1 /*
2 * Copyright (c) 1999, 2015, 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_linux.inline.hpp"
40 #include "os_share_linux.hpp"
41 #include "prims/jniFastGetField.hpp"
42 #include "prims/jvm.h"
43 #include "prims/jvm_misc.hpp"
44 #include "runtime/arguments.hpp"
45 #include "runtime/atomic.inline.hpp"
46 #include "runtime/extendedPC.hpp"
47 #include "runtime/globals.hpp"
48 #include "runtime/interfaceSupport.hpp"
49 #include "runtime/init.hpp"
50 #include "runtime/java.hpp"
51 #include "runtime/javaCalls.hpp"
52 #include "runtime/mutexLocker.hpp"
53 #include "runtime/objectMonitor.hpp"
54 #include "runtime/orderAccess.inline.hpp"
55 #include "runtime/osThread.hpp"
56 #include "runtime/perfMemory.hpp"
57 #include "runtime/sharedRuntime.hpp"
58 #include "runtime/statSampler.hpp"
59 #include "runtime/stubRoutines.hpp"
60 #include "runtime/thread.inline.hpp"
61 #include "runtime/threadCritical.hpp"
62 #include "runtime/timer.hpp"
63 #include "services/attachListener.hpp"
64 #include "services/memTracker.hpp"
65 #include "services/runtimeService.hpp"
66 #include "utilities/decoder.hpp"
67 #include "utilities/defaultStream.hpp"
68 #include "utilities/events.hpp"
69 #include "utilities/elfFile.hpp"
70 #include "utilities/growableArray.hpp"
71 #include "utilities/macros.hpp"
72 #include "utilities/semaphore.hpp"
73 #include "utilities/vmError.hpp"
74
75 // put OS-includes here
76 # include <sys/types.h>
77 # include <sys/mman.h>
78 # include <sys/stat.h>
79 # include <sys/select.h>
80 # include <pthread.h>
81 # include <signal.h>
82 # include <errno.h>
83 # include <dlfcn.h>
84 # include <stdio.h>
85 # include <unistd.h>
86 # include <sys/resource.h>
87 # include <pthread.h>
88 # include <sys/stat.h>
89 # include <sys/time.h>
90 # include <sys/times.h>
91 # include <sys/utsname.h>
92 # include <sys/socket.h>
93 # include <sys/wait.h>
94 # include <pwd.h>
95 # include <poll.h>
96 # include <semaphore.h>
97 # include <fcntl.h>
98 # include <string.h>
99 # include <syscall.h>
100 # include <sys/sysinfo.h>
101 # include <gnu/libc-version.h>
102 # include <sys/ipc.h>
103 # include <sys/shm.h>
104 # include <link.h>
105 # include <stdint.h>
106 # include <inttypes.h>
107 # include <sys/ioctl.h>
108
109 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
110
111 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
112 // getrusage() is prepared to handle the associated failure.
113 #ifndef RUSAGE_THREAD
114 #define RUSAGE_THREAD (1) /* only the calling thread */
115 #endif
116
117 #define MAX_PATH (2 * K)
118
119 #define MAX_SECS 100000000
120
121 // for timer info max values which include all bits
122 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
123
124 #define LARGEPAGES_BIT (1 << 6)
125 ////////////////////////////////////////////////////////////////////////////////
126 // global variables
127 julong os::Linux::_physical_memory = 0;
128
129 address os::Linux::_initial_thread_stack_bottom = NULL;
130 uintptr_t os::Linux::_initial_thread_stack_size = 0;
131
132 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
133 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
134 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
135 Mutex* os::Linux::_createThread_lock = NULL;
136 pthread_t os::Linux::_main_thread;
137 int os::Linux::_page_size = -1;
138 const int os::Linux::_vm_default_page_size = (8 * K);
139 bool os::Linux::_is_floating_stack = false;
140 bool os::Linux::_is_NPTL = false;
141 bool os::Linux::_supports_fast_thread_cpu_time = false;
142 const char * os::Linux::_glibc_version = NULL;
143 const char * os::Linux::_libpthread_version = NULL;
144 pthread_condattr_t os::Linux::_condattr[1];
145
146 static jlong initial_time_count=0;
147
148 static int clock_tics_per_sec = 100;
149
150 // For diagnostics to print a message once. see run_periodic_checks
151 static sigset_t check_signal_done;
152 static bool check_signals = true;
153
154 static pid_t _initial_pid = 0;
155
156 // Signal number used to suspend/resume a thread
157
158 // do not use any signal number less than SIGSEGV, see 4355769
159 static int SR_signum = SIGUSR2;
160 sigset_t SR_sigset;
161
162 // Declarations
163 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
164
165 // utility functions
166
167 static int SR_initialize();
168
169 julong os::available_memory() {
170 return Linux::available_memory();
171 }
172
173 julong os::Linux::available_memory() {
174 // values in struct sysinfo are "unsigned long"
175 struct sysinfo si;
176 sysinfo(&si);
177
178 return (julong)si.freeram * si.mem_unit;
179 }
180
181 julong os::physical_memory() {
182 return Linux::physical_memory();
183 }
184
185 // Return true if user is running as root.
186
187 bool os::have_special_privileges() {
188 static bool init = false;
189 static bool privileges = false;
190 if (!init) {
191 privileges = (getuid() != geteuid()) || (getgid() != getegid());
192 init = true;
193 }
194 return privileges;
195 }
196
197
198 #ifndef SYS_gettid
199 // i386: 224, ia64: 1105, amd64: 186, sparc 143
200 #ifdef __ia64__
201 #define SYS_gettid 1105
202 #else
203 #ifdef __i386__
204 #define SYS_gettid 224
205 #else
206 #ifdef __amd64__
207 #define SYS_gettid 186
208 #else
209 #ifdef __sparc__
210 #define SYS_gettid 143
211 #else
212 #error define gettid for the arch
213 #endif
214 #endif
215 #endif
216 #endif
217 #endif
218
219 // Cpu architecture string
220 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
221
222
223 // pid_t gettid()
224 //
225 // Returns the kernel thread id of the currently running thread. Kernel
226 // thread id is used to access /proc.
227 //
228 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
229 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
230 //
231 pid_t os::Linux::gettid() {
232 int rslt = syscall(SYS_gettid);
233 if (rslt == -1) {
234 // old kernel, no NPTL support
235 return getpid();
236 } else {
237 return (pid_t)rslt;
238 }
239 }
240
241 // Most versions of linux have a bug where the number of processors are
242 // determined by looking at the /proc file system. In a chroot environment,
243 // the system call returns 1. This causes the VM to act as if it is
244 // a single processor and elide locking (see is_MP() call).
245 static bool unsafe_chroot_detected = false;
246 static const char *unstable_chroot_error = "/proc file system not found.\n"
247 "Java may be unstable running multithreaded in a chroot "
248 "environment on Linux when /proc filesystem is not mounted.";
249
250 void os::Linux::initialize_system_info() {
251 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
252 if (processor_count() == 1) {
253 pid_t pid = os::Linux::gettid();
254 char fname[32];
255 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
256 FILE *fp = fopen(fname, "r");
257 if (fp == NULL) {
258 unsafe_chroot_detected = true;
259 } else {
260 fclose(fp);
261 }
262 }
263 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
264 assert(processor_count() > 0, "linux error");
265 }
266
267 void os::init_system_properties_values() {
268 // The next steps are taken in the product version:
269 //
270 // Obtain the JAVA_HOME value from the location of libjvm.so.
271 // This library should be located at:
272 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
273 //
274 // If "/jre/lib/" appears at the right place in the path, then we
275 // assume libjvm.so is installed in a JDK and we use this path.
276 //
277 // Otherwise exit with message: "Could not create the Java virtual machine."
278 //
279 // The following extra steps are taken in the debugging version:
280 //
281 // If "/jre/lib/" does NOT appear at the right place in the path
282 // instead of exit check for $JAVA_HOME environment variable.
283 //
284 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
285 // then we append a fake suffix "hotspot/libjvm.so" to this path so
286 // it looks like libjvm.so is installed there
287 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
288 //
289 // Otherwise exit.
290 //
291 // Important note: if the location of libjvm.so changes this
292 // code needs to be changed accordingly.
293
294 // See ld(1):
295 // The linker uses the following search paths to locate required
296 // shared libraries:
297 // 1: ...
298 // ...
299 // 7: The default directories, normally /lib and /usr/lib.
300 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
301 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
302 #else
303 #define DEFAULT_LIBPATH "/lib:/usr/lib"
304 #endif
305
306 // Base path of extensions installed on the system.
307 #define SYS_EXT_DIR "/usr/java/packages"
308 #define EXTENSIONS_DIR "/lib/ext"
309
310 // Buffer that fits several sprintfs.
311 // Note that the space for the colon and the trailing null are provided
312 // by the nulls included by the sizeof operator.
313 const size_t bufsize =
314 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
315 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
316 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
317
318 // sysclasspath, java_home, dll_dir
319 {
320 char *pslash;
321 os::jvm_path(buf, bufsize);
322
323 // Found the full path to libjvm.so.
324 // Now cut the path to <java_home>/jre if we can.
325 pslash = strrchr(buf, '/');
326 if (pslash != NULL) {
327 *pslash = '\0'; // Get rid of /libjvm.so.
328 }
329 pslash = strrchr(buf, '/');
330 if (pslash != NULL) {
331 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
332 }
333 Arguments::set_dll_dir(buf);
334
335 if (pslash != NULL) {
336 pslash = strrchr(buf, '/');
337 if (pslash != NULL) {
338 *pslash = '\0'; // Get rid of /<arch>.
339 pslash = strrchr(buf, '/');
340 if (pslash != NULL) {
341 *pslash = '\0'; // Get rid of /lib.
342 }
343 }
344 }
345 Arguments::set_java_home(buf);
346 set_boot_path('/', ':');
347 }
348
349 // Where to look for native libraries.
350 //
351 // Note: Due to a legacy implementation, most of the library path
352 // is set in the launcher. This was to accomodate linking restrictions
353 // on legacy Linux implementations (which are no longer supported).
354 // Eventually, all the library path setting will be done here.
355 //
356 // However, to prevent the proliferation of improperly built native
357 // libraries, the new path component /usr/java/packages is added here.
358 // Eventually, all the library path setting will be done here.
359 {
360 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
361 // should always exist (until the legacy problem cited above is
362 // addressed).
363 const char *v = ::getenv("LD_LIBRARY_PATH");
364 const char *v_colon = ":";
365 if (v == NULL) { v = ""; v_colon = ""; }
366 // That's +1 for the colon and +1 for the trailing '\0'.
367 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
368 strlen(v) + 1 +
369 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
370 mtInternal);
371 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
372 Arguments::set_library_path(ld_library_path);
373 FREE_C_HEAP_ARRAY(char, ld_library_path);
374 }
375
376 // Extensions directories.
377 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
378 Arguments::set_ext_dirs(buf);
379
380 FREE_C_HEAP_ARRAY(char, buf);
381
382 #undef DEFAULT_LIBPATH
383 #undef SYS_EXT_DIR
384 #undef EXTENSIONS_DIR
385 }
386
387 ////////////////////////////////////////////////////////////////////////////////
388 // breakpoint support
389
390 void os::breakpoint() {
391 BREAKPOINT;
392 }
393
394 extern "C" void breakpoint() {
395 // use debugger to set breakpoint here
396 }
397
398 ////////////////////////////////////////////////////////////////////////////////
399 // signal support
400
401 debug_only(static bool signal_sets_initialized = false);
402 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
403
404 bool os::Linux::is_sig_ignored(int sig) {
405 struct sigaction oact;
406 sigaction(sig, (struct sigaction*)NULL, &oact);
407 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
408 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
409 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
410 return true;
411 } else {
412 return false;
413 }
414 }
415
416 void os::Linux::signal_sets_init() {
417 // Should also have an assertion stating we are still single-threaded.
418 assert(!signal_sets_initialized, "Already initialized");
419 // Fill in signals that are necessarily unblocked for all threads in
420 // the VM. Currently, we unblock the following signals:
421 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
422 // by -Xrs (=ReduceSignalUsage));
423 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
424 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
425 // the dispositions or masks wrt these signals.
426 // Programs embedding the VM that want to use the above signals for their
427 // own purposes must, at this time, use the "-Xrs" option to prevent
428 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
429 // (See bug 4345157, and other related bugs).
430 // In reality, though, unblocking these signals is really a nop, since
431 // these signals are not blocked by default.
432 sigemptyset(&unblocked_sigs);
433 sigemptyset(&allowdebug_blocked_sigs);
434 sigaddset(&unblocked_sigs, SIGILL);
435 sigaddset(&unblocked_sigs, SIGSEGV);
436 sigaddset(&unblocked_sigs, SIGBUS);
437 sigaddset(&unblocked_sigs, SIGFPE);
438 #if defined(PPC64)
439 sigaddset(&unblocked_sigs, SIGTRAP);
440 #endif
441 sigaddset(&unblocked_sigs, SR_signum);
442
443 if (!ReduceSignalUsage) {
444 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
445 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
446 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
447 }
448 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
449 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
450 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
451 }
452 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
453 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
454 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
455 }
456 }
457 // Fill in signals that are blocked by all but the VM thread.
458 sigemptyset(&vm_sigs);
459 if (!ReduceSignalUsage) {
460 sigaddset(&vm_sigs, BREAK_SIGNAL);
461 }
462 debug_only(signal_sets_initialized = true);
463
464 }
465
466 // These are signals that are unblocked while a thread is running Java.
467 // (For some reason, they get blocked by default.)
468 sigset_t* os::Linux::unblocked_signals() {
469 assert(signal_sets_initialized, "Not initialized");
470 return &unblocked_sigs;
471 }
472
473 // These are the signals that are blocked while a (non-VM) thread is
474 // running Java. Only the VM thread handles these signals.
475 sigset_t* os::Linux::vm_signals() {
476 assert(signal_sets_initialized, "Not initialized");
477 return &vm_sigs;
478 }
479
480 // These are signals that are blocked during cond_wait to allow debugger in
481 sigset_t* os::Linux::allowdebug_blocked_signals() {
482 assert(signal_sets_initialized, "Not initialized");
483 return &allowdebug_blocked_sigs;
484 }
485
486 void os::Linux::hotspot_sigmask(Thread* thread) {
487
488 //Save caller's signal mask before setting VM signal mask
489 sigset_t caller_sigmask;
490 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
491
492 OSThread* osthread = thread->osthread();
493 osthread->set_caller_sigmask(caller_sigmask);
494
495 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
496
497 if (!ReduceSignalUsage) {
498 if (thread->is_VM_thread()) {
499 // Only the VM thread handles BREAK_SIGNAL ...
500 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
501 } else {
502 // ... all other threads block BREAK_SIGNAL
503 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
504 }
505 }
506 }
507
508 //////////////////////////////////////////////////////////////////////////////
509 // detecting pthread library
510
511 void os::Linux::libpthread_init() {
512 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
513 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
514 // generic name for earlier versions.
515 // Define macros here so we can build HotSpot on old systems.
516 #ifndef _CS_GNU_LIBC_VERSION
517 #define _CS_GNU_LIBC_VERSION 2
518 #endif
519 #ifndef _CS_GNU_LIBPTHREAD_VERSION
520 #define _CS_GNU_LIBPTHREAD_VERSION 3
521 #endif
522
523 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
524 if (n > 0) {
525 char *str = (char *)malloc(n, mtInternal);
526 confstr(_CS_GNU_LIBC_VERSION, str, n);
527 os::Linux::set_glibc_version(str);
528 } else {
529 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
530 static char _gnu_libc_version[32];
531 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
532 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
533 os::Linux::set_glibc_version(_gnu_libc_version);
534 }
535
536 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
537 if (n > 0) {
538 char *str = (char *)malloc(n, mtInternal);
539 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
540 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
541 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
542 // is the case. LinuxThreads has a hard limit on max number of threads.
543 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
544 // On the other hand, NPTL does not have such a limit, sysconf()
545 // will return -1 and errno is not changed. Check if it is really NPTL.
546 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
547 strstr(str, "NPTL") &&
548 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
549 free(str);
550 os::Linux::set_libpthread_version("linuxthreads");
551 } else {
552 os::Linux::set_libpthread_version(str);
553 }
554 } else {
555 // glibc before 2.3.2 only has LinuxThreads.
556 os::Linux::set_libpthread_version("linuxthreads");
557 }
558
559 if (strstr(libpthread_version(), "NPTL")) {
560 os::Linux::set_is_NPTL();
561 } else {
562 os::Linux::set_is_LinuxThreads();
563 }
564
565 // LinuxThreads have two flavors: floating-stack mode, which allows variable
566 // stack size; and fixed-stack mode. NPTL is always floating-stack.
567 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
568 os::Linux::set_is_floating_stack();
569 }
570 }
571
572 /////////////////////////////////////////////////////////////////////////////
573 // thread stack
574
575 // Force Linux kernel to expand current thread stack. If "bottom" is close
576 // to the stack guard, caller should block all signals.
