
3870 // code that used to be used.
3871 //
3872 // The protocol is quite simple:
3873 // - suspend:
3874 // - sends a signal to the target thread
3875 // - polls the suspend state of the osthread using a yield loop
3876 // - target thread signal handler (SR_handler) sets suspend state
3877 // and blocks in sigsuspend until continued
3878 // - resume:
3879 // - sets target osthread state to continue
3880 // - sends signal to end the sigsuspend loop in the SR_handler
3881 //
3882 // Note that the SR_lock plays no role in this suspend/resume protocol.
3883 //
3884
3885 static void resume_clear_context(OSThread *osthread) {
3886 osthread->set_ucontext(NULL);
3887 osthread->set_siginfo(NULL);
3888 }
3889
3890 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo, ucontext_t* context) {
3891 osthread->set_ucontext(context);
3892 osthread->set_siginfo(siginfo);
3893 }
3894
3895 //
3896 // Handler function invoked when a thread's execution is suspended or
3897 // resumed. We have to be careful that only async-safe functions are
3898 // called here (Note: most pthread functions are not async safe and
3899 // should be avoided.)
3900 //
3901 // Note: sigwait() is a more natural fit than sigsuspend() from an
3902 // interface point of view, but sigwait() prevents the signal hander
3903 // from being run. libpthread would get very confused by not having
3904 // its signal handlers run and prevents sigwait()'s use with the
3905 // mutex granting granting signal.
3906 //
3907 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3908 //
3909 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3910 // Save and restore errno to avoid confusing native code with EINTR
3911 // after sigsuspend.
3912 int old_errno = errno;
3913
3914 Thread* thread = Thread::current();
3915 OSThread* osthread = thread->osthread();
3916 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3917
3918 os::SuspendResume::State current = osthread->sr.state();
3919 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3920 suspend_save_context(osthread, siginfo, context);
3921
3922 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3923 os::SuspendResume::State state = osthread->sr.suspended();
3924 if (state == os::SuspendResume::SR_SUSPENDED) {
3925 sigset_t suspend_set; // signals for sigsuspend()
3926
3927 // get current set of blocked signals and unblock resume signal
3928 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3929 sigdelset(&suspend_set, SR_signum);
3930
3931 sr_semaphore.signal();
3932 // wait here until we are resumed
3933 while (1) {
3934 sigsuspend(&suspend_set);
3935
3936 os::SuspendResume::State result = osthread->sr.running();
3937 if (result == os::SuspendResume::SR_RUNNING) {
3938 sr_semaphore.signal();
3939 break;
3940 }
3941 }
3942
3943 } else if (state == os::SuspendResume::SR_RUNNING) {
3944 // request was cancelled, continue
3945 } else {
3946 ShouldNotReachHere();
3947 }
3948
3949 resume_clear_context(osthread);
3950 } else if (current == os::SuspendResume::SR_RUNNING) {
3951 // request was cancelled, continue
3952 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3953 // ignore
3954 } else {
3955 // ignore
3956 }
3957
3958 errno = old_errno;
3959 }
3960
3961
3962 static int SR_initialize() {
3963 struct sigaction act;
3964 char *s;
3965 /* Get signal number to use for suspend/resume */
3966 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
3967 int sig = ::strtol(s, 0, 10);
3968 if (sig > 0 || sig < _NSIG) {
3969 SR_signum = sig;
3970 }
3971 }
3972
3973 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
3974 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
3975
3976 sigemptyset(&SR_sigset);
3977 sigaddset(&SR_sigset, SR_signum);
3978
3979 /* Set up signal handler for suspend/resume */
3980 act.sa_flags = SA_RESTART|SA_SIGINFO;
3981 act.sa_handler = (void (*)(int)) SR_handler;
3982
3983 // SR_signum is blocked by default.
3984 // 4528190 - We also need to block pthread restart signal (32 on all
3985 // supported Linux platforms). Note that LinuxThreads need to block
3986 // this signal for all threads to work properly. So we don't have
3987 // to use hard-coded signal number when setting up the mask.
3988 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
3989
3990 if (sigaction(SR_signum, &act, 0) == -1) {
3991 return -1;
3992 }
3993
3994 // Save signal flag
3995 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
3996 return 0;
3997 }
3998
3999 static int sr_notify(OSThread* osthread) {
4000 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4001 assert_status(status == 0, status, "pthread_kill");
4002 return status;
4003 }
4004
4005 // "Randomly" selected value for how long we want to spin
4006 // before bailing out on suspending a thread, also how often
4007 // we send a signal to a thread we want to resume
4008 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4009 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4010
4011 // returns true on success and false on error - really an error is fatal
4012 // but this seems the normal response to library errors
4013 static bool do_suspend(OSThread* osthread) {
4014 assert(osthread->sr.is_running(), "thread should be running");
4015 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4016
4017 // mark as suspended and send signal
4018 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4019 // failed to switch, state wasn't running?
4020 ShouldNotReachHere();
4021 return false;
4022 }
4023
4024 if (sr_notify(osthread) != 0) {
4025 ShouldNotReachHere();
4026 }
4027
4028 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4029 while (true) {
4030 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4031 break;
4032 } else {
4033 // timeout
4034 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4035 if (cancelled == os::SuspendResume::SR_RUNNING) {
4036 return false;
4037 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4038 // make sure that we consume the signal on the semaphore as well
4039 sr_semaphore.wait();
4040 break;
4041 } else {
4042 ShouldNotReachHere();
4043 return false;
4044 }
4045 }
4046 }
4047
4048 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4049 return true;
4050 }
4051
4052 static void do_resume(OSThread* osthread) {
4053 assert(osthread->sr.is_suspended(), "thread should be suspended");
4054 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4055
4056 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4057 // failed to switch to WAKEUP_REQUEST
4058 ShouldNotReachHere();
4059 return;
4060 }
4061
4062 while (true) {
4063 if (sr_notify(osthread) == 0) {
4064 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4065 if (osthread->sr.is_running()) {
4066 return;
4067 }
4068 }
4069 } else {
4070 ShouldNotReachHere();
4071 }
4072 }
4073
4074 guarantee(osthread->sr.is_running(), "Must be running!");
4075 }
4076
4077 ///////////////////////////////////////////////////////////////////////////////////
4078 // signal handling (except suspend/resume)
4079
4080 // This routine may be used by user applications as a "hook" to catch signals.
4081 // The user-defined signal handler must pass unrecognized signals to this
4082 // routine, and if it returns true (non-zero), then the signal handler must
4083 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4084 // routine will never retun false (zero), but instead will execute a VM panic
4085 // routine kill the process.
4086 //
4087 // If this routine returns false, it is OK to call it again. This allows
4088 // the user-defined signal handler to perform checks either before or after
4089 // the VM performs its own checks. Naturally, the user code would be making
4090 // a serious error if it tried to handle an exception (such as a null check
4091 // or breakpoint) that the VM was generating for its own correct operation.
4092 //
4093 // This routine may recognize any of the following kinds of signals:
4094 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4095 // It should be consulted by handlers for any of those signals.
4096 //
4097 // The caller of this routine must pass in the three arguments supplied
4098 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4099 // field of the structure passed to sigaction(). This routine assumes that
4100 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4101 //
4102 // Note that the VM will print warnings if it detects conflicting signal
4103 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4104 //
4105 extern "C" JNIEXPORT int
4106 JVM_handle_linux_signal(int signo, siginfo_t* siginfo,
4107 void* ucontext, int abort_if_unrecognized);
4108
4109 void signalHandler(int sig, siginfo_t* info, void* uc) {
4110 assert(info != NULL && uc != NULL, "it must be old kernel");
4111 int orig_errno = errno; // Preserve errno value over signal handler.
4112 JVM_handle_linux_signal(sig, info, uc, true);
4113 errno = orig_errno;
4114 }
4115
4116
4117 // This boolean allows users to forward their own non-matching signals
4118 // to JVM_handle_linux_signal, harmlessly.
4119 bool os::Linux::signal_handlers_are_installed = false;
4120
4121 // For signal-chaining
4122 struct sigaction os::Linux::sigact[MAXSIGNUM];
4123 unsigned int os::Linux::sigs = 0;
4124 bool os::Linux::libjsig_is_loaded = false;
4125 typedef struct sigaction *(*get_signal_t)(int);
4126 get_signal_t os::Linux::get_signal_action = NULL;
4127
4128 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4129 struct sigaction *actp = NULL;
4130
4131 if (libjsig_is_loaded) {
4132 // Retrieve the old signal handler from libjsig
4133 actp = (*get_signal_action)(sig);
4134 }
4135 if (actp == NULL) {
4136 // Retrieve the preinstalled signal handler from jvm
4137 actp = get_preinstalled_handler(sig);
4138 }
4139
4140 return actp;
4141 }
4142
4143 static bool call_chained_handler(struct sigaction *actp, int sig,
4144 siginfo_t *siginfo, void *context) {
4145 // Call the old signal handler
4146 if (actp->sa_handler == SIG_DFL) {
4147 // It's more reasonable to let jvm treat it as an unexpected exception
4148 // instead of taking the default action.