577 //
578 // MAP_GROWSDOWN:
579 // A special mmap() flag that is used to implement thread stacks. It tells
580 // kernel that the memory region should extend downwards when needed. This
581 // allows early versions of LinuxThreads to only mmap the first few pages
582 // when creating a new thread. Linux kernel will automatically expand thread
583 // stack as needed (on page faults).
584 //
585 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
586 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
587 // region, it's hard to tell if the fault is due to a legitimate stack
588 // access or because of reading/writing non-exist memory (e.g. buffer
589 // overrun). As a rule, if the fault happens below current stack pointer,
590 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
591 // application (see Linux kernel fault.c).
592 //
593 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
594 // stack overflow detection.
595 //
596 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
597 // not use this flag. However, the stack of initial thread is not created
598 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
599 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
600 // and then attach the thread to JVM.
601 //
602 // To get around the problem and allow stack banging on Linux, we need to
603 // manually expand thread stack after receiving the SIGSEGV.
604 //
605 // There are two ways to expand thread stack to address "bottom", we used
606 // both of them in JVM before 1.5:
607 // 1. adjust stack pointer first so that it is below "bottom", and then
608 // touch "bottom"
609 // 2. mmap() the page in question
610 //
611 // Now alternate signal stack is gone, it's harder to use 2. For instance,
612 // if current sp is already near the lower end of page 101, and we need to
613 // call mmap() to map page 100, it is possible that part of the mmap() frame
614 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
615 // That will destroy the mmap() frame and cause VM to crash.
616 //
617 // The following code works by adjusting sp first, then accessing the "bottom"
618 // page to force a page fault. Linux kernel will then automatically expand the
619 // stack mapping.
620 //
621 // _expand_stack_to() assumes its frame size is less than page size, which
622 // should always be true if the function is not inlined.
623
624 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
625 #define NOINLINE
626 #else
627 #define NOINLINE __attribute__ ((noinline))
628 #endif
629
630 static void _expand_stack_to(address bottom) NOINLINE;
631
632 static void _expand_stack_to(address bottom) {
633 address sp;
634 size_t size;
635 volatile char *p;
636
637 // Adjust bottom to point to the largest address within the same page, it
638 // gives us a one-page buffer if alloca() allocates slightly more memory.
639 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
640 bottom += os::Linux::page_size() - 1;
641
642 // sp might be slightly above current stack pointer; if that's the case, we
643 // will alloca() a little more space than necessary, which is OK. Don't use
644 // os::current_stack_pointer(), as its result can be slightly below current
645 // stack pointer, causing us to not alloca enough to reach "bottom".
646 sp = (address)&sp;
647
648 if (sp > bottom) {
649 size = sp - bottom;
650 p = (volatile char *)alloca(size);
651 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
652 p[0] = '\0';
653 }
654 }
655
656 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
657 assert(t!=NULL, "just checking");
658 assert(t->osthread()->expanding_stack(), "expand should be set");
659 assert(t->stack_base() != NULL, "stack_base was not initialized");
660
661 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
662 sigset_t mask_all, old_sigset;
663 sigfillset(&mask_all);
664 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
665 _expand_stack_to(addr);
666 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
667 return true;
668 }
669 return false;
670 }
671
672 //////////////////////////////////////////////////////////////////////////////
673 // create new thread
674
675 static address highest_vm_reserved_address();
676
677 // check if it's safe to start a new thread
678 static bool _thread_safety_check(Thread* thread) {
679 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
680 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
681 // Heap is mmap'ed at lower end of memory space. Thread stacks are
682 // allocated (MAP_FIXED) from high address space. Every thread stack
683 // occupies a fixed size slot (usually 2Mbytes, but user can change
684 // it to other values if they rebuild LinuxThreads).
685 //
686 // Problem with MAP_FIXED is that mmap() can still succeed even part of
687 // the memory region has already been mmap'ed. That means if we have too
688 // many threads and/or very large heap, eventually thread stack will
689 // collide with heap.
690 //
691 // Here we try to prevent heap/stack collision by comparing current
692 // stack bottom with the highest address that has been mmap'ed by JVM
693 // plus a safety margin for memory maps created by native code.
694 //
695 // This feature can be disabled by setting ThreadSafetyMargin to 0
696 //
697 if (ThreadSafetyMargin > 0) {
698 address stack_bottom = os::current_stack_base() - os::current_stack_size();
699
700 // not safe if our stack extends below the safety margin
701 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
702 } else {
703 return true;
704 }
705 } else {
706 // Floating stack LinuxThreads or NPTL:
707 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
708 // there's not enough space left, pthread_create() will fail. If we come
709 // here, that means enough space has been reserved for stack.
710 return true;
711 }
712 }
713
714 // Thread start routine for all newly created threads
715 static void *java_start(Thread *thread) {
716 // Try to randomize the cache line index of hot stack frames.
717 // This helps when threads of the same stack traces evict each other's
718 // cache lines. The threads can be either from the same JVM instance, or
719 // from different JVM instances. The benefit is especially true for
720 // processors with hyperthreading technology.
721 static int counter = 0;
722 int pid = os::current_process_id();
723 alloca(((pid ^ counter++) & 7) * 128);
724
725 ThreadLocalStorage::set_thread(thread);
726
727 OSThread* osthread = thread->osthread();
728 Monitor* sync = osthread->startThread_lock();
729
730 // non floating stack LinuxThreads needs extra check, see above
731 if (!_thread_safety_check(thread)) {
732 // notify parent thread
733 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
734 osthread->set_state(ZOMBIE);
735 sync->notify_all();
736 return NULL;
737 }
738
739 // thread_id is kernel thread id (similar to Solaris LWP id)
740 osthread->set_thread_id(os::Linux::gettid());
741
742 if (UseNUMA) {
743 int lgrp_id = os::numa_get_group_id();
744 if (lgrp_id != -1) {
745 thread->set_lgrp_id(lgrp_id);
746 }
747 }
748 // initialize signal mask for this thread
749 os::Linux::hotspot_sigmask(thread);
750
751 // initialize floating point control register
752 os::Linux::init_thread_fpu_state();
753
754 // handshaking with parent thread
755 {
756 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
757
758 // notify parent thread
759 osthread->set_state(INITIALIZED);
760 sync->notify_all();
761
762 // wait until os::start_thread()
763 while (osthread->get_state() == INITIALIZED) {
764 sync->wait(Mutex::_no_safepoint_check_flag);
765 }
766 }
767
768 // call one more level start routine
769 thread->run();
770
771 return 0;
772 }
773
774 bool os::create_thread(Thread* thread, ThreadType thr_type,
775 size_t stack_size) {
776 assert(thread->osthread() == NULL, "caller responsible");
777
778 // Allocate the OSThread object
779 OSThread* osthread = new OSThread(NULL, NULL);
780 if (osthread == NULL) {
781 return false;
782 }
783
784 // set the correct thread state
785 osthread->set_thread_type(thr_type);
786
787 // Initial state is ALLOCATED but not INITIALIZED
788 osthread->set_state(ALLOCATED);
789
790 thread->set_osthread(osthread);
791
792 // init thread attributes
793 pthread_attr_t attr;
794 pthread_attr_init(&attr);
795 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
796
797 // stack size
798 if (os::Linux::supports_variable_stack_size()) {
799 // calculate stack size if it's not specified by caller
800 if (stack_size == 0) {
801 stack_size = os::Linux::default_stack_size(thr_type);
802
803 switch (thr_type) {
804 case os::java_thread:
805 // Java threads use ThreadStackSize which default value can be
806 // changed with the flag -Xss
807 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
808 stack_size = JavaThread::stack_size_at_create();
809 break;
810 case os::compiler_thread:
811 if (CompilerThreadStackSize > 0) {
812 stack_size = (size_t)(CompilerThreadStackSize * K);
813 break;
814 } // else fall through:
815 // use VMThreadStackSize if CompilerThreadStackSize is not defined
816 case os::vm_thread:
817 case os::pgc_thread:
818 case os::cgc_thread:
819 case os::watcher_thread:
820 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
821 break;
822 }
823 }
824
825 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
826 pthread_attr_setstacksize(&attr, stack_size);
827 } else {
828 // let pthread_create() pick the default value.
829 }
830
831 // glibc guard page
832 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
833
834 ThreadState state;
835
836 {
837 // Serialize thread creation if we are running with fixed stack LinuxThreads
838 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
839 if (lock) {
840 os::Linux::createThread_lock()->lock_without_safepoint_check();
841 }
842
843 pthread_t tid;
844 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
845
846 pthread_attr_destroy(&attr);
847
848 if (ret != 0) {
849 if (PrintMiscellaneous && (Verbose || WizardMode)) {
850 perror("pthread_create()");
851 }
852 // Need to clean up stuff we've allocated so far
853 thread->set_osthread(NULL);
854 delete osthread;
855 if (lock) os::Linux::createThread_lock()->unlock();
856 return false;
857 }
858
859 // Store pthread info into the OSThread
860 osthread->set_pthread_id(tid);
861
862 // Wait until child thread is either initialized or aborted
863 {
864 Monitor* sync_with_child = osthread->startThread_lock();
865 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
866 while ((state = osthread->get_state()) == ALLOCATED) {
867 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
868 }
869 }
870
871 if (lock) {
872 os::Linux::createThread_lock()->unlock();
873 }
874 }
875
876 // Aborted due to thread limit being reached
877 if (state == ZOMBIE) {
878 thread->set_osthread(NULL);
879 delete osthread;
880 return false;
881 }
882
883 // The thread is returned suspended (in state INITIALIZED),
884 // and is started higher up in the call chain
885 assert(state == INITIALIZED, "race condition");
886 return true;
887 }
888
889 /////////////////////////////////////////////////////////////////////////////
890 // attach existing thread
891
892 // bootstrap the main thread
893 bool os::create_main_thread(JavaThread* thread) {
894 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
895 return create_attached_thread(thread);
896 }
897
898 bool os::create_attached_thread(JavaThread* thread) {
899 #ifdef ASSERT
900 thread->verify_not_published();
901 #endif
902
903 // Allocate the OSThread object
904 OSThread* osthread = new OSThread(NULL, NULL);
905
906 if (osthread == NULL) {
907 return false;
908 }
909
910 // Store pthread info into the OSThread
911 osthread->set_thread_id(os::Linux::gettid());
912 osthread->set_pthread_id(::pthread_self());
913
914 // initialize floating point control register
915 os::Linux::init_thread_fpu_state();
916
917 // Initial thread state is RUNNABLE
918 osthread->set_state(RUNNABLE);
919
920 thread->set_osthread(osthread);
921
922 if (UseNUMA) {
923 int lgrp_id = os::numa_get_group_id();
924 if (lgrp_id != -1) {
925 thread->set_lgrp_id(lgrp_id);
926 }
927 }
928
929 if (os::Linux::is_initial_thread()) {
930 // If current thread is initial thread, its stack is mapped on demand,
931 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
932 // the entire stack region to avoid SEGV in stack banging.
933 // It is also useful to get around the heap-stack-gap problem on SuSE
934 // kernel (see 4821821 for details). We first expand stack to the top
935 // of yellow zone, then enable stack yellow zone (order is significant,
936 // enabling yellow zone first will crash JVM on SuSE Linux), so there
937 // is no gap between the last two virtual memory regions.
938
939 JavaThread *jt = (JavaThread *)thread;
940 address addr = jt->stack_yellow_zone_base();
941 assert(addr != NULL, "initialization problem?");
942 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
943
944 osthread->set_expanding_stack();
945 os::Linux::manually_expand_stack(jt, addr);
946 osthread->clear_expanding_stack();
947 }
948
949 // initialize signal mask for this thread
950 // and save the caller's signal mask
951 os::Linux::hotspot_sigmask(thread);
952
953 return true;
954 }
955
956 void os::pd_start_thread(Thread* thread) {
957 OSThread * osthread = thread->osthread();
958 assert(osthread->get_state() != INITIALIZED, "just checking");
959 Monitor* sync_with_child = osthread->startThread_lock();
960 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
961 sync_with_child->notify();
962 }
963
964 // Free Linux resources related to the OSThread
965 void os::free_thread(OSThread* osthread) {
966 assert(osthread != NULL, "osthread not set");
967
968 if (Thread::current()->osthread() == osthread) {
969 // Restore caller's signal mask
970 sigset_t sigmask = osthread->caller_sigmask();
971 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
972 }
973
974 delete osthread;
975 }
976
977 //////////////////////////////////////////////////////////////////////////////
978 // thread local storage
979
980 // Restore the thread pointer if the destructor is called. This is in case
981 // someone from JNI code sets up a destructor with pthread_key_create to run
982 // detachCurrentThread on thread death. Unless we restore the thread pointer we
983 // will hang or crash. When detachCurrentThread is called the key will be set
984 // to null and we will not be called again. If detachCurrentThread is never
985 // called we could loop forever depending on the pthread implementation.
986 static void restore_thread_pointer(void* p) {
987 Thread* thread = (Thread*) p;
988 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
989 }
990
991 int os::allocate_thread_local_storage() {
992 pthread_key_t key;
993 int rslt = pthread_key_create(&key, restore_thread_pointer);
994 assert(rslt == 0, "cannot allocate thread local storage");
995 return (int)key;
996 }
997
998 // Note: This is currently not used by VM, as we don't destroy TLS key
999 // on VM exit.
1000 void os::free_thread_local_storage(int index) {
1001 int rslt = pthread_key_delete((pthread_key_t)index);
1002 assert(rslt == 0, "invalid index");
1003 }
1004
1005 void os::thread_local_storage_at_put(int index, void* value) {
1006 int rslt = pthread_setspecific((pthread_key_t)index, value);
1007 assert(rslt == 0, "pthread_setspecific failed");
1008 }
1009
1010 extern "C" Thread* get_thread() {
1011 return ThreadLocalStorage::thread();
1012 }
1013
1014 //////////////////////////////////////////////////////////////////////////////
1015 // initial thread
1016
1017 // Check if current thread is the initial thread, similar to Solaris thr_main.
1018 bool os::Linux::is_initial_thread(void) {
1019 char dummy;
1020 // If called before init complete, thread stack bottom will be null.
1021 // Can be called if fatal error occurs before initialization.
1022 if (initial_thread_stack_bottom() == NULL) return false;
1023 assert(initial_thread_stack_bottom() != NULL &&
1024 initial_thread_stack_size() != 0,
1025 "os::init did not locate initial thread's stack region");
1026 if ((address)&dummy >= initial_thread_stack_bottom() &&
1027 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1028 return true;
1029 } else {
1030 return false;
1031 }
1032 }
1033
1034 // Find the virtual memory area that contains addr
1035 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1036 FILE *fp = fopen("/proc/self/maps", "r");
1037 if (fp) {
1038 address low, high;
1039 while (!feof(fp)) {
1040 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1041 if (low <= addr && addr < high) {
1042 if (vma_low) *vma_low = low;
1043 if (vma_high) *vma_high = high;
1044 fclose(fp);
1045 return true;
1046 }
1047 }
1048 for (;;) {
1049 int ch = fgetc(fp);
1050 if (ch == EOF || ch == (int)'\n') break;
1051 }
1052 }
1053 fclose(fp);
1054 }
1055 return false;
1056 }
1057
1058 // Locate initial thread stack. This special handling of initial thread stack
1059 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1060 // bogus value for initial thread.
1061 void os::Linux::capture_initial_stack(size_t max_size) {
1062 // stack size is the easy part, get it from RLIMIT_STACK
1063 size_t stack_size;
1064 struct rlimit rlim;
1065 getrlimit(RLIMIT_STACK, &rlim);
1066 stack_size = rlim.rlim_cur;
1067
1068 // 6308388: a bug in ld.so will relocate its own .data section to the
1069 // lower end of primordial stack; reduce ulimit -s value a little bit
1070 // so we won't install guard page on ld.so's data section.
1071 stack_size -= 2 * page_size();
1072
1073 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1074 // 7.1, in both cases we will get 2G in return value.
1075 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1076 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1077 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1078 // in case other parts in glibc still assumes 2M max stack size.
1079 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1080 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1081 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1082 stack_size = 2 * K * K IA64_ONLY(*2);
1083 }
1084 // Try to figure out where the stack base (top) is. This is harder.
1085 //
1086 // When an application is started, glibc saves the initial stack pointer in
1087 // a global variable "__libc_stack_end", which is then used by system
1088 // libraries. __libc_stack_end should be pretty close to stack top. The
1089 // variable is available since the very early days. However, because it is
1090 // a private interface, it could disappear in the future.
1091 //
1092 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1093 // to __libc_stack_end, it is very close to stack top, but isn't the real
1094 // stack top. Note that /proc may not exist if VM is running as a chroot
1095 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1096 // /proc/<pid>/stat could change in the future (though unlikely).
1097 //
1098 // We try __libc_stack_end first. If that doesn't work, look for
1099 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1100 // as a hint, which should work well in most cases.