4149 return false;
4150 } else if (actp->sa_handler != SIG_IGN) {
4151 if ((actp->sa_flags & SA_NODEFER) == 0) {
4152 // automaticlly block the signal
4153 sigaddset(&(actp->sa_mask), sig);
4154 }
4155
4156 sa_handler_t hand;
4157 sa_sigaction_t sa;
4158 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4159 // retrieve the chained handler
4160 if (siginfo_flag_set) {
4161 sa = actp->sa_sigaction;
4162 } else {
4163 hand = actp->sa_handler;
4164 }
4165
4166 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4167 actp->sa_handler = SIG_DFL;
4168 }
4169
4170 // try to honor the signal mask
4171 sigset_t oset;
4172 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4173
4174 // call into the chained handler
4175 if (siginfo_flag_set) {
4176 (*sa)(sig, siginfo, context);
4177 } else {
4178 (*hand)(sig);
4179 }
4180
4181 // restore the signal mask
4182 pthread_sigmask(SIG_SETMASK, &oset, 0);
4183 }
4184 // Tell jvm's signal handler the signal is taken care of.
4185 return true;
4186 }
4187
4188 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4189 bool chained = false;
4190 // signal-chaining
4191 if (UseSignalChaining) {
4192 struct sigaction *actp = get_chained_signal_action(sig);
4193 if (actp != NULL) {
4194 chained = call_chained_handler(actp, sig, siginfo, context);
4195 }
4196 }
4197 return chained;
4198 }
4199
4200 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4201 if ((( (unsigned int)1 << sig ) & sigs) != 0) {
4202 return &sigact[sig];
4203 }
4204 return NULL;
4205 }
4206
4207 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4208 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4209 sigact[sig] = oldAct;
4210 sigs |= (unsigned int)1 << sig;
4211 }
4212
4213 // for diagnostic
4214 int os::Linux::sigflags[MAXSIGNUM];
4215
4216 int os::Linux::get_our_sigflags(int sig) {
4217 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4218 return sigflags[sig];
4219 }
4220
4221 void os::Linux::set_our_sigflags(int sig, int flags) {
4222 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4223 sigflags[sig] = flags;
4224 }
4225
4226 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4227 // Check for overwrite.
4228 struct sigaction oldAct;
4229 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4230
4231 void* oldhand = oldAct.sa_sigaction
4232 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4233 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4234 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4235 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4236 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4237 if (AllowUserSignalHandlers || !set_installed) {
4238 // Do not overwrite; user takes responsibility to forward to us.
4239 return;
4240 } else if (UseSignalChaining) {
4241 // save the old handler in jvm
4242 save_preinstalled_handler(sig, oldAct);
4243 // libjsig also interposes the sigaction() call below and saves the
4244 // old sigaction on it own.
4245 } else {
4246 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4247 "%#lx for signal %d.", (long)oldhand, sig));
4248 }
4249 }
4250
4251 struct sigaction sigAct;
4252 sigfillset(&(sigAct.sa_mask));
4253 sigAct.sa_handler = SIG_DFL;
4254 if (!set_installed) {
4255 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4256 } else {
4257 sigAct.sa_sigaction = signalHandler;
4258 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4259 }
4260 // Save flags, which are set by ours
4261 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4262 sigflags[sig] = sigAct.sa_flags;
4263
4264 int ret = sigaction(sig, &sigAct, &oldAct);
4265 assert(ret == 0, "check");
4266
4267 void* oldhand2 = oldAct.sa_sigaction
4268 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4269 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4270 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4271 }
4272
4273 // install signal handlers for signals that HotSpot needs to
4274 // handle in order to support Java-level exception handling.
4275
4276 void os::Linux::install_signal_handlers() {
4277 if (!signal_handlers_are_installed) {
4278 signal_handlers_are_installed = true;
4279
4280 // signal-chaining
4281 typedef void (*signal_setting_t)();
4282 signal_setting_t begin_signal_setting = NULL;
4283 signal_setting_t end_signal_setting = NULL;
4284 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4285 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4286 if (begin_signal_setting != NULL) {
4287 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4288 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4289 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4290 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4291 libjsig_is_loaded = true;
4292 assert(UseSignalChaining, "should enable signal-chaining");
4293 }
4294 if (libjsig_is_loaded) {
4295 // Tell libjsig jvm is setting signal handlers
4296 (*begin_signal_setting)();
4297 }
4298
4299 set_signal_handler(SIGSEGV, true);
4300 set_signal_handler(SIGPIPE, true);
4301 set_signal_handler(SIGBUS, true);
4302 set_signal_handler(SIGILL, true);
4303 set_signal_handler(SIGFPE, true);
4304 #if defined(PPC64)
4305 set_signal_handler(SIGTRAP, true);
4306 #endif
4307 set_signal_handler(SIGXFSZ, true);
4308
4309 if (libjsig_is_loaded) {
4310 // Tell libjsig jvm finishes setting signal handlers
4311 (*end_signal_setting)();
4312 }
4313
4314 // We don't activate signal checker if libjsig is in place, we trust ourselves
4315 // and if UserSignalHandler is installed all bets are off.
4316 // Log that signal checking is off only if -verbose:jni is specified.
4317 if (CheckJNICalls) {
4318 if (libjsig_is_loaded) {
4319 if (PrintJNIResolving) {
4320 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4321 }
4322 check_signals = false;
4323 }
4324 if (AllowUserSignalHandlers) {
4325 if (PrintJNIResolving) {
4326 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4327 }
4328 check_signals = false;
4329 }
4330 }
4331 }
4332 }
4333
4334 // This is the fastest way to get thread cpu time on Linux.
4335 // Returns cpu time (user+sys) for any thread, not only for current.
4336 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4337 // It might work on 2.6.10+ with a special kernel/glibc patch.
4338 // For reference, please, see IEEE Std 1003.1-2004:
4339 // http://www.unix.org/single_unix_specification
4340
4341 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4342 struct timespec tp;
4343 int rc = os::Linux::clock_gettime(clockid, &tp);
4344 assert(rc == 0, "clock_gettime is expected to return 0 code");
4345
4346 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4347 }
4348
4349 /////
4350 // glibc on Linux platform uses non-documented flag
4351 // to indicate, that some special sort of signal
4352 // trampoline is used.
4353 // We will never set this flag, and we should
4354 // ignore this flag in our diagnostic
4355 #ifdef SIGNIFICANT_SIGNAL_MASK
4356 #undef SIGNIFICANT_SIGNAL_MASK
4357 #endif
4358 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4359
4360 static const char* get_signal_handler_name(address handler,
4361 char* buf, int buflen) {
4362 int offset;
4363 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4364 if (found) {
4365 // skip directory names
4366 const char *p1, *p2;
4367 p1 = buf;
4368 size_t len = strlen(os::file_separator());
4369 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4370 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4371 } else {
4372 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4373 }
4374 return buf;
4375 }
4376
4377 static void print_signal_handler(outputStream* st, int sig,
4378 char* buf, size_t buflen) {
4379 struct sigaction sa;
4380
4381 sigaction(sig, NULL, &sa);
4382
4383 // See comment for SIGNIFICANT_SIGNAL_MASK define
4384 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4385
4386 st->print("%s: ", os::exception_name(sig, buf, buflen));
4387
4388 address handler = (sa.sa_flags & SA_SIGINFO)
4389 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4390 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4391
4392 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4393 st->print("SIG_DFL");
4394 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4395 st->print("SIG_IGN");
4396 } else {
4397 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4398 }
4399
4400 st->print(", sa_mask[0]=");
4401 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4402
4403 address rh = VMError::get_resetted_sighandler(sig);
4404 // May be, handler was resetted by VMError?
4405 if(rh != NULL) {
4406 handler = rh;
4407 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4408 }
4409
4410 st->print(", sa_flags=");
4411 os::Posix::print_sa_flags(st, sa.sa_flags);
4412
4413 // Check: is it our handler?
4414 if(handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4415 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4416 // It is our signal handler
4417 // check for flags, reset system-used one!
4418 if((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4419 st->print(
4420 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4421 os::Linux::get_our_sigflags(sig));
4422 }
4423 }
4424 st->cr();
4425 }
4426
4427
4428 #define DO_SIGNAL_CHECK(sig) \
4429 if (!sigismember(&check_signal_done, sig)) \
4430 os::Linux::check_signal_handler(sig)
4431
4432 // This method is a periodic task to check for misbehaving JNI applications
4433 // under CheckJNI, we can add any periodic checks here
4434
4435 void os::run_periodic_checks() {
4436
4437 if (check_signals == false) return;
4438
4439 // SEGV and BUS if overridden could potentially prevent
4440 // generation of hs*.log in the event of a crash, debugging
4441 // such a case can be very challenging, so we absolutely
4442 // check the following for a good measure:
4443 DO_SIGNAL_CHECK(SIGSEGV);
4444 DO_SIGNAL_CHECK(SIGILL);
4445 DO_SIGNAL_CHECK(SIGFPE);
4446 DO_SIGNAL_CHECK(SIGBUS);
4447 DO_SIGNAL_CHECK(SIGPIPE);
4448 DO_SIGNAL_CHECK(SIGXFSZ);
4449 #if defined(PPC64)
4450 DO_SIGNAL_CHECK(SIGTRAP);
4451 #endif
4452
4453 // ReduceSignalUsage allows the user to override these handlers
4454 // see comments at the very top and jvm_solaris.h
4455 if (!ReduceSignalUsage) {
4456 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4457 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4458 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4459 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4460 }
4461
4462 DO_SIGNAL_CHECK(SR_signum);
4463 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4464 }
4465
4466 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4467
4468 static os_sigaction_t os_sigaction = NULL;
4469
4470 void os::Linux::check_signal_handler(int sig) {
4471 char buf[O_BUFLEN];
4472 address jvmHandler = NULL;
4473
4474
4475 struct sigaction act;
4476 if (os_sigaction == NULL) {
4477 // only trust the default sigaction, in case it has been interposed
4478 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4479 if (os_sigaction == NULL) return;
4480 }
4481
4482 os_sigaction(sig, (struct sigaction*)NULL, &act);
4483
4484
4485 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4486
4487 address thisHandler = (act.sa_flags & SA_SIGINFO)
4488 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4489 : CAST_FROM_FN_PTR(address, act.sa_handler) ;
4490
4491
4492 switch(sig) {
4493 case SIGSEGV:
4494 case SIGBUS:
4495 case SIGFPE:
4496 case SIGPIPE:
4497 case SIGILL:
4498 case SIGXFSZ:
4499 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4500 break;
4501
4502 case SHUTDOWN1_SIGNAL:
4503 case SHUTDOWN2_SIGNAL:
4504 case SHUTDOWN3_SIGNAL:
4505 case BREAK_SIGNAL:
4506 jvmHandler = (address)user_handler();
4507 break;
4508
4509 case INTERRUPT_SIGNAL:
4510 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4511 break;
4512
4513 default:
4514 if (sig == SR_signum) {
4515 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4516 } else {
4517 return;
4518 }
4519 break;
4520 }
4521
4522 if (thisHandler != jvmHandler) {
4523 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4524 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4525 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4526 // No need to check this sig any longer
4527 sigaddset(&check_signal_done, sig);
4528 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4529 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4530 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4531 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4532 // No need to check this sig any longer
4533 sigaddset(&check_signal_done, sig);
4534 }
4535
4536 // Dump all the signal
4537 if (sigismember(&check_signal_done, sig)) {
4538 print_signal_handlers(tty, buf, O_BUFLEN);
4539 }
4540 }
4541
4542 extern void report_error(char* file_name, int line_no, char* title, char* format, ...);
4543
4544 extern bool signal_name(int signo, char* buf, size_t len);
4545
4546 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4547 if (0 < exception_code && exception_code <= SIGRTMAX) {
4548 // signal
4549 if (!signal_name(exception_code, buf, size)) {
4550 jio_snprintf(buf, size, "SIG%d", exception_code);
4551 }
4552 return buf;
4553 } else {
4554 return NULL;
4555 }
4556 }
4557
4558 // this is called _before_ the most of global arguments have been parsed
4559 void os::init(void) {
4560 char dummy; /* used to get a guess on initial stack address */
4561 // first_hrtime = gethrtime();
4562
4563 // With LinuxThreads the JavaMain thread pid (primordial thread)
4564 // is different than the pid of the java launcher thread.
4565 // So, on Linux, the launcher thread pid is passed to the VM
4566 // via the sun.java.launcher.pid property.
4567 // Use this property instead of getpid() if it was correctly passed.
4568 // See bug 6351349.
4569 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4570
4571 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4572
4573 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4574
4575 init_random(1234567);
4576
4577 ThreadCritical::initialize();
4578
4579 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4580 if (Linux::page_size() == -1) {
4581 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4582 strerror(errno)));
4583 }
4584 init_page_sizes((size_t) Linux::page_size());
4585
4586 Linux::initialize_system_info();
4587
4588 // main_thread points to the aboriginal thread
4589 Linux::_main_thread = pthread_self();
4590
4591 Linux::clock_init();
4592 initial_time_count = javaTimeNanos();
4593
4594 // pthread_condattr initialization for monotonic clock
4595 int status;
4596 pthread_condattr_t* _condattr = os::Linux::condAttr();
4597 if ((status = pthread_condattr_init(_condattr)) != 0) {
4598 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4599 }
4600 // Only set the clock if CLOCK_MONOTONIC is available
4601 if (os::supports_monotonic_clock()) {
4602 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4603 if (status == EINVAL) {
4604 warning("Unable to use monotonic clock with relative timed-waits" \
4605 " - changes to the time-of-day clock may have adverse affects");
4606 } else {
4607 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4608 }
4609 }
4610 }
4611 // else it defaults to CLOCK_REALTIME
4612
4613 pthread_mutex_init(&dl_mutex, NULL);
4614
4615 // If the pagesize of the VM is greater than 8K determine the appropriate
4616 // number of initial guard pages. The user can change this with the
4617 // command line arguments, if needed.
4618 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4619 StackYellowPages = 1;
4620 StackRedPages = 1;
4621 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4622 }
4623 }
4624
4625 // To install functions for atexit system call
4626 extern "C" {
4627 static void perfMemory_exit_helper() {
4628 perfMemory_exit();
4629 }
4630 }
4631
4632 // this is called _after_ the global arguments have been parsed
4633 jint os::init_2(void)
4634 {
4635 Linux::fast_thread_clock_init();
4636
4637 // Allocate a single page and mark it as readable for safepoint polling
4638 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4639 guarantee( polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page" );
4640
4641 os::set_polling_page( polling_page );
4642
4643 #ifndef PRODUCT
4644 if(Verbose && PrintMiscellaneous)
4645 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n", (intptr_t)polling_page);
4646 #endif
4647
4648 if (!UseMembar) {
4649 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4650 guarantee( mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4651 os::set_memory_serialize_page( mem_serialize_page );
4652
4653 #ifndef PRODUCT
4654 if(Verbose && PrintMiscellaneous)
4655 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n", (intptr_t)mem_serialize_page);
4656 #endif
4657 }
4658
4659 // initialize suspend/resume support - must do this before signal_sets_init()
4660 if (SR_initialize() != 0) {
4661 perror("SR_initialize failed");
4662 return JNI_ERR;
4663 }
4664
4665 Linux::signal_sets_init();
4666 Linux::install_signal_handlers();
4667
4668 // Check minimum allowable stack size for thread creation and to initialize
4669 // the java system classes, including StackOverflowError - depends on page
4670 // size. Add a page for compiler2 recursion in main thread.
4671 // Add in 2*BytesPerWord times page size to account for VM stack during
4672 // class initialization depending on 32 or 64 bit VM.
4673 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4674 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4675 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4676
4677 size_t threadStackSizeInBytes = ThreadStackSize * K;
4678 if (threadStackSizeInBytes != 0 &&
4679 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4680 tty->print_cr("\nThe stack size specified is too small, "
4681 "Specify at least %dk",
4682 os::Linux::min_stack_allowed/ K);
4683 return JNI_ERR;
4684 }
4685
4686 // Make the stack size a multiple of the page size so that
4687 // the yellow/red zones can be guarded.
4688 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4689 vm_page_size()));
4690
4691 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4692
4693 #if defined(IA32)
4694 workaround_expand_exec_shield_cs_limit();
4695 #endif
4696
4697 Linux::libpthread_init();
4698 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4699 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4700 Linux::glibc_version(), Linux::libpthread_version(),
4701 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4702 }
4703
4704 if (UseNUMA) {
4705 if (!Linux::libnuma_init()) {
4706 UseNUMA = false;
4707 } else {
4708 if ((Linux::numa_max_node() < 1)) {
4709 // There's only one node(they start from 0), disable NUMA.
4710 UseNUMA = false;
4711 }
4712 }
4713 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4714 // we can make the adaptive lgrp chunk resizing work. If the user specified
4715 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4716 // disable adaptive resizing.
4717 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4718 if (FLAG_IS_DEFAULT(UseNUMA)) {
4719 UseNUMA = false;
4720 } else {
4721 if (FLAG_IS_DEFAULT(UseLargePages) &&
4722 FLAG_IS_DEFAULT(UseSHM) &&
4723 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4724 UseLargePages = false;
4725 } else {
4726 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4727 UseAdaptiveSizePolicy = false;
4728 UseAdaptiveNUMAChunkSizing = false;
4729 }
4730 }
4731 }
4732 if (!UseNUMA && ForceNUMA) {
4733 UseNUMA = true;
4734 }
4735 }
4736
4737 if (MaxFDLimit) {
4738 // set the number of file descriptors to max. print out error
4739 // if getrlimit/setrlimit fails but continue regardless.
4740 struct rlimit nbr_files;
4741 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4742 if (status != 0) {
4743 if (PrintMiscellaneous && (Verbose || WizardMode))
4744 perror("os::init_2 getrlimit failed");
4745 } else {
4746 nbr_files.rlim_cur = nbr_files.rlim_max;
4747 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4748 if (status != 0) {
4749 if (PrintMiscellaneous && (Verbose || WizardMode))
4750 perror("os::init_2 setrlimit failed");
4751 }
4752 }
4753 }
4754
4755 // Initialize lock used to serialize thread creation (see os::create_thread)
4756 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4757
4758 // at-exit methods are called in the reverse order of their registration.