1101
1102 uintptr_t stack_start;
1103
1104 // try __libc_stack_end first
1105 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1106 if (p && *p) {
1107 stack_start = *p;
1108 } else {
1109 // see if we can get the start_stack field from /proc/self/stat
1110 FILE *fp;
1111 int pid;
1112 char state;
1113 int ppid;
1114 int pgrp;
1115 int session;
1116 int nr;
1117 int tpgrp;
1118 unsigned long flags;
1119 unsigned long minflt;
1120 unsigned long cminflt;
1121 unsigned long majflt;
1122 unsigned long cmajflt;
1123 unsigned long utime;
1124 unsigned long stime;
1125 long cutime;
1126 long cstime;
1127 long prio;
1128 long nice;
1129 long junk;
1130 long it_real;
1131 uintptr_t start;
1132 uintptr_t vsize;
1133 intptr_t rss;
1134 uintptr_t rsslim;
1135 uintptr_t scodes;
1136 uintptr_t ecode;
1137 int i;
1138
1139 // Figure what the primordial thread stack base is. Code is inspired
1140 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1141 // followed by command name surrounded by parentheses, state, etc.
1142 char stat[2048];
1143 int statlen;
1144
1145 fp = fopen("/proc/self/stat", "r");
1146 if (fp) {
1147 statlen = fread(stat, 1, 2047, fp);
1148 stat[statlen] = '\0';
1149 fclose(fp);
1150
1151 // Skip pid and the command string. Note that we could be dealing with
1152 // weird command names, e.g. user could decide to rename java launcher
1153 // to "java 1.4.2 :)", then the stat file would look like
1154 // 1234 (java 1.4.2 :)) R ... ...
1155 // We don't really need to know the command string, just find the last
1156 // occurrence of ")" and then start parsing from there. See bug 4726580.
1157 char * s = strrchr(stat, ')');
1158
1159 i = 0;
1160 if (s) {
1161 // Skip blank chars
1162 do { s++; } while (s && isspace(*s));
1163
1164 #define _UFM UINTX_FORMAT
1165 #define _DFM INTX_FORMAT
1166
1167 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1168 // 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
1169 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,
1170 &state, // 3 %c
1171 &ppid, // 4 %d
1172 &pgrp, // 5 %d
1173 &session, // 6 %d
1174 &nr, // 7 %d
1175 &tpgrp, // 8 %d
1176 &flags, // 9 %lu
1177 &minflt, // 10 %lu
1178 &cminflt, // 11 %lu
1179 &majflt, // 12 %lu
1180 &cmajflt, // 13 %lu
1181 &utime, // 14 %lu
1182 &stime, // 15 %lu
1183 &cutime, // 16 %ld
1184 &cstime, // 17 %ld
1185 &prio, // 18 %ld
1186 &nice, // 19 %ld
1187 &junk, // 20 %ld
1188 &it_real, // 21 %ld
1189 &start, // 22 UINTX_FORMAT
1190 &vsize, // 23 UINTX_FORMAT
1191 &rss, // 24 INTX_FORMAT
1192 &rsslim, // 25 UINTX_FORMAT
1193 &scodes, // 26 UINTX_FORMAT
1194 &ecode, // 27 UINTX_FORMAT
1195 &stack_start); // 28 UINTX_FORMAT
1196 }
1197
1198 #undef _UFM
1199 #undef _DFM
1200
1201 if (i != 28 - 2) {
1202 assert(false, "Bad conversion from /proc/self/stat");
1203 // product mode - assume we are the initial thread, good luck in the
1204 // embedded case.
1205 warning("Can't detect initial thread stack location - bad conversion");
1206 stack_start = (uintptr_t) &rlim;
1207 }
1208 } else {
1209 // For some reason we can't open /proc/self/stat (for example, running on
1210 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1211 // most cases, so don't abort:
1212 warning("Can't detect initial thread stack location - no /proc/self/stat");
1213 stack_start = (uintptr_t) &rlim;
1214 }
1215 }
1216
1217 // Now we have a pointer (stack_start) very close to the stack top, the
1218 // next thing to do is to figure out the exact location of stack top. We
1219 // can find out the virtual memory area that contains stack_start by
1220 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1221 // and its upper limit is the real stack top. (again, this would fail if
1222 // running inside chroot, because /proc may not exist.)
1223
1224 uintptr_t stack_top;
1225 address low, high;
1226 if (find_vma((address)stack_start, &low, &high)) {
1227 // success, "high" is the true stack top. (ignore "low", because initial
1228 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1229 stack_top = (uintptr_t)high;
1230 } else {
1231 // failed, likely because /proc/self/maps does not exist
1232 warning("Can't detect initial thread stack location - find_vma failed");
1233 // best effort: stack_start is normally within a few pages below the real
1234 // stack top, use it as stack top, and reduce stack size so we won't put
1235 // guard page outside stack.
1236 stack_top = stack_start;
1237 stack_size -= 16 * page_size();
1238 }
1239
1240 // stack_top could be partially down the page so align it
1241 stack_top = align_size_up(stack_top, page_size());
1242
1243 if (max_size && stack_size > max_size) {
1244 _initial_thread_stack_size = max_size;
1245 } else {
1246 _initial_thread_stack_size = stack_size;
1247 }
1248
1249 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1250 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1251 }
1252
1253 ////////////////////////////////////////////////////////////////////////////////
1254 // time support
1255
1256 // Time since start-up in seconds to a fine granularity.
1257 // Used by VMSelfDestructTimer and the MemProfiler.
1258 double os::elapsedTime() {
1259
1260 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1261 }
1262
1263 jlong os::elapsed_counter() {
1264 return javaTimeNanos() - initial_time_count;
1265 }
1266
1267 jlong os::elapsed_frequency() {
1268 return NANOSECS_PER_SEC; // nanosecond resolution
1269 }
1270
1271 bool os::supports_vtime() { return true; }
1272 bool os::enable_vtime() { return false; }
1273 bool os::vtime_enabled() { return false; }
1274
1275 double os::elapsedVTime() {
1276 struct rusage usage;
1277 int retval = getrusage(RUSAGE_THREAD, &usage);
1278 if (retval == 0) {
1279 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);
1280 } else {
1281 // better than nothing, but not much
1282 return elapsedTime();
1283 }
1284 }
1285
1286 jlong os::javaTimeMillis() {
1287 timeval time;
1288 int status = gettimeofday(&time, NULL);
1289 assert(status != -1, "linux error");
1290 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1291 }
1292
1293 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1294 timeval time;
1295 int status = gettimeofday(&time, NULL);
1296 assert(status != -1, "linux error");
1297 seconds = jlong(time.tv_sec);
1298 nanos = jlong(time.tv_usec) * 1000;
1299 }
1300
1301
1302 #ifndef CLOCK_MONOTONIC
1303 #define CLOCK_MONOTONIC (1)
1304 #endif
1305
1306 void os::Linux::clock_init() {
1307 // we do dlopen's in this particular order due to bug in linux
1308 // dynamical loader (see 6348968) leading to crash on exit
1309 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1310 if (handle == NULL) {
1311 handle = dlopen("librt.so", RTLD_LAZY);
1312 }
1313
1314 if (handle) {
1315 int (*clock_getres_func)(clockid_t, struct timespec*) =
1316 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1317 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1318 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1319 if (clock_getres_func && clock_gettime_func) {
1320 // See if monotonic clock is supported by the kernel. Note that some
1321 // early implementations simply return kernel jiffies (updated every
1322 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1323 // for nano time (though the monotonic property is still nice to have).
1324 // It's fixed in newer kernels, however clock_getres() still returns
1325 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1326 // resolution for now. Hopefully as people move to new kernels, this
1327 // won't be a problem.
1328 struct timespec res;
1329 struct timespec tp;
1330 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1331 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1332 // yes, monotonic clock is supported
1333 _clock_gettime = clock_gettime_func;
1334 return;
1335 } else {
1336 // close librt if there is no monotonic clock
1337 dlclose(handle);
1338 }
1339 }
1340 }
1341 warning("No monotonic clock was available - timed services may " \
1342 "be adversely affected if the time-of-day clock changes");
1343 }
1344
1345 #ifndef SYS_clock_getres
1346 #if defined(IA32) || defined(AMD64)
1347 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1348 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1349 #else
1350 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1351 #define sys_clock_getres(x,y) -1
1352 #endif
1353 #else
1354 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1355 #endif
1356
1357 void os::Linux::fast_thread_clock_init() {
1358 if (!UseLinuxPosixThreadCPUClocks) {
1359 return;
1360 }
1361 clockid_t clockid;
1362 struct timespec tp;
1363 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1364 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1365
1366 // Switch to using fast clocks for thread cpu time if
1367 // the sys_clock_getres() returns 0 error code.
1368 // Note, that some kernels may support the current thread
1369 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1370 // returned by the pthread_getcpuclockid().
1371 // If the fast Posix clocks are supported then the sys_clock_getres()
1372 // must return at least tp.tv_sec == 0 which means a resolution
1373 // better than 1 sec. This is extra check for reliability.
1374
1375 if (pthread_getcpuclockid_func &&
1376 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1377 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1378 _supports_fast_thread_cpu_time = true;
1379 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1380 }
1381 }
1382
1383 jlong os::javaTimeNanos() {
1384 if (os::supports_monotonic_clock()) {
1385 struct timespec tp;
1386 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1387 assert(status == 0, "gettime error");
1388 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1389 return result;
1390 } else {
1391 timeval time;
1392 int status = gettimeofday(&time, NULL);
1393 assert(status != -1, "linux error");
1394 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1395 return 1000 * usecs;
1396 }
1397 }
1398
1399 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1400 if (os::supports_monotonic_clock()) {
1401 info_ptr->max_value = ALL_64_BITS;
1402
1403 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1404 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1405 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1406 } else {
1407 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1408 info_ptr->max_value = ALL_64_BITS;
1409
1410 // gettimeofday is a real time clock so it skips
1411 info_ptr->may_skip_backward = true;
1412 info_ptr->may_skip_forward = true;
1413 }
1414
1415 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1416 }
1417
1418 // Return the real, user, and system times in seconds from an
1419 // arbitrary fixed point in the past.
1420 bool os::getTimesSecs(double* process_real_time,
1421 double* process_user_time,
1422 double* process_system_time) {
1423 struct tms ticks;
1424 clock_t real_ticks = times(&ticks);
1425
1426 if (real_ticks == (clock_t) (-1)) {
1427 return false;
1428 } else {
1429 double ticks_per_second = (double) clock_tics_per_sec;
1430 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1431 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1432 *process_real_time = ((double) real_ticks) / ticks_per_second;
1433
1434 return true;
1435 }
1436 }
1437
1438
1439 char * os::local_time_string(char *buf, size_t buflen) {
1440 struct tm t;
1441 time_t long_time;
1442 time(&long_time);
1443 localtime_r(&long_time, &t);
1444 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1445 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1446 t.tm_hour, t.tm_min, t.tm_sec);
1447 return buf;
1448 }
1449
1450 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1451 return localtime_r(clock, res);
1452 }
1453
1454 ////////////////////////////////////////////////////////////////////////////////
1455 // runtime exit support
1456
1457 // Note: os::shutdown() might be called very early during initialization, or
1458 // called from signal handler. Before adding something to os::shutdown(), make
1459 // sure it is async-safe and can handle partially initialized VM.
1460 void os::shutdown() {
1461
1462 // allow PerfMemory to attempt cleanup of any persistent resources
1463 perfMemory_exit();
1464
1465 // needs to remove object in file system
1466 AttachListener::abort();
1467
1468 // flush buffered output, finish log files
1469 ostream_abort();
1470
1471 // Check for abort hook
1472 abort_hook_t abort_hook = Arguments::abort_hook();
1473 if (abort_hook != NULL) {
1474 abort_hook();
1475 }
1476
1477 }
1478
1479 // Note: os::abort() might be called very early during initialization, or
1480 // called from signal handler. Before adding something to os::abort(), make
1481 // sure it is async-safe and can handle partially initialized VM.
1482 void os::abort(bool dump_core) {
1483 abort(dump_core, NULL, NULL);
1484 }
1485
1486 void os::abort(bool dump_core, void* siginfo, void* context) {
1487 os::shutdown();
1488 if (dump_core) {
1489 #ifndef PRODUCT
1490 fdStream out(defaultStream::output_fd());
1491 out.print_raw("Current thread is ");
1492 char buf[16];
1493 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1494 out.print_raw_cr(buf);
1495 out.print_raw_cr("Dumping core ...");
1496 #endif
1497 ::abort(); // dump core
1498 }
1499
1500 ::exit(1);
1501 }
1502
1503 // Die immediately, no exit hook, no abort hook, no cleanup.
1504 void os::die() {
1505 // _exit() on LinuxThreads only kills current thread
1506 ::abort();
1507 }
1508
1509
1510 // This method is a copy of JDK's sysGetLastErrorString
1511 // from src/solaris/hpi/src/system_md.c
1512
1513 size_t os::lasterror(char *buf, size_t len) {
1514 if (errno == 0) return 0;
1515
1516 const char *s = ::strerror(errno);
1517 size_t n = ::strlen(s);
1518 if (n >= len) {
1519 n = len - 1;
1520 }
1521 ::strncpy(buf, s, n);
1522 buf[n] = '\0';
1523 return n;
1524 }
1525
1526 intx os::current_thread_id() { return (intx)pthread_self(); }
1527 int os::current_process_id() {
1528
1529 // Under the old linux thread library, linux gives each thread
1530 // its own process id. Because of this each thread will return
1531 // a different pid if this method were to return the result
1532 // of getpid(2). Linux provides no api that returns the pid
1533 // of the launcher thread for the vm. This implementation
1534 // returns a unique pid, the pid of the launcher thread
1535 // that starts the vm 'process'.
1536
1537 // Under the NPTL, getpid() returns the same pid as the
1538 // launcher thread rather than a unique pid per thread.
1539 // Use gettid() if you want the old pre NPTL behaviour.
1540
1541 // if you are looking for the result of a call to getpid() that
1542 // returns a unique pid for the calling thread, then look at the
1543 // OSThread::thread_id() method in osThread_linux.hpp file
1544
1545 return (int)(_initial_pid ? _initial_pid : getpid());
1546 }
1547
1548 // DLL functions
1549
1550 const char* os::dll_file_extension() { return ".so"; }
1551
1552 // This must be hard coded because it's the system's temporary
1553 // directory not the java application's temp directory, ala java.io.tmpdir.
1554 const char* os::get_temp_directory() { return "/tmp"; }
1555
1556 static bool file_exists(const char* filename) {
1557 struct stat statbuf;
1558 if (filename == NULL || strlen(filename) == 0) {
1559 return false;
1560 }
1561 return os::stat(filename, &statbuf) == 0;
1562 }
1563
1564 bool os::dll_build_name(char* buffer, size_t buflen,
1565 const char* pname, const char* fname) {
1566 bool retval = false;
1567 // Copied from libhpi
1568 const size_t pnamelen = pname ? strlen(pname) : 0;
1569
1570 // Return error on buffer overflow.
1571 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1572 return retval;
1573 }
1574
1575 if (pnamelen == 0) {
1576 snprintf(buffer, buflen, "lib%s.so", fname);
1577 retval = true;
1578 } else if (strchr(pname, *os::path_separator()) != NULL) {
1579 int n;
1580 char** pelements = split_path(pname, &n);
1581 if (pelements == NULL) {
1582 return false;
1583 }
1584 for (int i = 0; i < n; i++) {
1585 // Really shouldn't be NULL, but check can't hurt
1586 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1587 continue; // skip the empty path values
1588 }
1589 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1590 if (file_exists(buffer)) {
1591 retval = true;
1592 break;
1593 }
1594 }
1595 // release the storage
1596 for (int i = 0; i < n; i++) {
1597 if (pelements[i] != NULL) {
1598 FREE_C_HEAP_ARRAY(char, pelements[i]);
1599 }
1600 }
1601 if (pelements != NULL) {
1602 FREE_C_HEAP_ARRAY(char*, pelements);
1603 }
1604 } else {
1605 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1606 retval = true;
1607 }
1608 return retval;
1609 }
1610
1611 // check if addr is inside libjvm.so
1612 bool os::address_is_in_vm(address addr) {
1613 static address libjvm_base_addr;
1614 Dl_info dlinfo;
1615
1616 if (libjvm_base_addr == NULL) {
1617 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1618 libjvm_base_addr = (address)dlinfo.dli_fbase;
1619 }
1620 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1621 }
1622
1623 if (dladdr((void *)addr, &dlinfo) != 0) {
1624 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1625 }
1626
1627 return false;
1628 }
1629
1630 bool os::dll_address_to_function_name(address addr, char *buf,
1631 int buflen, int *offset) {
1632 // buf is not optional, but offset is optional
1633 assert(buf != NULL, "sanity check");
1634
1635 Dl_info dlinfo;
1636
1637 if (dladdr((void*)addr, &dlinfo) != 0) {
1638 // see if we have a matching symbol
1639 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1640 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1641 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1642 }
1643 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1644 return true;
1645 }
1646 // no matching symbol so try for just file info
1647 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1648 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1649 buf, buflen, offset, dlinfo.dli_fname)) {
1650 return true;
1651 }
1652 }
1653 }
1654
1655 buf[0] = '\0';
1656 if (offset != NULL) *offset = -1;
1657 return false;
1658 }
1659
1660 struct _address_to_library_name {
1661 address addr; // input : memory address
1662 size_t buflen; // size of fname
1663 char* fname; // output: library name
1664 address base; // library base addr
1665 };
1666
1667 static int address_to_library_name_callback(struct dl_phdr_info *info,
1668 size_t size, void *data) {
1669 int i;
1670 bool found = false;
1671 address libbase = NULL;
1672 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1673
1674 // iterate through all loadable segments
1675 for (i = 0; i < info->dlpi_phnum; i++) {
1676 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1677 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1678 // base address of a library is the lowest address of its loaded
1679 // segments.