4759 // atexit functions are called on return from main or as a result of a
4760 // call to exit(3C). There can be only 32 of these functions registered
4761 // and atexit() does not set errno.
4762
4763 if (PerfAllowAtExitRegistration) {
4764 // only register atexit functions if PerfAllowAtExitRegistration is set.
4765 // atexit functions can be delayed until process exit time, which
4766 // can be problematic for embedded VM situations. Embedded VMs should
4767 // call DestroyJavaVM() to assure that VM resources are released.
4768
4769 // note: perfMemory_exit_helper atexit function may be removed in
4770 // the future if the appropriate cleanup code can be added to the
4771 // VM_Exit VMOperation's doit method.
4772 if (atexit(perfMemory_exit_helper) != 0) {
4773 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4774 }
4775 }
4776
4777 // initialize thread priority policy
4778 prio_init();
4779
4780 return JNI_OK;
4781 }
4782
4783 // this is called at the end of vm_initialization
4784 void os::init_3(void) {
4785 #ifdef JAVASE_EMBEDDED
4786 // Start the MemNotifyThread
4787 if (LowMemoryProtection) {
4788 MemNotifyThread::start();
4789 }
4790 return;
4791 #endif
4792 }
4793
4794 // Mark the polling page as unreadable
4795 void os::make_polling_page_unreadable(void) {
4796 if( !guard_memory((char*)_polling_page, Linux::page_size()) )
4797 fatal("Could not disable polling page");
4798 };
4799
4800 // Mark the polling page as readable
4801 void os::make_polling_page_readable(void) {
4802 if( !linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4803 fatal("Could not enable polling page");
4804 }
4805 };
4806
4807 int os::active_processor_count() {
4808 // Linux doesn't yet have a (official) notion of processor sets,
4809 // so just return the number of online processors.
4810 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4811 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4812 return online_cpus;
4813 }
4814
4815 void os::set_native_thread_name(const char *name) {
4816 // Not yet implemented.
4817 return;
4818 }
4819
4820 bool os::distribute_processes(uint length, uint* distribution) {
4821 // Not yet implemented.
4822 return false;
4823 }
4824
4825 bool os::bind_to_processor(uint processor_id) {
4826 // Not yet implemented.
4827 return false;
4828 }
4829
4830 ///
4831
4832 void os::SuspendedThreadTask::internal_do_task() {
4833 if (do_suspend(_thread->osthread())) {
4834 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4835 do_task(context);
4836 do_resume(_thread->osthread());
4837 }
4838 }
4839
4840 class PcFetcher : public os::SuspendedThreadTask {
4841 public:
4842 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4843 ExtendedPC result();
4844 protected:
4845 void do_task(const os::SuspendedThreadTaskContext& context);
4846 private:
4847 ExtendedPC _epc;
4848 };
4849
4850 ExtendedPC PcFetcher::result() {
4851 guarantee(is_done(), "task is not done yet.");
4852 return _epc;
4853 }
4854
4855 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4856 Thread* thread = context.thread();
4857 OSThread* osthread = thread->osthread();
4858 if (osthread->ucontext() != NULL) {
4859 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4860 } else {
4861 // NULL context is unexpected, double-check this is the VMThread
4862 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4863 }
4864 }
4865
4866 // Suspends the target using the signal mechanism and then grabs the PC before
4867 // resuming the target. Used by the flat-profiler only
4868 ExtendedPC os::get_thread_pc(Thread* thread) {
4869 // Make sure that it is called by the watcher for the VMThread
4870 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4871 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4872
4873 PcFetcher fetcher(thread);
4874 fetcher.run();
4875 return fetcher.result();
4876 }
4877
4878 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond, pthread_mutex_t *_mutex, const struct timespec *_abstime)
4879 {
4880 if (is_NPTL()) {
4881 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4882 } else {
4883 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4884 // word back to default 64bit precision if condvar is signaled. Java
4885 // wants 53bit precision. Save and restore current value.
4886 int fpu = get_fpu_control_word();
4887 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4888 set_fpu_control_word(fpu);
4889 return status;
4890 }
4891 }
4892
4893 ////////////////////////////////////////////////////////////////////////////////
4894 // debug support
4895
4896 bool os::find(address addr, outputStream* st) {
4897 Dl_info dlinfo;
4898 memset(&dlinfo, 0, sizeof(dlinfo));
4899 if (dladdr(addr, &dlinfo) != 0) {
4900 st->print(PTR_FORMAT ": ", addr);
4901 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4902 st->print("%s+%#x", dlinfo.dli_sname,
4903 addr - (intptr_t)dlinfo.dli_saddr);
4904 } else if (dlinfo.dli_fbase != NULL) {
4905 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4906 } else {
4907 st->print("<absolute address>");
4908 }
4909 if (dlinfo.dli_fname != NULL) {
4910 st->print(" in %s", dlinfo.dli_fname);
4911 }
4912 if (dlinfo.dli_fbase != NULL) {
4913 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4914 }
4915 st->cr();
4916
4917 if (Verbose) {
4918 // decode some bytes around the PC
4919 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4920 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4921 address lowest = (address) dlinfo.dli_sname;
4922 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4923 if (begin < lowest) begin = lowest;
4924 Dl_info dlinfo2;
4925 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4926 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin)
4927 end = (address) dlinfo2.dli_saddr;
4928 Disassembler::decode(begin, end, st);
4929 }
4930 return true;
4931 }
4932 return false;
4933 }
4934
4935 ////////////////////////////////////////////////////////////////////////////////
4936 // misc
4937
4938 // This does not do anything on Linux. This is basically a hook for being
4939 // able to use structured exception handling (thread-local exception filters)
4940 // on, e.g., Win32.
4941 void
4942 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
4943 JavaCallArguments* args, Thread* thread) {
4944 f(value, method, args, thread);
4945 }
4946
4947 void os::print_statistics() {
4948 }
4949
4950 int os::message_box(const char* title, const char* message) {
4951 int i;
4952 fdStream err(defaultStream::error_fd());
4953 for (i = 0; i < 78; i++) err.print_raw("=");
4954 err.cr();
4955 err.print_raw_cr(title);
4956 for (i = 0; i < 78; i++) err.print_raw("-");
4957 err.cr();
4958 err.print_raw_cr(message);
4959 for (i = 0; i < 78; i++) err.print_raw("=");
4960 err.cr();
4961
4962 char buf[16];
4963 // Prevent process from exiting upon "read error" without consuming all CPU
4964 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
4965
4966 return buf[0] == 'y' || buf[0] == 'Y';
4967 }
4968
4969 int os::stat(const char *path, struct stat *sbuf) {
4970 char pathbuf[MAX_PATH];
4971 if (strlen(path) > MAX_PATH - 1) {
4972 errno = ENAMETOOLONG;
4973 return -1;
4974 }
4975 os::native_path(strcpy(pathbuf, path));
4976 return ::stat(pathbuf, sbuf);
4977 }
4978
4979 bool os::check_heap(bool force) {
4980 return true;
4981 }
4982
4983 int local_vsnprintf(char* buf, size_t count, const char* format, va_list args) {
4984 return ::vsnprintf(buf, count, format, args);
4985 }
4986
4987 // Is a (classpath) directory empty?
4988 bool os::dir_is_empty(const char* path) {
4989 DIR *dir = NULL;
4990 struct dirent *ptr;
4991
4992 dir = opendir(path);
4993 if (dir == NULL) return true;
4994
4995 /* Scan the directory */
4996 bool result = true;
4997 char buf[sizeof(struct dirent) + MAX_PATH];
4998 while (result && (ptr = ::readdir(dir)) != NULL) {
4999 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5000 result = false;
5001 }
5002 }
5003 closedir(dir);
5004 return result;
5005 }
5006
5007 // This code originates from JDK's sysOpen and open64_w
5008 // from src/solaris/hpi/src/system_md.c
5009
5010 #ifndef O_DELETE
5011 #define O_DELETE 0x10000
5012 #endif
5013
5014 // Open a file. Unlink the file immediately after open returns
5015 // if the specified oflag has the O_DELETE flag set.
5016 // O_DELETE is used only in j2se/src/share/native/java/util/zip/ZipFile.c
5017
5018 int os::open(const char *path, int oflag, int mode) {
5019
5020 if (strlen(path) > MAX_PATH - 1) {
5021 errno = ENAMETOOLONG;
5022 return -1;
5023 }
5024 int fd;
5025 int o_delete = (oflag & O_DELETE);
5026 oflag = oflag & ~O_DELETE;
5027
5028 fd = ::open64(path, oflag, mode);
5029 if (fd == -1) return -1;
5030
5031 //If the open succeeded, the file might still be a directory
5032 {
5033 struct stat64 buf64;
5034 int ret = ::fstat64(fd, &buf64);
5035 int st_mode = buf64.st_mode;
5036
5037 if (ret != -1) {
5038 if ((st_mode & S_IFMT) == S_IFDIR) {
5039 errno = EISDIR;
5040 ::close(fd);
5041 return -1;
5042 }
5043 } else {
5044 ::close(fd);
5045 return -1;
5046 }
5047 }
5048
5049 /*
5050 * All file descriptors that are opened in the JVM and not
5051 * specifically destined for a subprocess should have the
5052 * close-on-exec flag set. If we don't set it, then careless 3rd
5053 * party native code might fork and exec without closing all
5054 * appropriate file descriptors (e.g. as we do in closeDescriptors in
5055 * UNIXProcess.c), and this in turn might:
5056 *
5057 * - cause end-of-file to fail to be detected on some file
5058 * descriptors, resulting in mysterious hangs, or
5059 *
5060 * - might cause an fopen in the subprocess to fail on a system
5061 * suffering from bug 1085341.