1680 if (libbase == NULL || libbase > segbase) {
1681 libbase = segbase;
1682 }
1683 // see if 'addr' is within current segment
1684 if (segbase <= d->addr &&
1685 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1686 found = true;
1687 }
1688 }
1689 }
1690
1691 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1692 // so dll_address_to_library_name() can fall through to use dladdr() which
1693 // can figure out executable name from argv[0].
1694 if (found && info->dlpi_name && info->dlpi_name[0]) {
1695 d->base = libbase;
1696 if (d->fname) {
1697 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1698 }
1699 return 1;
1700 }
1701 return 0;
1702 }
1703
1704 bool os::dll_address_to_library_name(address addr, char* buf,
1705 int buflen, int* offset) {
1706 // buf is not optional, but offset is optional
1707 assert(buf != NULL, "sanity check");
1708
1709 Dl_info dlinfo;
1710 struct _address_to_library_name data;
1711
1712 // There is a bug in old glibc dladdr() implementation that it could resolve
1713 // to wrong library name if the .so file has a base address != NULL. Here
1714 // we iterate through the program headers of all loaded libraries to find
1715 // out which library 'addr' really belongs to. This workaround can be
1716 // removed once the minimum requirement for glibc is moved to 2.3.x.
1717 data.addr = addr;
1718 data.fname = buf;
1719 data.buflen = buflen;
1720 data.base = NULL;
1721 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1722
1723 if (rslt) {
1724 // buf already contains library name
1725 if (offset) *offset = addr - data.base;
1726 return true;
1727 }
1728 if (dladdr((void*)addr, &dlinfo) != 0) {
1729 if (dlinfo.dli_fname != NULL) {
1730 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1731 }
1732 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1733 *offset = addr - (address)dlinfo.dli_fbase;
1734 }
1735 return true;
1736 }
1737
1738 buf[0] = '\0';
1739 if (offset) *offset = -1;
1740 return false;
1741 }
1742
1743 // Loads .dll/.so and
1744 // in case of error it checks if .dll/.so was built for the
1745 // same architecture as Hotspot is running on
1746
1747
1748 // Remember the stack's state. The Linux dynamic linker will change
1749 // the stack to 'executable' at most once, so we must safepoint only once.
1750 bool os::Linux::_stack_is_executable = false;
1751
1752 // VM operation that loads a library. This is necessary if stack protection
1753 // of the Java stacks can be lost during loading the library. If we
1754 // do not stop the Java threads, they can stack overflow before the stacks
1755 // are protected again.
1756 class VM_LinuxDllLoad: public VM_Operation {
1757 private:
1758 const char *_filename;
1759 char *_ebuf;
1760 int _ebuflen;
1761 void *_lib;
1762 public:
1763 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1764 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1765 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1766 void doit() {
1767 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1768 os::Linux::_stack_is_executable = true;
1769 }
1770 void* loaded_library() { return _lib; }
1771 };
1772
1773 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1774 void * result = NULL;
1775 bool load_attempted = false;
1776
1777 // Check whether the library to load might change execution rights
1778 // of the stack. If they are changed, the protection of the stack
1779 // guard pages will be lost. We need a safepoint to fix this.
1780 //
1781 // See Linux man page execstack(8) for more info.
1782 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1783 ElfFile ef(filename);
1784 if (!ef.specifies_noexecstack()) {
1785 if (!is_init_completed()) {
1786 os::Linux::_stack_is_executable = true;
1787 // This is OK - No Java threads have been created yet, and hence no
1788 // stack guard pages to fix.
1789 //
1790 // This should happen only when you are building JDK7 using a very
1791 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1792 //
1793 // Dynamic loader will make all stacks executable after
1794 // this function returns, and will not do that again.
1795 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1796 } else {
1797 warning("You have loaded library %s which might have disabled stack guard. "
1798 "The VM will try to fix the stack guard now.\n"
1799 "It's highly recommended that you fix the library with "
1800 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1801 filename);
1802
1803 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1804 JavaThread *jt = JavaThread::current();
1805 if (jt->thread_state() != _thread_in_native) {
1806 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1807 // that requires ExecStack. Cannot enter safe point. Let's give up.
1808 warning("Unable to fix stack guard. Giving up.");
1809 } else {
1810 if (!LoadExecStackDllInVMThread) {
1811 // This is for the case where the DLL has an static
1812 // constructor function that executes JNI code. We cannot
1813 // load such DLLs in the VMThread.
1814 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1815 }
1816
1817 ThreadInVMfromNative tiv(jt);
1818 debug_only(VMNativeEntryWrapper vew;)
1819
1820 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1821 VMThread::execute(&op);
1822 if (LoadExecStackDllInVMThread) {
1823 result = op.loaded_library();
1824 }
1825 load_attempted = true;
1826 }
1827 }
1828 }
1829 }
1830
1831 if (!load_attempted) {
1832 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1833 }
1834
1835 if (result != NULL) {
1836 // Successful loading
1837 return result;
1838 }
1839
1840 Elf32_Ehdr elf_head;
1841 int diag_msg_max_length=ebuflen-strlen(ebuf);
1842 char* diag_msg_buf=ebuf+strlen(ebuf);
1843
1844 if (diag_msg_max_length==0) {
1845 // No more space in ebuf for additional diagnostics message
1846 return NULL;
1847 }
1848
1849
1850 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1851
1852 if (file_descriptor < 0) {
1853 // Can't open library, report dlerror() message
1854 return NULL;
1855 }
1856
1857 bool failed_to_read_elf_head=
1858 (sizeof(elf_head)!=
1859 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1860
1861 ::close(file_descriptor);
1862 if (failed_to_read_elf_head) {
1863 // file i/o error - report dlerror() msg
1864 return NULL;
1865 }
1866
1867 typedef struct {
1868 Elf32_Half code; // Actual value as defined in elf.h
1869 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1870 char elf_class; // 32 or 64 bit
1871 char endianess; // MSB or LSB
1872 char* name; // String representation
1873 } arch_t;
1874
1875 #ifndef EM_486
1876 #define EM_486 6 /* Intel 80486 */
1877 #endif
1878 #ifndef EM_AARCH64
1879 #define EM_AARCH64 183 /* ARM AARCH64 */
1880 #endif
1881
1882 static const arch_t arch_array[]={
1883 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1884 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1885 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1886 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1887 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1888 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1889 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1890 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1891 #if defined(VM_LITTLE_ENDIAN)
1892 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1893 #else
1894 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1895 #endif
1896 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1897 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1898 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1899 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1900 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1901 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1902 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1903 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1904 };
1905
1906 #if (defined IA32)
1907 static Elf32_Half running_arch_code=EM_386;
1908 #elif (defined AMD64)
1909 static Elf32_Half running_arch_code=EM_X86_64;
1910 #elif (defined IA64)
1911 static Elf32_Half running_arch_code=EM_IA_64;
1912 #elif (defined __sparc) && (defined _LP64)
1913 static Elf32_Half running_arch_code=EM_SPARCV9;
1914 #elif (defined __sparc) && (!defined _LP64)
1915 static Elf32_Half running_arch_code=EM_SPARC;
1916 #elif (defined __powerpc64__)
1917 static Elf32_Half running_arch_code=EM_PPC64;
1918 #elif (defined __powerpc__)
1919 static Elf32_Half running_arch_code=EM_PPC;
1920 #elif (defined ARM)
1921 static Elf32_Half running_arch_code=EM_ARM;
1922 #elif (defined S390)
1923 static Elf32_Half running_arch_code=EM_S390;
1924 #elif (defined ALPHA)
1925 static Elf32_Half running_arch_code=EM_ALPHA;
1926 #elif (defined MIPSEL)
1927 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1928 #elif (defined PARISC)
1929 static Elf32_Half running_arch_code=EM_PARISC;
1930 #elif (defined MIPS)
1931 static Elf32_Half running_arch_code=EM_MIPS;
1932 #elif (defined M68K)
1933 static Elf32_Half running_arch_code=EM_68K;
1934 #elif (defined AARCH64)
1935 static Elf32_Half running_arch_code=EM_AARCH64;
1936 #else
1937 #error Method os::dll_load requires that one of following is defined:\
1938 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1939 #endif
1940
1941 // Identify compatability class for VM's architecture and library's architecture
1942 // Obtain string descriptions for architectures
1943
1944 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1945 int running_arch_index=-1;
1946
1947 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1948 if (running_arch_code == arch_array[i].code) {
1949 running_arch_index = i;
1950 }
1951 if (lib_arch.code == arch_array[i].code) {
1952 lib_arch.compat_class = arch_array[i].compat_class;
1953 lib_arch.name = arch_array[i].name;
1954 }
1955 }
1956
1957 assert(running_arch_index != -1,
1958 "Didn't find running architecture code (running_arch_code) in arch_array");
1959 if (running_arch_index == -1) {
1960 // Even though running architecture detection failed
1961 // we may still continue with reporting dlerror() message
1962 return NULL;
1963 }
1964
1965 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1966 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1967 return NULL;
1968 }
1969
1970 #ifndef S390
1971 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1972 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1973 return NULL;
1974 }
1975 #endif // !S390
1976
1977 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1978 if (lib_arch.name!=NULL) {
1979 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1980 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1981 lib_arch.name, arch_array[running_arch_index].name);
1982 } else {
1983 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1984 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1985 lib_arch.code,
1986 arch_array[running_arch_index].name);
1987 }
1988 }
1989
1990 return NULL;
1991 }
1992
1993 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1994 int ebuflen) {
1995 void * result = ::dlopen(filename, RTLD_LAZY);
1996 if (result == NULL) {
1997 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1998 ebuf[ebuflen-1] = '\0';
1999 }
2000 return result;
2001 }
2002
2003 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2004 int ebuflen) {
2005 void * result = NULL;
2006 if (LoadExecStackDllInVMThread) {
2007 result = dlopen_helper(filename, ebuf, ebuflen);
2008 }
2009
2010 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2011 // library that requires an executable stack, or which does not have this
2012 // stack attribute set, dlopen changes the stack attribute to executable. The
2013 // read protection of the guard pages gets lost.
2014 //
2015 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2016 // may have been queued at the same time.
2017
2018 if (!_stack_is_executable) {
2019 JavaThread *jt = Threads::first();
2020
2021 while (jt) {
2022 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2023 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2024 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2025 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2026 warning("Attempt to reguard stack yellow zone failed.");
2027 }
2028 }
2029 jt = jt->next();
2030 }
2031 }
2032
2033 return result;
2034 }
2035
2036 void* os::dll_lookup(void* handle, const char* name) {
2037 void* res = dlsym(handle, name);
2038 return res;
2039 }
2040
2041 void* os::get_default_process_handle() {
2042 return (void*)::dlopen(NULL, RTLD_LAZY);
2043 }
2044
2045 static bool _print_ascii_file(const char* filename, outputStream* st) {
2046 int fd = ::open(filename, O_RDONLY);
2047 if (fd == -1) {
2048 return false;
2049 }
2050
2051 char buf[32];
2052 int bytes;
2053 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2054 st->print_raw(buf, bytes);
2055 }
2056
2057 ::close(fd);
2058
2059 return true;
2060 }
2061
2062 void os::print_dll_info(outputStream *st) {
2063 st->print_cr("Dynamic libraries:");
2064
2065 char fname[32];
2066 pid_t pid = os::Linux::gettid();
2067
2068 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2069
2070 if (!_print_ascii_file(fname, st)) {
2071 st->print("Can not get library information for pid = %d\n", pid);
2072 }
2073 }
2074
2075 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2076 FILE *procmapsFile = NULL;
2077
2078 // Open the procfs maps file for the current process
2079 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2080 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2081 char line[PATH_MAX + 100];
2082
2083 // Read line by line from 'file'
2084 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2085 u8 base, top, offset, inode;
2086 char permissions[5];
2087 char device[6];
2088 char name[PATH_MAX + 1];
2089
2090 // Parse fields from line
2091 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2092
2093 // Filter by device id '00:00' so that we only get file system mapped files.
2094 if (strcmp(device, "00:00") != 0) {
2095
2096 // Call callback with the fields of interest
2097 if(callback(name, (address)base, (address)top, param)) {
2098 // Oops abort, callback aborted
2099 fclose(procmapsFile);
2100 return 1;
2101 }
2102 }
2103 }
2104 fclose(procmapsFile);
2105 }
2106 return 0;
2107 }
2108
2109 void os::print_os_info_brief(outputStream* st) {
2110 os::Linux::print_distro_info(st);
2111
2112 os::Posix::print_uname_info(st);
2113
2114 os::Linux::print_libversion_info(st);
2115
2116 }
2117
2118 void os::print_os_info(outputStream* st) {
2119 st->print("OS:");
2120
2121 os::Linux::print_distro_info(st);
2122
2123 os::Posix::print_uname_info(st);
2124
2125 // Print warning if unsafe chroot environment detected
2126 if (unsafe_chroot_detected) {
2127 st->print("WARNING!! ");
2128 st->print_cr("%s", unstable_chroot_error);
2129 }
2130
2131 os::Linux::print_libversion_info(st);
2132
2133 os::Posix::print_rlimit_info(st);
2134
2135 os::Posix::print_load_average(st);
2136
2137 os::Linux::print_full_memory_info(st);
2138 }
2139
2140 // Try to identify popular distros.
2141 // Most Linux distributions have a /etc/XXX-release file, which contains
2142 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2143 // file that also contains the OS version string. Some have more than one
2144 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2145 // /etc/redhat-release.), so the order is important.
2146 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2147 // their own specific XXX-release file as well as a redhat-release file.
2148 // Because of this the XXX-release file needs to be searched for before the
2149 // redhat-release file.
2150 // Since Red Hat has a lsb-release file that is not very descriptive the
2151 // search for redhat-release needs to be before lsb-release.
2152 // Since the lsb-release file is the new standard it needs to be searched
2153 // before the older style release files.
2154 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2155 // next to last resort. The os-release file is a new standard that contains
2156 // distribution information and the system-release file seems to be an old
2157 // standard that has been replaced by the lsb-release and os-release files.
2158 // Searching for the debian_version file is the last resort. It contains
2159 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2160 // "Debian " is printed before the contents of the debian_version file.
2161 void os::Linux::print_distro_info(outputStream* st) {
2162 if (!_print_ascii_file("/etc/oracle-release", st) &&
2163 !_print_ascii_file("/etc/mandriva-release", st) &&
2164 !_print_ascii_file("/etc/mandrake-release", st) &&
2165 !_print_ascii_file("/etc/sun-release", st) &&
2166 !_print_ascii_file("/etc/redhat-release", st) &&
2167 !_print_ascii_file("/etc/lsb-release", st) &&
2168 !_print_ascii_file("/etc/SuSE-release", st) &&
2169 !_print_ascii_file("/etc/turbolinux-release", st) &&
2170 !_print_ascii_file("/etc/gentoo-release", st) &&
2171 !_print_ascii_file("/etc/ltib-release", st) &&
2172 !_print_ascii_file("/etc/angstrom-version", st) &&
2173 !_print_ascii_file("/etc/system-release", st) &&
2174 !_print_ascii_file("/etc/os-release", st)) {
2175
2176 if (file_exists("/etc/debian_version")) {
2177 st->print("Debian ");
2178 _print_ascii_file("/etc/debian_version", st);
2179 } else {
2180 st->print("Linux");
2181 }
2182 }
2183 st->cr();
2184 }
2185
2186 void os::Linux::print_libversion_info(outputStream* st) {
2187 // libc, pthread
2188 st->print("libc:");
2189 st->print("%s ", os::Linux::glibc_version());
2190 st->print("%s ", os::Linux::libpthread_version());
2191 if (os::Linux::is_LinuxThreads()) {
2192 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2193 }
2194 st->cr();
2195 }
2196
2197 void os::Linux::print_full_memory_info(outputStream* st) {
2198 st->print("\n/proc/meminfo:\n");
2199 _print_ascii_file("/proc/meminfo", st);
2200 st->cr();
2201 }
2202
2203 void os::print_memory_info(outputStream* st) {
2204
2205 st->print("Memory:");
2206 st->print(" %dk page", os::vm_page_size()>>10);
2207
2208 // values in struct sysinfo are "unsigned long"
2209 struct sysinfo si;
2210 sysinfo(&si);
2211
2212 st->print(", physical " UINT64_FORMAT "k",
2213 os::physical_memory() >> 10);
2214 st->print("(" UINT64_FORMAT "k free)",
2215 os::available_memory() >> 10);
2216 st->print(", swap " UINT64_FORMAT "k",
2217 ((jlong)si.totalswap * si.mem_unit) >> 10);
2218 st->print("(" UINT64_FORMAT "k free)",
2219 ((jlong)si.freeswap * si.mem_unit) >> 10);
2220 st->cr();
2221 }
2222
2223 void os::pd_print_cpu_info(outputStream* st) {
2224 st->print("\n/proc/cpuinfo:\n");
2225 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2226 st->print(" <Not Available>");
2227 }
2228 st->cr();
2229 }
2230
2231 void os::print_siginfo(outputStream* st, void* siginfo) {
2232 const siginfo_t* si = (const siginfo_t*)siginfo;
2233
2234 os::Posix::print_siginfo_brief(st, si);
2235 #if INCLUDE_CDS
2236 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2237 UseSharedSpaces) {
2238 FileMapInfo* mapinfo = FileMapInfo::current_info();
2239 if (mapinfo->is_in_shared_space(si->si_addr)) {
2240 st->print("\n\nError accessing class data sharing archive." \
2241 " Mapped file inaccessible during execution, " \
2242 " possible disk/network problem.");
2243 }
2244 }
2245 #endif
2246 st->cr();
2247 }
2248
2249
2250 static void print_signal_handler(outputStream* st, int sig,
2251 char* buf, size_t buflen);
2252
2253 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2254 st->print_cr("Signal Handlers:");
2255 print_signal_handler(st, SIGSEGV, buf, buflen);
2256 print_signal_handler(st, SIGBUS , buf, buflen);
2257 print_signal_handler(st, SIGFPE , buf, buflen);
2258 print_signal_handler(st, SIGPIPE, buf, buflen);
2259 print_signal_handler(st, SIGXFSZ, buf, buflen);
2260 print_signal_handler(st, SIGILL , buf, buflen);
2261 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2262 print_signal_handler(st, SR_signum, buf, buflen);
2263 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2264 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2265 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2266 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2267 #if defined(PPC64)
2268 print_signal_handler(st, SIGTRAP, buf, buflen);
2269 #endif
2270 }
2271
2272 static char saved_jvm_path[MAXPATHLEN] = {0};
2273
2274 // Find the full path to the current module, libjvm.so
2275 void os::jvm_path(char *buf, jint buflen) {
2276 // Error checking.
2277 if (buflen < MAXPATHLEN) {
2278 assert(false, "must use a large-enough buffer");
2279 buf[0] = '\0';
2280 return;
2281 }
2282 // Lazy resolve the path to current module.
2283 if (saved_jvm_path[0] != 0) {
2284 strcpy(buf, saved_jvm_path);
2285 return;
2286 }
2287
2288 char dli_fname[MAXPATHLEN];
2289 bool ret = dll_address_to_library_name(
2290 CAST_FROM_FN_PTR(address, os::jvm_path),
2291 dli_fname, sizeof(dli_fname), NULL);
2292 assert(ret, "cannot locate libjvm");
2293 char *rp = NULL;
2294 if (ret && dli_fname[0] != '\0') {
2295 rp = realpath(dli_fname, buf);
2296 }
2297 if (rp == NULL) {
2298 return;
2299 }
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 if (p == NULL) {
2325 return;
2326 }
2327 assert(strstr(p, "/libjvm") == p, "invalid library name");
2328
2329 rp = realpath(java_home_var, buf);
2330 if (rp == NULL) {
2331 return;
2332 }
2333
2334 // determine if this is a legacy image or modules image
2335 // modules image doesn't have "jre" subdirectory
2336 len = strlen(buf);
2337 assert(len < buflen, "Ran out of buffer room");
2338 jrelib_p = buf + len;
2339 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2340 if (0 != access(buf, F_OK)) {
2341 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2342 }
2343
2344 if (0 == access(buf, F_OK)) {
2345 // Use current module name "libjvm.so"
2346 len = strlen(buf);
2347 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2348 } else {
2349 // Go back to path of .so
2350 rp = realpath(dli_fname, buf);
2351 if (rp == NULL) {
2352 return;
2353 }
2354 }
2355 }
2356 }
2357 }
2358
2359 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2360 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2361 }
2362
2363 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2364 // no prefix required, not even "_"
2365 }
2366
2367 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2368 // no suffix required
2369 }
2370
2371 ////////////////////////////////////////////////////////////////////////////////
2372 // sun.misc.Signal support
2373
2374 static volatile jint sigint_count = 0;
2375
2376 static void UserHandler(int sig, void *siginfo, void *context) {
2377 // 4511530 - sem_post is serialized and handled by the manager thread. When
2378 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2379 // don't want to flood the manager thread with sem_post requests.
2380 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2381 return;
2382 }
2383
2384 // Ctrl-C is pressed during error reporting, likely because the error
2385 // handler fails to abort. Let VM die immediately.
2386 if (sig == SIGINT && is_error_reported()) {
2387 os::die();
2388 }
2389
2390 os::signal_notify(sig);
2391 }
2392
2393 void* os::user_handler() {
2394 return CAST_FROM_FN_PTR(void*, UserHandler);
2395 }
2396
2397 Semaphore::Semaphore(uint value, uint max) {
2398 guarantee(value <= max, "value lower than max");
2399 guarantee(max == Semaphore::NoMaxCount || max <= SEM_VALUE_MAX, "Max value set too high");
2400
2401 int ret = sem_init(&_semaphore, 0, value);
2402 guarantee(ret == 0, "Failed to initialize semaphore");
2403 }
2404
2405 Semaphore::~Semaphore() {
2406 sem_destroy(&_semaphore);
2407 }
2408
2409 void Semaphore::signal() {
2410 int ret = sem_post(&_semaphore);
2411 guarantee(ret == 0, err_msg("sem_post failed: %d", ret));
2412 }
2413
2414 void Semaphore::signal(uint count) {
2415 for (uint i = 0; i < count; i++) {
2416 signal();
2417 }
2418 }
2419
2420 void Semaphore::wait() {
2421 while (sem_wait(&_semaphore) == -1 && errno == EINTR) {
2422 // Retry if the wait was interrupted by a signal.
2423 }
2424 }
2425
2426 bool Semaphore::trywait() {
2427 return sem_trywait(&_semaphore) == 0;
2428 }
2429
2430 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2431
2432 struct timespec ts;
2433 // Semaphore's are always associated with CLOCK_REALTIME
2434 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2435 // see unpackTime for discussion on overflow checking
2436 if (sec >= MAX_SECS) {
2437 ts.tv_sec += MAX_SECS;
2438 ts.tv_nsec = 0;
2439 } else {
2440 ts.tv_sec += sec;
2441 ts.tv_nsec += nsec;
2442 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2443 ts.tv_nsec -= NANOSECS_PER_SEC;
2444 ++ts.tv_sec; // note: this must be <= max_secs
2445 }
2446 }
2447
2448 while (1) {
2449 int result = sem_timedwait(&_semaphore, &ts);
2450 if (result == 0) {
2451 return true;
2452 } else if (errno == EINTR) {
2453 continue;
2454 } else if (errno == ETIMEDOUT) {
2455 return false;
2456 } else {
2457 return false;
2458 }
2459 }
2460 }
2461
2462 extern "C" {
2463 typedef void (*sa_handler_t)(int);
2464 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2465 }
2466
2467 void* os::signal(int signal_number, void* handler) {
2468 struct sigaction sigAct, oldSigAct;
2469
2470 sigfillset(&(sigAct.sa_mask));
2471 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2472 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2473
2474 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2475 // -1 means registration failed
2476 return (void *)-1;
2477 }
2478
2479 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2480 }
2481
2482 void os::signal_raise(int signal_number) {
2483 ::raise(signal_number);
2484 }
2485
2486 // The following code is moved from os.cpp for making this
2487 // code platform specific, which it is by its very nature.
2488
2489 // Will be modified when max signal is changed to be dynamic
2490 int os::sigexitnum_pd() {
2491 return NSIG;
2492 }
2493
2494 // a counter for each possible signal value
2495 static volatile jint pending_signals[NSIG+1] = { 0 };
2496
2497 // Linux(POSIX) specific hand shaking semaphore.
2498 static sem_t sig_sem;
2499 static Semaphore sr_semaphore;
2500
2501 void os::signal_init_pd() {
2502 // Initialize signal structures
2503 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2504
2505 // Initialize signal semaphore
2506 ::sem_init(&sig_sem, 0, 0);
2507 }
2508
2509 void os::signal_notify(int sig) {
2510 Atomic::inc(&pending_signals[sig]);
2511 ::sem_post(&sig_sem);
2512 }
2513
2514 static int check_pending_signals(bool wait) {
2515 Atomic::store(0, &sigint_count);
2516 for (;;) {
2517 for (int i = 0; i < NSIG + 1; i++) {
2518 jint n = pending_signals[i];
2519 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2520 return i;
2521 }
2522 }
2523 if (!wait) {
2524 return -1;
2525 }
2526 JavaThread *thread = JavaThread::current();
2527 ThreadBlockInVM tbivm(thread);
2528
2529 bool threadIsSuspended;
2530 do {
2531 thread->set_suspend_equivalent();
2532 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2533 ::sem_wait(&sig_sem);
2534
2535 // were we externally suspended while we were waiting?
2536 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2537 if (threadIsSuspended) {
2538 // The semaphore has been incremented, but while we were waiting
2539 // another thread suspended us. We don't want to continue running
2540 // while suspended because that would surprise the thread that
2541 // suspended us.
2542 ::sem_post(&sig_sem);
2543
2544 thread->java_suspend_self();
2545 }
2546 } while (threadIsSuspended);
2547 }
2548 }
2549
2550 int os::signal_lookup() {
2551 return check_pending_signals(false);
2552 }
2553
2554 int os::signal_wait() {
2555 return check_pending_signals(true);
2556 }
2557
2558 ////////////////////////////////////////////////////////////////////////////////
2559 // Virtual Memory
2560
2561 int os::vm_page_size() {
2562 // Seems redundant as all get out
2563 assert(os::Linux::page_size() != -1, "must call os::init");
2564 return os::Linux::page_size();
2565 }
2566
2567 // Solaris allocates memory by pages.
2568 int os::vm_allocation_granularity() {
2569 assert(os::Linux::page_size() != -1, "must call os::init");
2570 return os::Linux::page_size();
2571 }
2572
2573 // Rationale behind this function:
2574 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2575 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2576 // samples for JITted code. Here we create private executable mapping over the code cache
2577 // and then we can use standard (well, almost, as mapping can change) way to provide
2578 // info for the reporting script by storing timestamp and location of symbol
2579 void linux_wrap_code(char* base, size_t size) {
2580 static volatile jint cnt = 0;
2581
2582 if (!UseOprofile) {
2583 return;
2584 }
2585
2586 char buf[PATH_MAX+1];
2587 int num = Atomic::add(1, &cnt);
2588
2589 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2590 os::get_temp_directory(), os::current_process_id(), num);
2591 unlink(buf);
2592
2593 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2594
2595 if (fd != -1) {
2596 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2597 if (rv != (off_t)-1) {
2598 if (::write(fd, "", 1) == 1) {
2599 mmap(base, size,
2600 PROT_READ|PROT_WRITE|PROT_EXEC,
2601 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2602 }
2603 }
2604 ::close(fd);
2605 unlink(buf);
2606 }
2607 }
2608
2609 static bool recoverable_mmap_error(int err) {
2610 // See if the error is one we can let the caller handle. This
2611 // list of errno values comes from JBS-6843484. I can't find a
2612 // Linux man page that documents this specific set of errno
2613 // values so while this list currently matches Solaris, it may
2614 // change as we gain experience with this failure mode.
2615 switch (err) {
2616 case EBADF:
2617 case EINVAL:
2618 case ENOTSUP:
2619 // let the caller deal with these errors
2620 return true;
2621
2622 default:
2623 // Any remaining errors on this OS can cause our reserved mapping
2624 // to be lost. That can cause confusion where different data
2625 // structures think they have the same memory mapped. The worst
2626 // scenario is if both the VM and a library think they have the
2627 // same memory mapped.
2628 return false;
2629 }
2630 }
2631
2632 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2633 int err) {
2634 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2635 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2636 strerror(err), err);
2637 }
2638
2639 static void warn_fail_commit_memory(char* addr, size_t size,
2640 size_t alignment_hint, bool exec,
2641 int err) {
2642 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2643 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2644 alignment_hint, exec, strerror(err), err);
2645 }
2646
2647 // NOTE: Linux kernel does not really reserve the pages for us.
2648 // All it does is to check if there are enough free pages
2649 // left at the time of mmap(). This could be a potential
2650 // problem.
2651 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2652 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2653 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2654 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2655 if (res != (uintptr_t) MAP_FAILED) {
2656 if (UseNUMAInterleaving) {
2657 numa_make_global(addr, size);
2658 }
2659 return 0;
2660 }
2661
2662 int err = errno; // save errno from mmap() call above
2663
2664 if (!recoverable_mmap_error(err)) {
2665 warn_fail_commit_memory(addr, size, exec, err);
2666 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2667 }
2668
2669 return err;
2670 }
2671
2672 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2673 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2674 }
2675
2676 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2677 const char* mesg) {
2678 assert(mesg != NULL, "mesg must be specified");
2679 int err = os::Linux::commit_memory_impl(addr, size, exec);
2680 if (err != 0) {
2681 // the caller wants all commit errors to exit with the specified mesg:
2682 warn_fail_commit_memory(addr, size, exec, err);
2683 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2684 }
2685 }
2686
2687 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2688 #ifndef MAP_HUGETLB
2689 #define MAP_HUGETLB 0x40000
2690 #endif
2691
2692 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2693 #ifndef MADV_HUGEPAGE
2694 #define MADV_HUGEPAGE 14
2695 #endif
2696
2697 int os::Linux::commit_memory_impl(char* addr, size_t size,
2698 size_t alignment_hint, bool exec) {
2699 int err = os::Linux::commit_memory_impl(addr, size, exec);
2700 if (err == 0) {
2701 realign_memory(addr, size, alignment_hint);
2702 }
2703 return err;
2704 }
2705
2706 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2707 bool exec) {
2708 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2709 }
2710
2711 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2712 size_t alignment_hint, bool exec,
2713 const char* mesg) {
2714 assert(mesg != NULL, "mesg must be specified");
2715 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2716 if (err != 0) {
2717 // the caller wants all commit errors to exit with the specified mesg:
2718 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2719 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2720 }
2721 }
2722
2723 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2724 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2725 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2726 // be supported or the memory may already be backed by huge pages.
2727 ::madvise(addr, bytes, MADV_HUGEPAGE);
2728 }
2729 }
2730
2731 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2732 // This method works by doing an mmap over an existing mmaping and effectively discarding
2733 // the existing pages. However it won't work for SHM-based large pages that cannot be
2734 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2735 // small pages on top of the SHM segment. This method always works for small pages, so we
2736 // allow that in any case.
2737 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2738 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2739 }
2740 }
2741
2742 void os::numa_make_global(char *addr, size_t bytes) {
2743 Linux::numa_interleave_memory(addr, bytes);
2744 }
2745
2746 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2747 // bind policy to MPOL_PREFERRED for the current thread.
2748 #define USE_MPOL_PREFERRED 0
2749
2750 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2751 // To make NUMA and large pages more robust when both enabled, we need to ease
2752 // the requirements on where the memory should be allocated. MPOL_BIND is the
2753 // default policy and it will force memory to be allocated on the specified
2754 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2755 // the specified node, but will not force it. Using this policy will prevent
2756 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2757 // free large pages.
2758 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2759 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2760 }
2761
2762 bool os::numa_topology_changed() { return false; }
2763
2764 size_t os::numa_get_groups_num() {
2765 int max_node = Linux::numa_max_node();
2766 return max_node > 0 ? max_node + 1 : 1;
2767 }
2768
2769 int os::numa_get_group_id() {
2770 int cpu_id = Linux::sched_getcpu();
2771 if (cpu_id != -1) {
2772 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2773 if (lgrp_id != -1) {
2774 return lgrp_id;
2775 }
2776 }
2777 return 0;
2778 }
2779
2780 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2781 for (size_t i = 0; i < size; i++) {
2782 ids[i] = i;
2783 }
2784 return size;
2785 }
2786
2787 bool os::get_page_info(char *start, page_info* info) {
2788 return false;
2789 }
2790
2791 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2792 page_info* page_found) {
2793 return end;
2794 }
2795
2796
2797 int os::Linux::sched_getcpu_syscall(void) {
2798 unsigned int cpu;
2799 int retval = -1;
2800
2801 #if defined(IA32)
2802 #ifndef SYS_getcpu
2803 #define SYS_getcpu 318
2804 #endif
2805 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2806 #elif defined(AMD64)
2807 // Unfortunately we have to bring all these macros here from vsyscall.h
2808 // to be able to compile on old linuxes.