5062 *
5063 * (Yes, the default setting of the close-on-exec flag is a Unix
5064 * design flaw)
5065 *
5066 * See:
5067 * 1085341: 32-bit stdio routines should support file descriptors >255
5068 * 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5069 * 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5070 */
5071 #ifdef FD_CLOEXEC
5072 {
5073 int flags = ::fcntl(fd, F_GETFD);
5074 if (flags != -1)
5075 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5076 }
5077 #endif
5078
5079 if (o_delete != 0) {
5080 ::unlink(path);
5081 }
5082 return fd;
5083 }
5084
5085
5086 // create binary file, rewriting existing file if required
5087 int os::create_binary_file(const char* path, bool rewrite_existing) {
5088 int oflags = O_WRONLY | O_CREAT;
5089 if (!rewrite_existing) {
5090 oflags |= O_EXCL;
5091 }
5092 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5093 }
5094
5095 // return current position of file pointer
5096 jlong os::current_file_offset(int fd) {
5097 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5098 }
5099
5100 // move file pointer to the specified offset
5101 jlong os::seek_to_file_offset(int fd, jlong offset) {
5102 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5103 }
5104
5105 // This code originates from JDK's sysAvailable
5106 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5107
5108 int os::available(int fd, jlong *bytes) {
5109 jlong cur, end;
5110 int mode;
5111 struct stat64 buf64;
5112
5113 if (::fstat64(fd, &buf64) >= 0) {
5114 mode = buf64.st_mode;
5115 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5116 /*
5117 * XXX: is the following call interruptible? If so, this might
5118 * need to go through the INTERRUPT_IO() wrapper as for other
5119 * blocking, interruptible calls in this file.
5120 */
5121 int n;
5122 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5123 *bytes = n;
5124 return 1;
5125 }
5126 }
5127 }
5128 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5129 return 0;
5130 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5131 return 0;
5132 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5133 return 0;
5134 }
5135 *bytes = end - cur;
5136 return 1;
5137 }
5138
5139 int os::socket_available(int fd, jint *pbytes) {
5140 // Linux doc says EINTR not returned, unlike Solaris
5141 int ret = ::ioctl(fd, FIONREAD, pbytes);
5142
5143 //%% note ioctl can return 0 when successful, JVM_SocketAvailable
5144 // is expected to return 0 on failure and 1 on success to the jdk.
5145 return (ret < 0) ? 0 : 1;
5146 }
5147
5148 // Map a block of memory.
5149 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5150 char *addr, size_t bytes, bool read_only,
5151 bool allow_exec) {
5152 int prot;
5153 int flags = MAP_PRIVATE;
5154
5155 if (read_only) {
5156 prot = PROT_READ;
5157 } else {
5158 prot = PROT_READ | PROT_WRITE;
5159 }
5160
5161 if (allow_exec) {
5162 prot |= PROT_EXEC;
5163 }
5164
5165 if (addr != NULL) {
5166 flags |= MAP_FIXED;
5167 }
5168
5169 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5170 fd, file_offset);
5171 if (mapped_address == MAP_FAILED) {
5172 return NULL;
5173 }
5174 return mapped_address;
5175 }
5176
5177
5178 // Remap a block of memory.
5179 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5180 char *addr, size_t bytes, bool read_only,
5181 bool allow_exec) {
5182 // same as map_memory() on this OS
5183 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5184 allow_exec);
5185 }
5186
5187
5188 // Unmap a block of memory.
5189 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5190 return munmap(addr, bytes) == 0;
5191 }
5192
5193 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5194
5195 static clockid_t thread_cpu_clockid(Thread* thread) {
5196 pthread_t tid = thread->osthread()->pthread_id();
5197 clockid_t clockid;
5198
5199 // Get thread clockid
5200 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5201 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5202 return clockid;
5203 }
5204
5205 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5206 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5207 // of a thread.
5208 //
5209 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5210 // the fast estimate available on the platform.
5211
5212 jlong os::current_thread_cpu_time() {
5213 if (os::Linux::supports_fast_thread_cpu_time()) {
5214 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5215 } else {
5216 // return user + sys since the cost is the same
5217 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5218 }
5219 }
5220
5221 jlong os::thread_cpu_time(Thread* thread) {
5222 // consistent with what current_thread_cpu_time() returns
5223 if (os::Linux::supports_fast_thread_cpu_time()) {
5224 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5225 } else {
5226 return slow_thread_cpu_time(thread, true /* user + sys */);
5227 }
5228 }
5229
5230 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5231 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5232 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5233 } else {
5234 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5235 }
5236 }
5237
5238 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5239 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5240 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5241 } else {
5242 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5243 }
5244 }
5245
5246 //
5247 // -1 on error.
5248 //
5249
5250 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5251 static bool proc_task_unchecked = true;
5252 pid_t tid = thread->osthread()->thread_id();
5253 char *s;
5254 char stat[2048];
5255 int statlen;
5256 char proc_name[64];
5257 int count;
5258 long sys_time, user_time;
5259 char cdummy;
5260 int idummy;
5261 long ldummy;
5262 FILE *fp;
5263
5264 snprintf(proc_name, 64, "/proc/%d/stat", tid);
5265
5266 // The /proc/<tid>/stat aggregates per-process usage on
5267 // new Linux kernels 2.6+ where NPTL is supported.
5268 // The /proc/self/task/<tid>/stat still has the per-thread usage.
5269 // See bug 6328462.
5270 // There possibly can be cases where there is no directory
5271 // /proc/self/task, so we check its availability.
5272 if (proc_task_unchecked && os::Linux::is_NPTL()) {
5273 // This is executed only once
5274 proc_task_unchecked = false;
5275 fp = fopen("/proc/self/task", "r");
5276 if (fp != NULL) {
5277 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5278 fclose(fp);
5279 }
5280 }
5281
5282 fp = fopen(proc_name, "r");
5283 if ( fp == NULL ) return -1;
5284 statlen = fread(stat, 1, 2047, fp);
5285 stat[statlen] = '\0';
5286 fclose(fp);
5287
5288 // Skip pid and the command string. Note that we could be dealing with
5289 // weird command names, e.g. user could decide to rename java launcher
5290 // to "java 1.4.2 :)", then the stat file would look like
5291 // 1234 (java 1.4.2 :)) R ... ...
5292 // We don't really need to know the command string, just find the last
5293 // occurrence of ")" and then start parsing from there. See bug 4726580.
5294 s = strrchr(stat, ')');
5295 if (s == NULL ) return -1;
5296
5297 // Skip blank chars
5298 do s++; while (isspace(*s));
5299
5300 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5301 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5302 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5303 &user_time, &sys_time);
5304 if ( count != 13 ) return -1;
5305 if (user_sys_cpu_time) {
5306 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5307 } else {
5308 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5309 }
5310 }
5311
5312 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5313 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5314 info_ptr->may_skip_backward = false; // elapsed time not wall time
5315 info_ptr->may_skip_forward = false; // elapsed time not wall time
5316 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5317 }
5318
5319 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5320 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5321 info_ptr->may_skip_backward = false; // elapsed time not wall time
5322 info_ptr->may_skip_forward = false; // elapsed time not wall time
5323 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5324 }
5325
5326 bool os::is_thread_cpu_time_supported() {
5327 return true;
5328 }
5329
5330 // System loadavg support. Returns -1 if load average cannot be obtained.
5331 // Linux doesn't yet have a (official) notion of processor sets,
5332 // so just return the system wide load average.
5333 int os::loadavg(double loadavg[], int nelem) {
5334 return ::getloadavg(loadavg, nelem);
5335 }
5336
5337 void os::pause() {
5338 char filename[MAX_PATH];
5339 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5340 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5341 } else {
5342 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5343 }
5344
5345 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5346 if (fd != -1) {
5347 struct stat buf;
5348 ::close(fd);
5349 while (::stat(filename, &buf) == 0) {
5350 (void)::poll(NULL, 0, 100);
5351 }
5352 } else {
5353 jio_fprintf(stderr,
5354 "Could not open pause file '%s', continuing immediately.\n", filename);
5355 }
5356 }
5357
5358
5359 // Refer to the comments in os_solaris.cpp park-unpark.
5360 //
5361 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5362 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5363 // For specifics regarding the bug see GLIBC BUGID 261237 :
5364 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5365 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5366 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5367 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5368 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5369 // and monitorenter when we're using 1-0 locking. All those operations may result in
5370 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5371 // of libpthread avoids the problem, but isn't practical.
5372 //
5373 // Possible remedies:
5374 //
5375 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5376 // This is palliative and probabilistic, however. If the thread is preempted
5377 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5378 // than the minimum period may have passed, and the abstime may be stale (in the
5379 // past) resultin in a hang. Using this technique reduces the odds of a hang
5380 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5381 //
5382 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5383 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5384 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5385 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5386 // thread.
5387 //
5388 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5389 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5390 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5391 // This also works well. In fact it avoids kernel-level scalability impediments
5392 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5393 // timers in a graceful fashion.
5394 //
5395 // 4. When the abstime value is in the past it appears that control returns
5396 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5397 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5398 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5399 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5400 // It may be possible to avoid reinitialization by checking the return
5401 // value from pthread_cond_timedwait(). In addition to reinitializing the
5402 // condvar we must establish the invariant that cond_signal() is only called
5403 // within critical sections protected by the adjunct mutex. This prevents
5404 // cond_signal() from "seeing" a condvar that's in the midst of being
5405 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5406 // desirable signal-after-unlock optimization that avoids futile context switching.
5407 //
5408 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5409 // structure when a condvar is used or initialized. cond_destroy() would
5410 // release the helper structure. Our reinitialize-after-timedwait fix
5411 // put excessive stress on malloc/free and locks protecting the c-heap.
5412 //
5413 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5414 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5415 // and only enabling the work-around for vulnerable environments.
5416
5417 // utility to compute the abstime argument to timedwait:
5418 // millis is the relative timeout time
5419 // abstime will be the absolute timeout time
5420 // TODO: replace compute_abstime() with unpackTime()
5421
5422 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5423 if (millis < 0) millis = 0;
5424
5425 jlong seconds = millis / 1000;
5426 millis %= 1000;
5427 if (seconds > 50000000) { // see man cond_timedwait(3T)
5428 seconds = 50000000;
5429 }
5430
5431 if (os::supports_monotonic_clock()) {
5432 struct timespec now;
5433 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5434 assert_status(status == 0, status, "clock_gettime");
5435 abstime->tv_sec = now.tv_sec + seconds;
5436 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5437 if (nanos >= NANOSECS_PER_SEC) {
5438 abstime->tv_sec += 1;
5439 nanos -= NANOSECS_PER_SEC;
5440 }
5441 abstime->tv_nsec = nanos;
5442 } else {
5443 struct timeval now;
5444 int status = gettimeofday(&now, NULL);
5445 assert(status == 0, "gettimeofday");
5446 abstime->tv_sec = now.tv_sec + seconds;
5447 long usec = now.tv_usec + millis * 1000;
5448 if (usec >= 1000000) {
5449 abstime->tv_sec += 1;
5450 usec -= 1000000;
5451 }
5452 abstime->tv_nsec = usec * 1000;
5453 }
5454 return abstime;
5455 }
5456
5457
5458 // Test-and-clear _Event, always leaves _Event set to 0, returns immediately.
5459 // Conceptually TryPark() should be equivalent to park(0).
5460
5461 int os::PlatformEvent::TryPark() {
5462 for (;;) {
5463 const int v = _Event ;
5464 guarantee ((v == 0) || (v == 1), "invariant") ;
5465 if (Atomic::cmpxchg (0, &_Event, v) == v) return v ;
5466 }
5467 }
5468
5469 void os::PlatformEvent::park() { // AKA "down()"
5470 // Invariant: Only the thread associated with the Event/PlatformEvent
5471 // may call park().
5472 // TODO: assert that _Assoc != NULL or _Assoc == Self
5473 int v ;
5474 for (;;) {
5475 v = _Event ;
5476 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5477 }
5478 guarantee (v >= 0, "invariant") ;
5479 if (v == 0) {
5480 // Do this the hard way by blocking ...
5481 int status = pthread_mutex_lock(_mutex);
5482 assert_status(status == 0, status, "mutex_lock");
5483 guarantee (_nParked == 0, "invariant") ;
5484 ++ _nParked ;
5485 while (_Event < 0) {
5486 status = pthread_cond_wait(_cond, _mutex);
5487 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5488 // Treat this the same as if the wait was interrupted
5489 if (status == ETIME) { status = EINTR; }
5490 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5491 }
5492 -- _nParked ;
5493
5494 _Event = 0 ;
5495 status = pthread_mutex_unlock(_mutex);
5496 assert_status(status == 0, status, "mutex_unlock");
5497 // Paranoia to ensure our locked and lock-free paths interact
5498 // correctly with each other.
5499 OrderAccess::fence();
5500 }
5501 guarantee (_Event >= 0, "invariant") ;
5502 }
5503
5504 int os::PlatformEvent::park(jlong millis) {
5505 guarantee (_nParked == 0, "invariant") ;
5506
5507 int v ;
5508 for (;;) {
5509 v = _Event ;
5510 if (Atomic::cmpxchg (v-1, &_Event, v) == v) break ;
5511 }
5512 guarantee (v >= 0, "invariant") ;
5513 if (v != 0) return OS_OK ;
5514
5515 // We do this the hard way, by blocking the thread.
5516 // Consider enforcing a minimum timeout value.
5517 struct timespec abst;
5518 compute_abstime(&abst, millis);
5519
5520 int ret = OS_TIMEOUT;
5521 int status = pthread_mutex_lock(_mutex);
5522 assert_status(status == 0, status, "mutex_lock");
5523 guarantee (_nParked == 0, "invariant") ;
5524 ++_nParked ;
5525
5526 // Object.wait(timo) will return because of
5527 // (a) notification
5528 // (b) timeout
5529 // (c) thread.interrupt
5530 //
5531 // Thread.interrupt and object.notify{All} both call Event::set.
5532 // That is, we treat thread.interrupt as a special case of notification.
5533 // The underlying Solaris implementation, cond_timedwait, admits
5534 // spurious/premature wakeups, but the JLS/JVM spec prevents the
5535 // JVM from making those visible to Java code. As such, we must
5536 // filter out spurious wakeups. We assume all ETIME returns are valid.
5537 //
5538 // TODO: properly differentiate simultaneous notify+interrupt.
5539 // In that case, we should propagate the notify to another waiter.
5540
5541 while (_Event < 0) {
5542 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5543 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5544 pthread_cond_destroy (_cond);
5545 pthread_cond_init (_cond, os::Linux::condAttr()) ;
5546 }
5547 assert_status(status == 0 || status == EINTR ||
5548 status == ETIME || status == ETIMEDOUT,
5549 status, "cond_timedwait");
5550 if (!FilterSpuriousWakeups) break ; // previous semantics
5551 if (status == ETIME || status == ETIMEDOUT) break ;
5552 // We consume and ignore EINTR and spurious wakeups.
5553 }
5554 --_nParked ;
5555 if (_Event >= 0) {
5556 ret = OS_OK;
5557 }
5558 _Event = 0 ;
5559 status = pthread_mutex_unlock(_mutex);
5560 assert_status(status == 0, status, "mutex_unlock");
5561 assert (_nParked == 0, "invariant") ;
5562 // Paranoia to ensure our locked and lock-free paths interact
5563 // correctly with each other.
5564 OrderAccess::fence();
5565 return ret;
5566 }
5567
5568 void os::PlatformEvent::unpark() {
5569 // Transitions for _Event:
5570 // 0 :=> 1
5571 // 1 :=> 1
5572 // -1 :=> either 0 or 1; must signal target thread
5573 // That is, we can safely transition _Event from -1 to either
5574 // 0 or 1. Forcing 1 is slightly more efficient for back-to-back
5575 // unpark() calls.
5576 // See also: "Semaphores in Plan 9" by Mullender & Cox
5577 //
5578 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5579 // that it will take two back-to-back park() calls for the owning
5580 // thread to block. This has the benefit of forcing a spurious return
5581 // from the first park() call after an unpark() call which will help
5582 // shake out uses of park() and unpark() without condition variables.
5583
5584 if (Atomic::xchg(1, &_Event) >= 0) return;
5585
5586 // Wait for the thread associated with the event to vacate
5587 int status = pthread_mutex_lock(_mutex);
5588 assert_status(status == 0, status, "mutex_lock");
5589 int AnyWaiters = _nParked;
5590 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5591 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5592 AnyWaiters = 0;
5593 pthread_cond_signal(_cond);
5594 }
5595 status = pthread_mutex_unlock(_mutex);
5596 assert_status(status == 0, status, "mutex_unlock");
5597 if (AnyWaiters != 0) {
5598 status = pthread_cond_signal(_cond);
5599 assert_status(status == 0, status, "cond_signal");
5600 }
5601
5602 // Note that we signal() _after dropping the lock for "immortal" Events.
5603 // This is safe and avoids a common class of futile wakeups. In rare
5604 // circumstances this can cause a thread to return prematurely from
5605 // cond_{timed}wait() but the spurious wakeup is benign and the victim will
5606 // simply re-test the condition and re-park itself.
5607 }
5608
5609
5610 // JSR166
5611 // -------------------------------------------------------
5612
5613 /*
5614 * The solaris and linux implementations of park/unpark are fairly
5615 * conservative for now, but can be improved. They currently use a
5616 * mutex/condvar pair, plus a a count.