2809 #define __NR_vgetcpu 2
2810 #define VSYSCALL_START (-10UL << 20)
2811 #define VSYSCALL_SIZE 1024
2812 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2813 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2814 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2815 retval = vgetcpu(&cpu, NULL, NULL);
2816 #endif
2817
2818 return (retval == -1) ? retval : cpu;
2819 }
2820
2821 // Something to do with the numa-aware allocator needs these symbols
2822 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2823 extern "C" JNIEXPORT void numa_error(char *where) { }
2824 extern "C" JNIEXPORT int fork1() { return fork(); }
2825
2826
2827 // If we are running with libnuma version > 2, then we should
2828 // be trying to use symbols with versions 1.1
2829 // If we are running with earlier version, which did not have symbol versions,
2830 // we should use the base version.
2831 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2832 void *f = dlvsym(handle, name, "libnuma_1.1");
2833 if (f == NULL) {
2834 f = dlsym(handle, name);
2835 }
2836 return f;
2837 }
2838
2839 bool os::Linux::libnuma_init() {
2840 // sched_getcpu() should be in libc.
2841 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2842 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2843
2844 // If it's not, try a direct syscall.
2845 if (sched_getcpu() == -1) {
2846 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2847 (void*)&sched_getcpu_syscall));
2848 }
2849
2850 if (sched_getcpu() != -1) { // Does it work?
2851 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2852 if (handle != NULL) {
2853 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2854 libnuma_dlsym(handle, "numa_node_to_cpus")));
2855 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2856 libnuma_dlsym(handle, "numa_max_node")));
2857 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2858 libnuma_dlsym(handle, "numa_available")));
2859 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2860 libnuma_dlsym(handle, "numa_tonode_memory")));
2861 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2862 libnuma_dlsym(handle, "numa_interleave_memory")));
2863 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2864 libnuma_dlsym(handle, "numa_set_bind_policy")));
2865
2866
2867 if (numa_available() != -1) {
2868 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2869 // Create a cpu -> node mapping
2870 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2871 rebuild_cpu_to_node_map();
2872 return true;
2873 }
2874 }
2875 }
2876 return false;
2877 }
2878
2879 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2880 // The table is later used in get_node_by_cpu().
2881 void os::Linux::rebuild_cpu_to_node_map() {
2882 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2883 // in libnuma (possible values are starting from 16,
2884 // and continuing up with every other power of 2, but less
2885 // than the maximum number of CPUs supported by kernel), and
2886 // is a subject to change (in libnuma version 2 the requirements
2887 // are more reasonable) we'll just hardcode the number they use
2888 // in the library.
2889 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2890
2891 size_t cpu_num = os::active_processor_count();
2892 size_t cpu_map_size = NCPUS / BitsPerCLong;
2893 size_t cpu_map_valid_size =
2894 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2895
2896 cpu_to_node()->clear();
2897 cpu_to_node()->at_grow(cpu_num - 1);
2898 size_t node_num = numa_get_groups_num();
2899
2900 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2901 for (size_t i = 0; i < node_num; i++) {
2902 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2903 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2904 if (cpu_map[j] != 0) {
2905 for (size_t k = 0; k < BitsPerCLong; k++) {
2906 if (cpu_map[j] & (1UL << k)) {
2907 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2908 }
2909 }
2910 }
2911 }
2912 }
2913 }
2914 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2915 }
2916
2917 int os::Linux::get_node_by_cpu(int cpu_id) {
2918 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2919 return cpu_to_node()->at(cpu_id);
2920 }
2921 return -1;
2922 }
2923
2924 GrowableArray<int>* os::Linux::_cpu_to_node;
2925 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2926 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2927 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2928 os::Linux::numa_available_func_t os::Linux::_numa_available;
2929 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2930 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2931 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2932 unsigned long* os::Linux::_numa_all_nodes;
2933
2934 bool os::pd_uncommit_memory(char* addr, size_t size) {
2935 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2936 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2937 return res != (uintptr_t) MAP_FAILED;
2938 }
2939
2940 static address get_stack_commited_bottom(address bottom, size_t size) {
2941 address nbot = bottom;
2942 address ntop = bottom + size;
2943
2944 size_t page_sz = os::vm_page_size();
2945 unsigned pages = size / page_sz;
2946
2947 unsigned char vec[1];
2948 unsigned imin = 1, imax = pages + 1, imid;
2949 int mincore_return_value = 0;
2950
2951 assert(imin <= imax, "Unexpected page size");
2952
2953 while (imin < imax) {
2954 imid = (imax + imin) / 2;
2955 nbot = ntop - (imid * page_sz);
2956
2957 // Use a trick with mincore to check whether the page is mapped or not.
2958 // mincore sets vec to 1 if page resides in memory and to 0 if page
2959 // is swapped output but if page we are asking for is unmapped
2960 // it returns -1,ENOMEM
2961 mincore_return_value = mincore(nbot, page_sz, vec);
2962
2963 if (mincore_return_value == -1) {
2964 // Page is not mapped go up
2965 // to find first mapped page
2966 if (errno != EAGAIN) {
2967 assert(errno == ENOMEM, "Unexpected mincore errno");
2968 imax = imid;
2969 }
2970 } else {
2971 // Page is mapped go down
2972 // to find first not mapped page
2973 imin = imid + 1;
2974 }
2975 }
2976
2977 nbot = nbot + page_sz;
2978
2979 // Adjust stack bottom one page up if last checked page is not mapped
2980 if (mincore_return_value == -1) {
2981 nbot = nbot + page_sz;
2982 }
2983
2984 return nbot;
2985 }
2986
2987
2988 // Linux uses a growable mapping for the stack, and if the mapping for
2989 // the stack guard pages is not removed when we detach a thread the
2990 // stack cannot grow beyond the pages where the stack guard was
2991 // mapped. If at some point later in the process the stack expands to
2992 // that point, the Linux kernel cannot expand the stack any further
2993 // because the guard pages are in the way, and a segfault occurs.
2994 //
2995 // However, it's essential not to split the stack region by unmapping
2996 // a region (leaving a hole) that's already part of the stack mapping,
2997 // so if the stack mapping has already grown beyond the guard pages at
2998 // the time we create them, we have to truncate the stack mapping.
2999 // So, we need to know the extent of the stack mapping when
3000 // create_stack_guard_pages() is called.
3001
3002 // We only need this for stacks that are growable: at the time of
3003 // writing thread stacks don't use growable mappings (i.e. those
3004 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3005 // only applies to the main thread.
3006
3007 // If the (growable) stack mapping already extends beyond the point
3008 // where we're going to put our guard pages, truncate the mapping at
3009 // that point by munmap()ping it. This ensures that when we later
3010 // munmap() the guard pages we don't leave a hole in the stack
3011 // mapping. This only affects the main/initial thread
3012
3013 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3014 if (os::Linux::is_initial_thread()) {
3015 // As we manually grow stack up to bottom inside create_attached_thread(),
3016 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3017 // we don't need to do anything special.
3018 // Check it first, before calling heavy function.
3019 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3020 unsigned char vec[1];
3021
3022 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3023 // Fallback to slow path on all errors, including EAGAIN
3024 stack_extent = (uintptr_t) get_stack_commited_bottom(
3025 os::Linux::initial_thread_stack_bottom(),
3026 (size_t)addr - stack_extent);
3027 }
3028
3029 if (stack_extent < (uintptr_t)addr) {
3030 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3031 }
3032 }
3033
3034 return os::commit_memory(addr, size, !ExecMem);
3035 }
3036
3037 // If this is a growable mapping, remove the guard pages entirely by
3038 // munmap()ping them. If not, just call uncommit_memory(). This only
3039 // affects the main/initial thread, but guard against future OS changes
3040 // It's safe to always unmap guard pages for initial thread because we
3041 // always place it right after end of the mapped region
3042
3043 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3044 uintptr_t stack_extent, stack_base;
3045
3046 if (os::Linux::is_initial_thread()) {
3047 return ::munmap(addr, size) == 0;
3048 }
3049
3050 return os::uncommit_memory(addr, size);
3051 }
3052
3053 static address _highest_vm_reserved_address = NULL;
3054
3055 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3056 // at 'requested_addr'. If there are existing memory mappings at the same
3057 // location, however, they will be overwritten. If 'fixed' is false,
3058 // 'requested_addr' is only treated as a hint, the return value may or
3059 // may not start from the requested address. Unlike Linux mmap(), this
3060 // function returns NULL to indicate failure.
3061 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3062 char * addr;
3063 int flags;
3064
3065 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3066 if (fixed) {
3067 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3068 flags |= MAP_FIXED;
3069 }
3070
3071 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3072 // touch an uncommitted page. Otherwise, the read/write might
3073 // succeed if we have enough swap space to back the physical page.
3074 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3075 flags, -1, 0);
3076
3077 if (addr != MAP_FAILED) {
3078 // anon_mmap() should only get called during VM initialization,
3079 // don't need lock (actually we can skip locking even it can be called
3080 // from multiple threads, because _highest_vm_reserved_address is just a
3081 // hint about the upper limit of non-stack memory regions.)
3082 if ((address)addr + bytes > _highest_vm_reserved_address) {
3083 _highest_vm_reserved_address = (address)addr + bytes;
3084 }
3085 }
3086
3087 return addr == MAP_FAILED ? NULL : addr;
3088 }
3089
3090 // Don't update _highest_vm_reserved_address, because there might be memory
3091 // regions above addr + size. If so, releasing a memory region only creates
3092 // a hole in the address space, it doesn't help prevent heap-stack collision.
3093 //
3094 static int anon_munmap(char * addr, size_t size) {
3095 return ::munmap(addr, size) == 0;
3096 }
3097
3098 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3099 size_t alignment_hint) {
3100 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3101 }
3102
3103 bool os::pd_release_memory(char* addr, size_t size) {
3104 return anon_munmap(addr, size);
3105 }
3106
3107 static address highest_vm_reserved_address() {
3108 return _highest_vm_reserved_address;
3109 }
3110
3111 static bool linux_mprotect(char* addr, size_t size, int prot) {
3112 // Linux wants the mprotect address argument to be page aligned.
3113 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3114
3115 // According to SUSv3, mprotect() should only be used with mappings
3116 // established by mmap(), and mmap() always maps whole pages. Unaligned
3117 // 'addr' likely indicates problem in the VM (e.g. trying to change
3118 // protection of malloc'ed or statically allocated memory). Check the
3119 // caller if you hit this assert.
3120 assert(addr == bottom, "sanity check");
3121
3122 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3123 return ::mprotect(bottom, size, prot) == 0;
3124 }
3125
3126 // Set protections specified
3127 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3128 bool is_committed) {
3129 unsigned int p = 0;
3130 switch (prot) {
3131 case MEM_PROT_NONE: p = PROT_NONE; break;
3132 case MEM_PROT_READ: p = PROT_READ; break;
3133 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3134 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3135 default:
3136 ShouldNotReachHere();
3137 }
3138 // is_committed is unused.
3139 return linux_mprotect(addr, bytes, p);
3140 }
3141
3142 bool os::guard_memory(char* addr, size_t size) {
3143 return linux_mprotect(addr, size, PROT_NONE);
3144 }
3145
3146 bool os::unguard_memory(char* addr, size_t size) {
3147 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3148 }
3149
3150 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3151 size_t page_size) {
3152 bool result = false;
3153 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3154 MAP_ANONYMOUS|MAP_PRIVATE,
3155 -1, 0);
3156 if (p != MAP_FAILED) {
3157 void *aligned_p = align_ptr_up(p, page_size);
3158
3159 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3160
3161 munmap(p, page_size * 2);
3162 }
3163
3164 if (warn && !result) {
3165 warning("TransparentHugePages is not supported by the operating system.");
3166 }
3167
3168 return result;
3169 }
3170
3171 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3172 bool result = false;
3173 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3174 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3175 -1, 0);
3176
3177 if (p != MAP_FAILED) {
3178 // We don't know if this really is a huge page or not.
3179 FILE *fp = fopen("/proc/self/maps", "r");
3180 if (fp) {
3181 while (!feof(fp)) {
3182 char chars[257];
3183 long x = 0;
3184 if (fgets(chars, sizeof(chars), fp)) {
3185 if (sscanf(chars, "%lx-%*x", &x) == 1
3186 && x == (long)p) {
3187 if (strstr (chars, "hugepage")) {
3188 result = true;
3189 break;
3190 }
3191 }
3192 }
3193 }
3194 fclose(fp);
3195 }
3196 munmap(p, page_size);
3197 }
3198
3199 if (warn && !result) {
3200 warning("HugeTLBFS is not supported by the operating system.");
3201 }
3202
3203 return result;
3204 }
3205
3206 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3207 //
3208 // From the coredump_filter documentation:
3209 //
3210 // - (bit 0) anonymous private memory
3211 // - (bit 1) anonymous shared memory
3212 // - (bit 2) file-backed private memory
3213 // - (bit 3) file-backed shared memory
3214 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3215 // effective only if the bit 2 is cleared)
3216 // - (bit 5) hugetlb private memory
3217 // - (bit 6) hugetlb shared memory
3218 //
3219 static void set_coredump_filter(void) {
3220 FILE *f;
3221 long cdm;
3222
3223 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3224 return;
3225 }
3226
3227 if (fscanf(f, "%lx", &cdm) != 1) {
3228 fclose(f);
3229 return;
3230 }
3231
3232 rewind(f);
3233
3234 if ((cdm & LARGEPAGES_BIT) == 0) {
3235 cdm |= LARGEPAGES_BIT;
3236 fprintf(f, "%#lx", cdm);
3237 }
3238
3239 fclose(f);
3240 }
3241
3242 // Large page support
3243
3244 static size_t _large_page_size = 0;
3245
3246 size_t os::Linux::find_large_page_size() {
3247 size_t large_page_size = 0;
3248
3249 // large_page_size on Linux is used to round up heap size. x86 uses either
3250 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3251 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3252 // page as large as 256M.
3253 //
3254 // Here we try to figure out page size by parsing /proc/meminfo and looking
3255 // for a line with the following format:
3256 // Hugepagesize: 2048 kB
3257 //
3258 // If we can't determine the value (e.g. /proc is not mounted, or the text
3259 // format has been changed), we'll use the largest page size supported by
3260 // the processor.
3261
3262 #ifndef ZERO
3263 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3264 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3265 #endif // ZERO
3266
3267 FILE *fp = fopen("/proc/meminfo", "r");
3268 if (fp) {
3269 while (!feof(fp)) {
3270 int x = 0;
3271 char buf[16];
3272 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3273 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3274 large_page_size = x * K;
3275 break;
3276 }
3277 } else {
3278 // skip to next line
3279 for (;;) {
3280 int ch = fgetc(fp);
3281 if (ch == EOF || ch == (int)'\n') break;
3282 }
3283 }
3284 }
3285 fclose(fp);
3286 }
3287
3288 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3289 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3290 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3291 proper_unit_for_byte_size(large_page_size));
3292 }
3293
3294 return large_page_size;
3295 }
3296
3297 size_t os::Linux::setup_large_page_size() {
3298 _large_page_size = Linux::find_large_page_size();
3299 const size_t default_page_size = (size_t)Linux::page_size();
3300 if (_large_page_size > default_page_size) {
3301 _page_sizes[0] = _large_page_size;
3302 _page_sizes[1] = default_page_size;
3303 _page_sizes[2] = 0;
3304 }
3305
3306 return _large_page_size;
3307 }
3308
3309 bool os::Linux::setup_large_page_type(size_t page_size) {
3310 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3311 FLAG_IS_DEFAULT(UseSHM) &&
3312 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3313
3314 // The type of large pages has not been specified by the user.
3315
3316 // Try UseHugeTLBFS and then UseSHM.
3317 UseHugeTLBFS = UseSHM = true;
3318
3319 // Don't try UseTransparentHugePages since there are known
3320 // performance issues with it turned on. This might change in the future.
3321 UseTransparentHugePages = false;
3322 }
3323
3324 if (UseTransparentHugePages) {
3325 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3326 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3327 UseHugeTLBFS = false;
3328 UseSHM = false;
3329 return true;
3330 }
3331 UseTransparentHugePages = false;
3332 }
3333
3334 if (UseHugeTLBFS) {
3335 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3336 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3337 UseSHM = false;
3338 return true;
3339 }
3340 UseHugeTLBFS = false;
3341 }
3342
3343 return UseSHM;
3344 }
3345
3346 void os::large_page_init() {
3347 if (!UseLargePages &&
3348 !UseTransparentHugePages &&
3349 !UseHugeTLBFS &&
3350 !UseSHM) {
3351 // Not using large pages.
3352 return;
3353 }
3354
3355 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3356 // The user explicitly turned off large pages.
3357 // Ignore the rest of the large pages flags.
3358 UseTransparentHugePages = false;
3359 UseHugeTLBFS = false;
3360 UseSHM = false;
3361 return;
3362 }
3363
3364 size_t large_page_size = Linux::setup_large_page_size();
3365 UseLargePages = Linux::setup_large_page_type(large_page_size);
3366
3367 set_coredump_filter();
3368 }
3369
3370 #ifndef SHM_HUGETLB
3371 #define SHM_HUGETLB 04000
3372 #endif
3373
3374 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3375 char* req_addr, bool exec) {
3376 // "exec" is passed in but not used. Creating the shared image for
3377 // the code cache doesn't have an SHM_X executable permission to check.
3378 assert(UseLargePages && UseSHM, "only for SHM large pages");
3379 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3380
3381 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3382 return NULL; // Fallback to small pages.
3383 }
3384
3385 key_t key = IPC_PRIVATE;
3386 char *addr;
3387
3388 bool warn_on_failure = UseLargePages &&
3389 (!FLAG_IS_DEFAULT(UseLargePages) ||
3390 !FLAG_IS_DEFAULT(UseSHM) ||
3391 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3392 char msg[128];
3393
3394 // Create a large shared memory region to attach to based on size.
3395 // Currently, size is the total size of the heap
3396 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3397 if (shmid == -1) {
3398 // Possible reasons for shmget failure:
3399 // 1. shmmax is too small for Java heap.
3400 // > check shmmax value: cat /proc/sys/kernel/shmmax
3401 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3402 // 2. not enough large page memory.
3403 // > check available large pages: cat /proc/meminfo
3404 // > increase amount of large pages:
3405 // echo new_value > /proc/sys/vm/nr_hugepages
3406 // Note 1: different Linux may use different name for this property,
3407 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3408 // Note 2: it's possible there's enough physical memory available but
3409 // they are so fragmented after a long run that they can't
3410 // coalesce into large pages. Try to reserve large pages when
3411 // the system is still "fresh".
3412 if (warn_on_failure) {
3413 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3414 warning("%s", msg);
3415 }
3416 return NULL;
3417 }
3418
3419 // attach to the region
3420 addr = (char*)shmat(shmid, req_addr, 0);
3421 int err = errno;
3422
3423 // Remove shmid. If shmat() is successful, the actual shared memory segment
3424 // will be deleted when it's detached by shmdt() or when the process
3425 // terminates. If shmat() is not successful this will remove the shared
3426 // segment immediately.
3427 shmctl(shmid, IPC_RMID, NULL);
3428
3429 if ((intptr_t)addr == -1) {
3430 if (warn_on_failure) {
3431 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3432 warning("%s", msg);
3433 }
3434 return NULL;
3435 }
3436
3437 return addr;
3438 }
3439
3440 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3441 int error) {
3442 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3443
3444 bool warn_on_failure = UseLargePages &&
3445 (!FLAG_IS_DEFAULT(UseLargePages) ||
3446 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3447 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3448
3449 if (warn_on_failure) {
3450 char msg[128];
3451 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3452 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3453 warning("%s", msg);
3454 }
3455 }
3456
3457 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3458 char* req_addr,
3459 bool exec) {
3460 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3461 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3462 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3463
3464 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3465 char* addr = (char*)::mmap(req_addr, bytes, prot,
3466 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3467 -1, 0);
3468
3469 if (addr == MAP_FAILED) {
3470 warn_on_large_pages_failure(req_addr, bytes, errno);
3471 return NULL;
3472 }
3473
3474 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3475
3476 return addr;
3477 }
3478
3479 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3480 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3481 // (req_addr != NULL) or with a given alignment.
3482 // - bytes shall be a multiple of alignment.
3483 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3484 // - alignment sets the alignment at which memory shall be allocated.
3485 // It must be a multiple of allocation granularity.
3486 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3487 // req_addr or NULL.
3488 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3489
3490 size_t extra_size = bytes;
3491 if (req_addr == NULL && alignment > 0) {
3492 extra_size += alignment;
3493 }
3494
3495 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3496 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3497 -1, 0);
3498 if (start == MAP_FAILED) {
3499 start = NULL;
3500 } else {
3501 if (req_addr != NULL) {
3502 if (start != req_addr) {
3503 ::munmap(start, extra_size);
3504 start = NULL;
3505 }
3506 } else {
3507 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3508 char* const end_aligned = start_aligned + bytes;
3509 char* const end = start + extra_size;
3510 if (start_aligned > start) {
3511 ::munmap(start, start_aligned - start);
3512 }
3513 if (end_aligned < end) {
3514 ::munmap(end_aligned, end - end_aligned);
3515 }
3516 start = start_aligned;
3517 }
3518 }
3519 return start;
3520
3521 }
3522
3523 // Reserve memory using mmap(MAP_HUGETLB).
3524 // - bytes shall be a multiple of alignment.
3525 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3526 // - alignment sets the alignment at which memory shall be allocated.
3527 // It must be a multiple of allocation granularity.
3528 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3529 // req_addr or NULL.
3530 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3531 size_t alignment,
3532 char* req_addr,
3533 bool exec) {
3534 size_t large_page_size = os::large_page_size();
3535 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3536
3537 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3538 assert(is_size_aligned(bytes, alignment), "Must be");
3539
3540 // First reserve - but not commit - the address range in small pages.
3541 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3542
3543 if (start == NULL) {
3544 return NULL;
3545 }
3546
3547 assert(is_ptr_aligned(start, alignment), "Must be");
3548
3549 char* end = start + bytes;
3550
3551 // Find the regions of the allocated chunk that can be promoted to large pages.
3552 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3553 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3554
3555 size_t lp_bytes = lp_end - lp_start;
3556
3557 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3558
3559 if (lp_bytes == 0) {
3560 // The mapped region doesn't even span the start and the end of a large page.
3561 // Fall back to allocate a non-special area.
3562 ::munmap(start, end - start);
3563 return NULL;
3564 }
3565
3566 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3567
3568 void* result;
3569
3570 // Commit small-paged leading area.
3571 if (start != lp_start) {
3572 result = ::mmap(start, lp_start - start, prot,
3573 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3574 -1, 0);
3575 if (result == MAP_FAILED) {
3576 ::munmap(lp_start, end - lp_start);
3577 return NULL;
3578 }
3579 }
3580
3581 // Commit large-paged area.
3582 result = ::mmap(lp_start, lp_bytes, prot,
3583 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3584 -1, 0);
3585 if (result == MAP_FAILED) {
3586 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3587 // If the mmap above fails, the large pages region will be unmapped and we
3588 // have regions before and after with small pages. Release these regions.
3589 //
3590 // | mapped | unmapped | mapped |
3591 // ^ ^ ^ ^
3592 // start lp_start lp_end end
3593 //
3594 ::munmap(start, lp_start - start);
3595 ::munmap(lp_end, end - lp_end);
3596 return NULL;
3597 }
3598
3599 // Commit small-paged trailing area.
3600 if (lp_end != end) {
3601 result = ::mmap(lp_end, end - lp_end, prot,
3602 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3603 -1, 0);
3604 if (result == MAP_FAILED) {
3605 ::munmap(start, lp_end - start);
3606 return NULL;
3607 }
3608 }
3609
3610 return start;
3611 }
3612
3613 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3614 size_t alignment,
3615 char* req_addr,
3616 bool exec) {
3617 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3618 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3619 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3620 assert(is_power_of_2(os::large_page_size()), "Must be");
3621 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3622
3623 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3624 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3625 } else {
3626 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3627 }
3628 }
3629
3630 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3631 char* req_addr, bool exec) {
3632 assert(UseLargePages, "only for large pages");
3633
3634 char* addr;
3635 if (UseSHM) {
3636 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3637 } else {
3638 assert(UseHugeTLBFS, "must be");
3639 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3640 }
3641
3642 if (addr != NULL) {
3643 if (UseNUMAInterleaving) {
3644 numa_make_global(addr, bytes);
3645 }
3646
3647 // The memory is committed
3648 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3649 }
3650
3651 return addr;
3652 }
3653
3654 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3655 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3656 return shmdt(base) == 0;
3657 }
3658
3659 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3660 return pd_release_memory(base, bytes);
3661 }
3662
3663 bool os::release_memory_special(char* base, size_t bytes) {
3664 bool res;
3665 if (MemTracker::tracking_level() > NMT_minimal) {
3666 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3667 res = os::Linux::release_memory_special_impl(base, bytes);
3668 if (res) {
3669 tkr.record((address)base, bytes);
3670 }
3671
3672 } else {
3673 res = os::Linux::release_memory_special_impl(base, bytes);
3674 }
3675 return res;
3676 }
3677
3678 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3679 assert(UseLargePages, "only for large pages");
3680 bool res;
3681
3682 if (UseSHM) {
3683 res = os::Linux::release_memory_special_shm(base, bytes);
3684 } else {
3685 assert(UseHugeTLBFS, "must be");
3686 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3687 }
3688 return res;
3689 }
3690
3691 size_t os::large_page_size() {
3692 return _large_page_size;
3693 }
3694
3695 // With SysV SHM the entire memory region must be allocated as shared
3696 // memory.
3697 // HugeTLBFS allows application to commit large page memory on demand.
3698 // However, when committing memory with HugeTLBFS fails, the region
3699 // that was supposed to be committed will lose the old reservation
3700 // and allow other threads to steal that memory region. Because of this
3701 // behavior we can't commit HugeTLBFS memory.
3702 bool os::can_commit_large_page_memory() {
3703 return UseTransparentHugePages;
3704 }
3705
3706 bool os::can_execute_large_page_memory() {
3707 return UseTransparentHugePages || UseHugeTLBFS;
3708 }
3709
3710 // Reserve memory at an arbitrary address, only if that area is
3711 // available (and not reserved for something else).
3712
3713 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3714 const int max_tries = 10;
3715 char* base[max_tries];
3716 size_t size[max_tries];
3717 const size_t gap = 0x000000;
3718
3719 // Assert only that the size is a multiple of the page size, since
3720 // that's all that mmap requires, and since that's all we really know
3721 // about at this low abstraction level. If we need higher alignment,
3722 // we can either pass an alignment to this method or verify alignment
3723 // in one of the methods further up the call chain. See bug 5044738.
3724 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3725
3726 // Repeatedly allocate blocks until the block is allocated at the
3727 // right spot. Give up after max_tries. Note that reserve_memory() will
3728 // automatically update _highest_vm_reserved_address if the call is
3729 // successful. The variable tracks the highest memory address every reserved
3730 // by JVM. It is used to detect heap-stack collision if running with
3731 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3732 // space than needed, it could confuse the collision detecting code. To
3733 // solve the problem, save current _highest_vm_reserved_address and
3734 // calculate the correct value before return.
3735 address old_highest = _highest_vm_reserved_address;
3736
3737 // Linux mmap allows caller to pass an address as hint; give it a try first,
3738 // if kernel honors the hint then we can return immediately.
3739 char * addr = anon_mmap(requested_addr, bytes, false);
3740 if (addr == requested_addr) {
3741 return requested_addr;
3742 }
3743
3744 if (addr != NULL) {
3745 // mmap() is successful but it fails to reserve at the requested address
3746 anon_munmap(addr, bytes);
3747 }
3748
3749 int i;
3750 for (i = 0; i < max_tries; ++i) {
3751 base[i] = reserve_memory(bytes);
3752
3753 if (base[i] != NULL) {
3754 // Is this the block we wanted?
3755 if (base[i] == requested_addr) {
3756 size[i] = bytes;
3757 break;
3758 }
3759
3760 // Does this overlap the block we wanted? Give back the overlapped
3761 // parts and try again.
3762
3763 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3764 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3765 unmap_memory(base[i], top_overlap);
3766 base[i] += top_overlap;
3767 size[i] = bytes - top_overlap;
3768 } else {
3769 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3770 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3771 unmap_memory(requested_addr, bottom_overlap);
3772 size[i] = bytes - bottom_overlap;
3773 } else {
3774 size[i] = bytes;
3775 }
3776 }
3777 }
3778 }
3779
3780 // Give back the unused reserved pieces.
3781
3782 for (int j = 0; j < i; ++j) {
3783 if (base[j] != NULL) {
3784 unmap_memory(base[j], size[j]);
3785 }
3786 }
3787
3788 if (i < max_tries) {
3789 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3790 return requested_addr;
3791 } else {
3792 _highest_vm_reserved_address = old_highest;
3793 return NULL;
3794 }
3795 }
3796
3797 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3798 return ::read(fd, buf, nBytes);
3799 }
3800
3801 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3802 return ::pread(fd, buf, nBytes, offset);
3803 }
3804
3805 // Short sleep, direct OS call.
3806 //
3807 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3808 // sched_yield(2) will actually give up the CPU:
3809 //
3810 // * Alone on this pariticular CPU, keeps running.
3811 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3812 // (pre 2.6.39).
3813 //
3814 // So calling this with 0 is an alternative.
3815 //
3816 void os::naked_short_sleep(jlong ms) {
3817 struct timespec req;
3818
3819 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3820 req.tv_sec = 0;
3821 if (ms > 0) {
3822 req.tv_nsec = (ms % 1000) * 1000000;
3823 } else {
3824 req.tv_nsec = 1;
3825 }
3826
3827 nanosleep(&req, NULL);
3828
3829 return;
3830 }
3831
3832 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3833 void os::infinite_sleep() {
3834 while (true) { // sleep forever ...
3835 ::sleep(100); // ... 100 seconds at a time
3836 }
3837 }
3838
3839 // Used to convert frequent JVM_Yield() to nops
3840 bool os::dont_yield() {
3841 return DontYieldALot;
3842 }
3843
3844 void os::naked_yield() {
3845 sched_yield();
3846 }
3847
3848 ////////////////////////////////////////////////////////////////////////////////
3849 // thread priority support
3850
3851 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3852 // only supports dynamic priority, static priority must be zero. For real-time
3853 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3854 // However, for large multi-threaded applications, SCHED_RR is not only slower
3855 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3856 // of 5 runs - Sep 2005).
3857 //
3858 // The following code actually changes the niceness of kernel-thread/LWP. It
3859 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3860 // not the entire user process, and user level threads are 1:1 mapped to kernel
3861 // threads. It has always been the case, but could change in the future. For
3862 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3863 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3864
3865 int os::java_to_os_priority[CriticalPriority + 1] = {
3866 19, // 0 Entry should never be used
3867
3868 4, // 1 MinPriority
3869 3, // 2
3870 2, // 3
3871
3872 1, // 4
3873 0, // 5 NormPriority
3874 -1, // 6
3875
3876 -2, // 7
3877 -3, // 8
3878 -4, // 9 NearMaxPriority
3879
3880 -5, // 10 MaxPriority
3881
3882 -5 // 11 CriticalPriority
3883 };
3884
3885 static int prio_init() {
3886 if (ThreadPriorityPolicy == 1) {
3887 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3888 // if effective uid is not root. Perhaps, a more elegant way of doing
3889 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3890 if (geteuid() != 0) {
3891 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3892 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3893 }
3894 ThreadPriorityPolicy = 0;
3895 }
3896 }
3897 if (UseCriticalJavaThreadPriority) {
3898 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3899 }
3900 return 0;
3901 }
3902
3903 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3904 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3905
3906 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3907 return (ret == 0) ? OS_OK : OS_ERR;
3908 }
3909
3910 OSReturn os::get_native_priority(const Thread* const thread,
3911 int *priority_ptr) {
3912 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3913 *priority_ptr = java_to_os_priority[NormPriority];
3914 return OS_OK;
3915 }
3916
3917 errno = 0;
3918 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3919 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3920 }
3921
3922 // Hint to the underlying OS that a task switch would not be good.
3923 // Void return because it's a hint and can fail.
3924 void os::hint_no_preempt() {}
3925
3926 ////////////////////////////////////////////////////////////////////////////////
3927 // suspend/resume support
3928
3929 // the low-level signal-based suspend/resume support is a remnant from the
3930 // old VM-suspension that used to be for java-suspension, safepoints etc,
3931 // within hotspot. Now there is a single use-case for this:
3932 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3933 // that runs in the watcher thread.
3934 // The remaining code is greatly simplified from the more general suspension
3935 // code that used to be used.
3936 //
3937 // The protocol is quite simple:
3938 // - suspend:
3939 // - sends a signal to the target thread
3940 // - polls the suspend state of the osthread using a yield loop
3941 // - target thread signal handler (SR_handler) sets suspend state
3942 // and blocks in sigsuspend until continued
3943 // - resume:
3944 // - sets target osthread state to continue
3945 // - sends signal to end the sigsuspend loop in the SR_handler
3946 //
3947 // Note that the SR_lock plays no role in this suspend/resume protocol.