5617 * Park decrements count if > 0, else does a condvar wait. Unpark
5618 * sets count to 1 and signals condvar. Only one thread ever waits
5619 * on the condvar. Contention seen when trying to park implies that someone
5620 * is unparking you, so don't wait. And spurious returns are fine, so there
5621 * is no need to track notifications.
5622 */
5623
5624 #define MAX_SECS 100000000
5625 /*
5626 * This code is common to linux and solaris and will be moved to a
5627 * common place in dolphin.
5628 *
5629 * The passed in time value is either a relative time in nanoseconds
5630 * or an absolute time in milliseconds. Either way it has to be unpacked
5631 * into suitable seconds and nanoseconds components and stored in the
5632 * given timespec structure.
5633 * Given time is a 64-bit value and the time_t used in the timespec is only
5634 * a signed-32-bit value (except on 64-bit Linux) we have to watch for
5635 * overflow if times way in the future are given. Further on Solaris versions
5636 * prior to 10 there is a restriction (see cond_timedwait) that the specified
5637 * number of seconds, in abstime, is less than current_time + 100,000,000.
5638 * As it will be 28 years before "now + 100000000" will overflow we can
5639 * ignore overflow and just impose a hard-limit on seconds using the value
5640 * of "now + 100,000,000". This places a limit on the timeout of about 3.17
5641 * years from "now".
5642 */
5643
5644 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5645 assert (time > 0, "convertTime");
5646 time_t max_secs = 0;
5647
5648 if (!os::supports_monotonic_clock() || isAbsolute) {
5649 struct timeval now;
5650 int status = gettimeofday(&now, NULL);
5651 assert(status == 0, "gettimeofday");
5652
5653 max_secs = now.tv_sec + MAX_SECS;
5654
5655 if (isAbsolute) {
5656 jlong secs = time / 1000;
5657 if (secs > max_secs) {
5658 absTime->tv_sec = max_secs;
5659 } else {
5660 absTime->tv_sec = secs;
5661 }
5662 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5663 } else {
5664 jlong secs = time / NANOSECS_PER_SEC;
5665 if (secs >= MAX_SECS) {
5666 absTime->tv_sec = max_secs;
5667 absTime->tv_nsec = 0;
5668 } else {
5669 absTime->tv_sec = now.tv_sec + secs;
5670 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5671 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5672 absTime->tv_nsec -= NANOSECS_PER_SEC;
5673 ++absTime->tv_sec; // note: this must be <= max_secs
5674 }
5675 }
5676 }
5677 } else {
5678 // must be relative using monotonic clock
5679 struct timespec now;
5680 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5681 assert_status(status == 0, status, "clock_gettime");
5682 max_secs = now.tv_sec + MAX_SECS;
5683 jlong secs = time / NANOSECS_PER_SEC;
5684 if (secs >= MAX_SECS) {
5685 absTime->tv_sec = max_secs;
5686 absTime->tv_nsec = 0;
5687 } else {
5688 absTime->tv_sec = now.tv_sec + secs;
5689 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5690 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5691 absTime->tv_nsec -= NANOSECS_PER_SEC;
5692 ++absTime->tv_sec; // note: this must be <= max_secs
5693 }
5694 }
5695 }
5696 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5697 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5698 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5699 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5700 }
5701
5702 void Parker::park(bool isAbsolute, jlong time) {
5703 // Ideally we'd do something useful while spinning, such
5704 // as calling unpackTime().
5705
5706 // Optional fast-path check:
5707 // Return immediately if a permit is available.
5708 // We depend on Atomic::xchg() having full barrier semantics
5709 // since we are doing a lock-free update to _counter.
5710 if (Atomic::xchg(0, &_counter) > 0) return;
5711
5712 Thread* thread = Thread::current();
5713 assert(thread->is_Java_thread(), "Must be JavaThread");
5714 JavaThread *jt = (JavaThread *)thread;
5715
5716 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5717 // Check interrupt before trying to wait
5718 if (Thread::is_interrupted(thread, false)) {
5719 return;
5720 }
5721
5722 // Next, demultiplex/decode time arguments
5723 timespec absTime;
5724 if (time < 0 || (isAbsolute && time == 0) ) { // don't wait at all
5725 return;
5726 }
5727 if (time > 0) {
5728 unpackTime(&absTime, isAbsolute, time);
5729 }
5730
5731
5732 // Enter safepoint region
5733 // Beware of deadlocks such as 6317397.
5734 // The per-thread Parker:: mutex is a classic leaf-lock.
5735 // In particular a thread must never block on the Threads_lock while
5736 // holding the Parker:: mutex. If safepoints are pending both the
5737 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5738 ThreadBlockInVM tbivm(jt);
5739
5740 // Don't wait if cannot get lock since interference arises from
5741 // unblocking. Also. check interrupt before trying wait
5742 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5743 return;
5744 }
5745
5746 int status ;
5747 if (_counter > 0) { // no wait needed
5748 _counter = 0;
5749 status = pthread_mutex_unlock(_mutex);
5750 assert (status == 0, "invariant") ;
5751 // Paranoia to ensure our locked and lock-free paths interact
5752 // correctly with each other and Java-level accesses.
5753 OrderAccess::fence();
5754 return;
5755 }
5756
5757 #ifdef ASSERT
5758 // Don't catch signals while blocked; let the running threads have the signals.
5759 // (This allows a debugger to break into the running thread.)
5760 sigset_t oldsigs;
5761 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5762 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5763 #endif
5764
5765 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5766 jt->set_suspend_equivalent();
5767 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5768
5769 assert(_cur_index == -1, "invariant");
5770 if (time == 0) {
5771 _cur_index = REL_INDEX; // arbitrary choice when not timed
5772 status = pthread_cond_wait (&_cond[_cur_index], _mutex) ;
5773 } else {
5774 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5775 status = os::Linux::safe_cond_timedwait (&_cond[_cur_index], _mutex, &absTime) ;
5776 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5777 pthread_cond_destroy (&_cond[_cur_index]) ;
5778 pthread_cond_init (&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5779 }
5780 }
5781 _cur_index = -1;
5782 assert_status(status == 0 || status == EINTR ||
5783 status == ETIME || status == ETIMEDOUT,
5784 status, "cond_timedwait");
5785
5786 #ifdef ASSERT
5787 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5788 #endif
5789
5790 _counter = 0 ;
5791 status = pthread_mutex_unlock(_mutex) ;
5792 assert_status(status == 0, status, "invariant") ;
5793 // Paranoia to ensure our locked and lock-free paths interact
5794 // correctly with each other and Java-level accesses.
5795 OrderAccess::fence();
5796
5797 // If externally suspended while waiting, re-suspend
5798 if (jt->handle_special_suspend_equivalent_condition()) {
5799 jt->java_suspend_self();
5800 }
5801 }
5802
5803 void Parker::unpark() {
5804 int s, status ;
5805 status = pthread_mutex_lock(_mutex);
5806 assert (status == 0, "invariant") ;
5807 s = _counter;
5808 _counter = 1;
5809 if (s < 1) {
5810 // thread might be parked
5811 if (_cur_index != -1) {
5812 // thread is definitely parked
5813 if (WorkAroundNPTLTimedWaitHang) {
5814 status = pthread_cond_signal (&_cond[_cur_index]);
5815 assert (status == 0, "invariant");
5816 status = pthread_mutex_unlock(_mutex);
5817 assert (status == 0, "invariant");
5818 } else {
5819 status = pthread_mutex_unlock(_mutex);
5820 assert (status == 0, "invariant");
5821 status = pthread_cond_signal (&_cond[_cur_index]);
5822 assert (status == 0, "invariant");
5823 }
5824 } else {
5825 pthread_mutex_unlock(_mutex);
5826 assert (status == 0, "invariant") ;
5827 }
5828 } else {
5829 pthread_mutex_unlock(_mutex);
5830 assert (status == 0, "invariant") ;
5831 }
5832 }
5833
5834
5835 extern char** environ;
5836
5837 #ifndef __NR_fork
5838 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57)
5839 #endif
5840
5841 #ifndef __NR_execve
5842 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59)
5843 #endif
5844
5845 // Run the specified command in a separate process. Return its exit value,
5846 // or -1 on failure (e.g. can't fork a new process).
5847 // Unlike system(), this function can be called from signal handler. It
5848 // doesn't block SIGINT et al.
5849 int os::fork_and_exec(char* cmd) {
5850 const char * argv[4] = {"sh", "-c", cmd, NULL};
5851
5852 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5853 // pthread_atfork handlers and reset pthread library. All we need is a
5854 // separate process to execve. Make a direct syscall to fork process.
5855 // On IA64 there's no fork syscall, we have to use fork() and hope for
5856 // the best...
5857 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5858 IA64_ONLY(fork();)
5859
5860 if (pid < 0) {
5861 // fork failed
5862 return -1;
5863
5864 } else if (pid == 0) {
5865 // child process
5866
5867 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5868 // first to kill every thread on the thread list. Because this list is
5869 // not reset by fork() (see notes above), execve() will instead kill
5870 // every thread in the parent process. We know this is the only thread
5871 // in the new process, so make a system call directly.
5872 // IA64 should use normal execve() from glibc to match the glibc fork()
5873 // above.
5874 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5875 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5876
5877 // execve failed
5878 _exit(-1);
5879
5880 } else {
5881 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5882 // care about the actual exit code, for now.
5883
5884 int status;
5885
5886 // Wait for the child process to exit. This returns immediately if
5887 // the child has already exited. */
5888 while (waitpid(pid, &status, 0) < 0) {
5889 switch (errno) {
5890 case ECHILD: return 0;
5891 case EINTR: break;
5892 default: return -1;
5893 }
5894 }
5895
5896 if (WIFEXITED(status)) {
5897 // The child exited normally; get its exit code.
5898 return WEXITSTATUS(status);
5899 } else if (WIFSIGNALED(status)) {
5900 // The child exited because of a signal
5901 // The best value to return is 0x80 + signal number,
5902 // because that is what all Unix shells do, and because
5903 // it allows callers to distinguish between process exit and
5904 // process death by signal.
5905 return 0x80 + WTERMSIG(status);
5906 } else {
5907 // Unknown exit code; pass it through
5908 return status;
5909 }
5910 }
5911 }
5912
5913 // is_headless_jre()
5914 //
5915 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5916 // in order to report if we are running in a headless jre
5917 //
5918 // Since JDK8 xawt/libmawt.so was moved into the same directory
5919 // as libawt.so, and renamed libawt_xawt.so
5920 //
5921 bool os::is_headless_jre() {
5922 struct stat statbuf;
5923 char buf[MAXPATHLEN];
5924 char libmawtpath[MAXPATHLEN];
5925 const char *xawtstr = "/xawt/libmawt.so";
5926 const char *new_xawtstr = "/libawt_xawt.so";
5927 char *p;
5928
5929 // Get path to libjvm.so
5930 os::jvm_path(buf, sizeof(buf));
5931
5932 // Get rid of libjvm.so
5933 p = strrchr(buf, '/');
5934 if (p == NULL) return false;
5935 else *p = '\0';
5936
5937 // Get rid of client or server
5938 p = strrchr(buf, '/');
5939 if (p == NULL) return false;
5940 else *p = '\0';
5941
5942 // check xawt/libmawt.so
5943 strcpy(libmawtpath, buf);
5944 strcat(libmawtpath, xawtstr);
5945 if (::stat(libmawtpath, &statbuf) == 0) return false;
5946
5947 // check libawt_xawt.so
5948 strcpy(libmawtpath, buf);
5949 strcat(libmawtpath, new_xawtstr);
5950 if (::stat(libmawtpath, &statbuf) == 0) return false;
5951
5952 return true;
5953 }
5954
5955 // Get the default path to the core file
5956 // Returns the length of the string
5957 int os::get_core_path(char* buffer, size_t bufferSize) {
5958 const char* p = get_current_directory(buffer, bufferSize);
5959
5960 if (p == NULL) {
5961 assert(p != NULL, "failed to get current directory");
5962 return 0;
5963 }
5964
5965 return strlen(buffer);
5966 }
5967
5968 #ifdef JAVASE_EMBEDDED
5969 //
5970 // A thread to watch the '/dev/mem_notify' device, which will tell us when the OS is running low on memory.
5971 //
5972 MemNotifyThread* MemNotifyThread::_memnotify_thread = NULL;
5973
5974 // ctor
5975 //
5976 MemNotifyThread::MemNotifyThread(int fd): Thread() {
5977 assert(memnotify_thread() == NULL, "we can only allocate one MemNotifyThread");
5978 _fd = fd;
5979
5980 if (os::create_thread(this, os::os_thread)) {
5981 _memnotify_thread = this;
5982 os::set_priority(this, NearMaxPriority);
5983 os::start_thread(this);
5984 }
5985 }
5986
5987 // Where all the work gets done
5988 //
5989 void MemNotifyThread::run() {
5990 assert(this == memnotify_thread(), "expected the singleton MemNotifyThread");
5991
5992 // Set up the select arguments
5993 fd_set rfds;
5994 if (_fd != -1) {
5995 FD_ZERO(&rfds);
5996 FD_SET(_fd, &rfds);
5997 }
5998
5999 // Now wait for the mem_notify device to wake up
6000 while (1) {
6001 // Wait for the mem_notify device to signal us..
6002 int rc = select(_fd+1, _fd != -1 ? &rfds : NULL, NULL, NULL, NULL);
6003 if (rc == -1) {
6004 perror("select!\n");
6005 break;
6006 } else if (rc) {
6007 //ssize_t free_before = os::available_memory();
6008 //tty->print ("Notified: Free: %dK \n",os::available_memory()/1024);
6009
6010 // The kernel is telling us there is not much memory left...
6011 // try to do something about that
6012
6013 // If we are not already in a GC, try one.
6014 if (!Universe::heap()->is_gc_active()) {
6015 Universe::heap()->collect(GCCause::_allocation_failure);
6016
6017 //ssize_t free_after = os::available_memory();
6018 //tty->print ("Post-Notify: Free: %dK\n",free_after/1024);
6019 //tty->print ("GC freed: %dK\n", (free_after - free_before)/1024);
6020 }
6021 // We might want to do something like the following if we find the GC's are not helping...
6022 // Universe::heap()->size_policy()->set_gc_time_limit_exceeded(true);
6023 }
6024 }
6025 }
6026
6027 //
6028 // See if the /dev/mem_notify device exists, and if so, start a thread to monitor it.
6029 //
6030 void MemNotifyThread::start() {
6031 int fd;
6032 fd = open ("/dev/mem_notify", O_RDONLY, 0);
6033 if (fd < 0) {
6034 return;
6035 }
6036
6037 if (memnotify_thread() == NULL) {
6038 new MemNotifyThread(fd);
6039 }
6040 }
6041
6042 #endif // JAVASE_EMBEDDED
6043
6044
6045 /////////////// Unit tests ///////////////
6046
6047 #ifndef PRODUCT
6048
6049 #define test_log(...) \
6050 do {\
6051 if (VerboseInternalVMTests) { \
6052 tty->print_cr(__VA_ARGS__); \
6053 tty->flush(); \
6054 }\
6055 } while (false)
6056
6057 class TestReserveMemorySpecial : AllStatic {
6058 public:
6059 static void small_page_write(void* addr, size_t size) {
6060 size_t page_size = os::vm_page_size();
6061
6062 char* end = (char*)addr + size;
6063 for (char* p = (char*)addr; p < end; p += page_size) {
6064 *p = 1;
6065 }
6066 }
6067
6068 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6069 if (!UseHugeTLBFS) {
6070 return;
6071 }
6072
6073 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6074
6075 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6076
6077 if (addr != NULL) {
6078 small_page_write(addr, size);
6079
6080 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6081 }
6082 }
6083
6084 static void test_reserve_memory_special_huge_tlbfs_only() {
6085 if (!UseHugeTLBFS) {
6086 return;
6087 }
6088
6089 size_t lp = os::large_page_size();
6090
6091 for (size_t size = lp; size <= lp * 10; size += lp) {
6092 test_reserve_memory_special_huge_tlbfs_only(size);
6093 }
6094 }
6095
6096 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6097 if (!UseHugeTLBFS) {
6098 return;
6099 }
6100
6101 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6102 size, alignment);
6103
6104 assert(size >= os::large_page_size(), "Incorrect input to test");
6105
6106 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6107
6108 if (addr != NULL) {
6109 small_page_write(addr, size);
6110
6111 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6112 }
6113 }
6114
6115 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6116 size_t lp = os::large_page_size();
6117 size_t ag = os::vm_allocation_granularity();
6118
6119 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6120 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6121 }
6122 }
6123
6124 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6125 size_t lp = os::large_page_size();
6126 size_t ag = os::vm_allocation_granularity();
6127
6128 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6129 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6130 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6131 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6132 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
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 + lp / 2);
6135 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6136 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6137 }
6138
6139 static void test_reserve_memory_special_huge_tlbfs() {
6140 if (!UseHugeTLBFS) {
6141 return;
6142 }
6143
6144 test_reserve_memory_special_huge_tlbfs_only();
6145 test_reserve_memory_special_huge_tlbfs_mixed();
6146 }
6147
6148 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6149 if (!UseSHM) {
6150 return;
6151 }
6152
6153 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6154
6155 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6156
6157 if (addr != NULL) {
6158 assert(is_ptr_aligned(addr, alignment), "Check");
6159 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6160
6161 small_page_write(addr, size);
6162
6163 os::Linux::release_memory_special_shm(addr, size);
6164 }
6165 }
6166
6167 static void test_reserve_memory_special_shm() {
6168 size_t lp = os::large_page_size();
6169 size_t ag = os::vm_allocation_granularity();
6170
6171 for (size_t size = ag; size < lp * 3; size += ag) {
6172 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6173 test_reserve_memory_special_shm(size, alignment);
6174 }
6175 }
6176 }
6177
6178 static void test() {
6179 test_reserve_memory_special_huge_tlbfs();
6180 test_reserve_memory_special_shm();
6181 }
6182 };
6183
6184 void TestReserveMemorySpecial_test() {
6185 TestReserveMemorySpecial::test();
6186 }
6187
6188 #endif