3948
3949 static void resume_clear_context(OSThread *osthread) {
3950 osthread->set_ucontext(NULL);
3951 osthread->set_siginfo(NULL);
3952 }
3953
3954 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3955 ucontext_t* context) {
3956 osthread->set_ucontext(context);
3957 osthread->set_siginfo(siginfo);
3958 }
3959
3960 // Handler function invoked when a thread's execution is suspended or
3961 // resumed. We have to be careful that only async-safe functions are
3962 // called here (Note: most pthread functions are not async safe and
3963 // should be avoided.)
3964 //
3965 // Note: sigwait() is a more natural fit than sigsuspend() from an
3966 // interface point of view, but sigwait() prevents the signal hander
3967 // from being run. libpthread would get very confused by not having
3968 // its signal handlers run and prevents sigwait()'s use with the
3969 // mutex granting granting signal.
3970 //
3971 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3972 //
3973 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3974 // Save and restore errno to avoid confusing native code with EINTR
3975 // after sigsuspend.
3976 int old_errno = errno;
3977
3978 Thread* thread = Thread::current();
3979 OSThread* osthread = thread->osthread();
3980 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3981
3982 os::SuspendResume::State current = osthread->sr.state();
3983 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3984 suspend_save_context(osthread, siginfo, context);
3985
3986 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3987 os::SuspendResume::State state = osthread->sr.suspended();
3988 if (state == os::SuspendResume::SR_SUSPENDED) {
3989 sigset_t suspend_set; // signals for sigsuspend()
3990
3991 // get current set of blocked signals and unblock resume signal
3992 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3993 sigdelset(&suspend_set, SR_signum);
3994
3995 sr_semaphore.signal();
3996 // wait here until we are resumed
3997 while (1) {
3998 sigsuspend(&suspend_set);
3999
4000 os::SuspendResume::State result = osthread->sr.running();
4001 if (result == os::SuspendResume::SR_RUNNING) {
4002 sr_semaphore.signal();
4003 break;
4004 }
4005 }
4006
4007 } else if (state == os::SuspendResume::SR_RUNNING) {
4008 // request was cancelled, continue
4009 } else {
4010 ShouldNotReachHere();
4011 }
4012
4013 resume_clear_context(osthread);
4014 } else if (current == os::SuspendResume::SR_RUNNING) {
4015 // request was cancelled, continue
4016 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4017 // ignore
4018 } else {
4019 // ignore
4020 }
4021
4022 errno = old_errno;
4023 }
4024
4025
4026 static int SR_initialize() {
4027 struct sigaction act;
4028 char *s;
4029 // Get signal number to use for suspend/resume
4030 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4031 int sig = ::strtol(s, 0, 10);
4032 if (sig > 0 || sig < _NSIG) {
4033 SR_signum = sig;
4034 }
4035 }
4036
4037 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4038 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4039
4040 sigemptyset(&SR_sigset);
4041 sigaddset(&SR_sigset, SR_signum);
4042
4043 // Set up signal handler for suspend/resume
4044 act.sa_flags = SA_RESTART|SA_SIGINFO;
4045 act.sa_handler = (void (*)(int)) SR_handler;
4046
4047 // SR_signum is blocked by default.
4048 // 4528190 - We also need to block pthread restart signal (32 on all
4049 // supported Linux platforms). Note that LinuxThreads need to block
4050 // this signal for all threads to work properly. So we don't have
4051 // to use hard-coded signal number when setting up the mask.
4052 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4053
4054 if (sigaction(SR_signum, &act, 0) == -1) {
4055 return -1;
4056 }
4057
4058 // Save signal flag
4059 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4060 return 0;
4061 }
4062
4063 static int sr_notify(OSThread* osthread) {
4064 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4065 assert_status(status == 0, status, "pthread_kill");
4066 return status;
4067 }
4068
4069 // "Randomly" selected value for how long we want to spin
4070 // before bailing out on suspending a thread, also how often
4071 // we send a signal to a thread we want to resume
4072 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4073 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4074
4075 // returns true on success and false on error - really an error is fatal
4076 // but this seems the normal response to library errors
4077 static bool do_suspend(OSThread* osthread) {
4078 assert(osthread->sr.is_running(), "thread should be running");
4079 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4080
4081 // mark as suspended and send signal
4082 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4083 // failed to switch, state wasn't running?
4084 ShouldNotReachHere();
4085 return false;
4086 }
4087
4088 if (sr_notify(osthread) != 0) {
4089 ShouldNotReachHere();
4090 }
4091
4092 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4093 while (true) {
4094 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4095 break;
4096 } else {
4097 // timeout
4098 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4099 if (cancelled == os::SuspendResume::SR_RUNNING) {
4100 return false;
4101 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4102 // make sure that we consume the signal on the semaphore as well
4103 sr_semaphore.wait();
4104 break;
4105 } else {
4106 ShouldNotReachHere();
4107 return false;
4108 }
4109 }
4110 }
4111
4112 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4113 return true;
4114 }
4115
4116 static void do_resume(OSThread* osthread) {
4117 assert(osthread->sr.is_suspended(), "thread should be suspended");
4118 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4119
4120 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4121 // failed to switch to WAKEUP_REQUEST
4122 ShouldNotReachHere();
4123 return;
4124 }
4125
4126 while (true) {
4127 if (sr_notify(osthread) == 0) {
4128 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4129 if (osthread->sr.is_running()) {
4130 return;
4131 }
4132 }
4133 } else {
4134 ShouldNotReachHere();
4135 }
4136 }
4137
4138 guarantee(osthread->sr.is_running(), "Must be running!");
4139 }
4140
4141 ///////////////////////////////////////////////////////////////////////////////////
4142 // signal handling (except suspend/resume)
4143
4144 // This routine may be used by user applications as a "hook" to catch signals.
4145 // The user-defined signal handler must pass unrecognized signals to this
4146 // routine, and if it returns true (non-zero), then the signal handler must
4147 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4148 // routine will never retun false (zero), but instead will execute a VM panic
4149 // routine kill the process.
4150 //
4151 // If this routine returns false, it is OK to call it again. This allows
4152 // the user-defined signal handler to perform checks either before or after
4153 // the VM performs its own checks. Naturally, the user code would be making
4154 // a serious error if it tried to handle an exception (such as a null check
4155 // or breakpoint) that the VM was generating for its own correct operation.
4156 //
4157 // This routine may recognize any of the following kinds of signals:
4158 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4159 // It should be consulted by handlers for any of those signals.
4160 //
4161 // The caller of this routine must pass in the three arguments supplied
4162 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4163 // field of the structure passed to sigaction(). This routine assumes that
4164 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4165 //
4166 // Note that the VM will print warnings if it detects conflicting signal
4167 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4168 //
4169 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4170 siginfo_t* siginfo,
4171 void* ucontext,
4172 int abort_if_unrecognized);
4173
4174 void signalHandler(int sig, siginfo_t* info, void* uc) {
4175 assert(info != NULL && uc != NULL, "it must be old kernel");
4176 int orig_errno = errno; // Preserve errno value over signal handler.
4177 JVM_handle_linux_signal(sig, info, uc, true);
4178 errno = orig_errno;
4179 }
4180
4181
4182 // This boolean allows users to forward their own non-matching signals
4183 // to JVM_handle_linux_signal, harmlessly.
4184 bool os::Linux::signal_handlers_are_installed = false;
4185
4186 // For signal-chaining
4187 struct sigaction os::Linux::sigact[MAXSIGNUM];
4188 unsigned int os::Linux::sigs = 0;
4189 bool os::Linux::libjsig_is_loaded = false;
4190 typedef struct sigaction *(*get_signal_t)(int);
4191 get_signal_t os::Linux::get_signal_action = NULL;
4192
4193 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4194 struct sigaction *actp = NULL;
4195
4196 if (libjsig_is_loaded) {
4197 // Retrieve the old signal handler from libjsig
4198 actp = (*get_signal_action)(sig);
4199 }
4200 if (actp == NULL) {
4201 // Retrieve the preinstalled signal handler from jvm
4202 actp = get_preinstalled_handler(sig);
4203 }
4204
4205 return actp;
4206 }
4207
4208 static bool call_chained_handler(struct sigaction *actp, int sig,
4209 siginfo_t *siginfo, void *context) {
4210 // Call the old signal handler
4211 if (actp->sa_handler == SIG_DFL) {
4212 // It's more reasonable to let jvm treat it as an unexpected exception
4213 // instead of taking the default action.
4214 return false;
4215 } else if (actp->sa_handler != SIG_IGN) {
4216 if ((actp->sa_flags & SA_NODEFER) == 0) {
4217 // automaticlly block the signal
4218 sigaddset(&(actp->sa_mask), sig);
4219 }
4220
4221 sa_handler_t hand;
4222 sa_sigaction_t sa;
4223 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4224 // retrieve the chained handler
4225 if (siginfo_flag_set) {
4226 sa = actp->sa_sigaction;
4227 } else {
4228 hand = actp->sa_handler;
4229 }
4230
4231 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4232 actp->sa_handler = SIG_DFL;
4233 }
4234
4235 // try to honor the signal mask
4236 sigset_t oset;
4237 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4238
4239 // call into the chained handler
4240 if (siginfo_flag_set) {
4241 (*sa)(sig, siginfo, context);
4242 } else {
4243 (*hand)(sig);
4244 }
4245
4246 // restore the signal mask
4247 pthread_sigmask(SIG_SETMASK, &oset, 0);
4248 }
4249 // Tell jvm's signal handler the signal is taken care of.
4250 return true;
4251 }
4252
4253 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4254 bool chained = false;
4255 // signal-chaining
4256 if (UseSignalChaining) {
4257 struct sigaction *actp = get_chained_signal_action(sig);
4258 if (actp != NULL) {
4259 chained = call_chained_handler(actp, sig, siginfo, context);
4260 }
4261 }
4262 return chained;
4263 }
4264
4265 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4266 if ((((unsigned int)1 << sig) & sigs) != 0) {
4267 return &sigact[sig];
4268 }
4269 return NULL;
4270 }
4271
4272 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4273 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4274 sigact[sig] = oldAct;
4275 sigs |= (unsigned int)1 << sig;
4276 }
4277
4278 // for diagnostic
4279 int os::Linux::sigflags[MAXSIGNUM];
4280
4281 int os::Linux::get_our_sigflags(int sig) {
4282 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4283 return sigflags[sig];
4284 }
4285
4286 void os::Linux::set_our_sigflags(int sig, int flags) {
4287 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4288 sigflags[sig] = flags;
4289 }
4290
4291 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4292 // Check for overwrite.
4293 struct sigaction oldAct;
4294 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4295
4296 void* oldhand = oldAct.sa_sigaction
4297 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4298 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4299 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4300 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4301 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4302 if (AllowUserSignalHandlers || !set_installed) {
4303 // Do not overwrite; user takes responsibility to forward to us.
4304 return;
4305 } else if (UseSignalChaining) {
4306 // save the old handler in jvm
4307 save_preinstalled_handler(sig, oldAct);
4308 // libjsig also interposes the sigaction() call below and saves the
4309 // old sigaction on it own.
4310 } else {
4311 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4312 "%#lx for signal %d.", (long)oldhand, sig));
4313 }
4314 }
4315
4316 struct sigaction sigAct;
4317 sigfillset(&(sigAct.sa_mask));
4318 sigAct.sa_handler = SIG_DFL;
4319 if (!set_installed) {
4320 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4321 } else {
4322 sigAct.sa_sigaction = signalHandler;
4323 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4324 }
4325 // Save flags, which are set by ours
4326 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4327 sigflags[sig] = sigAct.sa_flags;
4328
4329 int ret = sigaction(sig, &sigAct, &oldAct);
4330 assert(ret == 0, "check");
4331
4332 void* oldhand2 = oldAct.sa_sigaction
4333 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4334 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4335 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4336 }
4337
4338 // install signal handlers for signals that HotSpot needs to
4339 // handle in order to support Java-level exception handling.
4340
4341 void os::Linux::install_signal_handlers() {
4342 if (!signal_handlers_are_installed) {
4343 signal_handlers_are_installed = true;
4344
4345 // signal-chaining
4346 typedef void (*signal_setting_t)();
4347 signal_setting_t begin_signal_setting = NULL;
4348 signal_setting_t end_signal_setting = NULL;
4349 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4350 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4351 if (begin_signal_setting != NULL) {
4352 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4353 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4354 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4355 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4356 libjsig_is_loaded = true;
4357 assert(UseSignalChaining, "should enable signal-chaining");
4358 }
4359 if (libjsig_is_loaded) {
4360 // Tell libjsig jvm is setting signal handlers
4361 (*begin_signal_setting)();
4362 }
4363
4364 set_signal_handler(SIGSEGV, true);
4365 set_signal_handler(SIGPIPE, true);
4366 set_signal_handler(SIGBUS, true);
4367 set_signal_handler(SIGILL, true);
4368 set_signal_handler(SIGFPE, true);
4369 #if defined(PPC64)
4370 set_signal_handler(SIGTRAP, true);
4371 #endif
4372 set_signal_handler(SIGXFSZ, true);
4373
4374 if (libjsig_is_loaded) {
4375 // Tell libjsig jvm finishes setting signal handlers
4376 (*end_signal_setting)();
4377 }
4378
4379 // We don't activate signal checker if libjsig is in place, we trust ourselves
4380 // and if UserSignalHandler is installed all bets are off.
4381 // Log that signal checking is off only if -verbose:jni is specified.
4382 if (CheckJNICalls) {
4383 if (libjsig_is_loaded) {
4384 if (PrintJNIResolving) {
4385 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4386 }
4387 check_signals = false;
4388 }
4389 if (AllowUserSignalHandlers) {
4390 if (PrintJNIResolving) {
4391 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4392 }
4393 check_signals = false;
4394 }
4395 }
4396 }
4397 }
4398
4399 // This is the fastest way to get thread cpu time on Linux.
4400 // Returns cpu time (user+sys) for any thread, not only for current.
4401 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4402 // It might work on 2.6.10+ with a special kernel/glibc patch.
4403 // For reference, please, see IEEE Std 1003.1-2004:
4404 // http://www.unix.org/single_unix_specification
4405
4406 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4407 struct timespec tp;
4408 int rc = os::Linux::clock_gettime(clockid, &tp);
4409 assert(rc == 0, "clock_gettime is expected to return 0 code");
4410
4411 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4412 }
4413
4414 /////
4415 // glibc on Linux platform uses non-documented flag
4416 // to indicate, that some special sort of signal
4417 // trampoline is used.
4418 // We will never set this flag, and we should
4419 // ignore this flag in our diagnostic
4420 #ifdef SIGNIFICANT_SIGNAL_MASK
4421 #undef SIGNIFICANT_SIGNAL_MASK
4422 #endif
4423 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4424
4425 static const char* get_signal_handler_name(address handler,
4426 char* buf, int buflen) {
4427 int offset;
4428 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4429 if (found) {
4430 // skip directory names
4431 const char *p1, *p2;
4432 p1 = buf;
4433 size_t len = strlen(os::file_separator());
4434 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4435 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4436 } else {
4437 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4438 }
4439 return buf;
4440 }
4441
4442 static void print_signal_handler(outputStream* st, int sig,
4443 char* buf, size_t buflen) {
4444 struct sigaction sa;
4445
4446 sigaction(sig, NULL, &sa);
4447
4448 // See comment for SIGNIFICANT_SIGNAL_MASK define
4449 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4450
4451 st->print("%s: ", os::exception_name(sig, buf, buflen));
4452
4453 address handler = (sa.sa_flags & SA_SIGINFO)
4454 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4455 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4456
4457 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4458 st->print("SIG_DFL");
4459 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4460 st->print("SIG_IGN");
4461 } else {
4462 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4463 }
4464
4465 st->print(", sa_mask[0]=");
4466 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4467
4468 address rh = VMError::get_resetted_sighandler(sig);
4469 // May be, handler was resetted by VMError?
4470 if (rh != NULL) {
4471 handler = rh;
4472 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4473 }
4474
4475 st->print(", sa_flags=");
4476 os::Posix::print_sa_flags(st, sa.sa_flags);
4477
4478 // Check: is it our handler?
4479 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4480 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4481 // It is our signal handler
4482 // check for flags, reset system-used one!
4483 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4484 st->print(
4485 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4486 os::Linux::get_our_sigflags(sig));
4487 }
4488 }
4489 st->cr();
4490 }
4491
4492
4493 #define DO_SIGNAL_CHECK(sig) \
4494 do { \
4495 if (!sigismember(&check_signal_done, sig)) { \
4496 os::Linux::check_signal_handler(sig); \
4497 } \
4498 } while (0)
4499
4500 // This method is a periodic task to check for misbehaving JNI applications
4501 // under CheckJNI, we can add any periodic checks here
4502
4503 void os::run_periodic_checks() {
4504 if (check_signals == false) return;
4505
4506 // SEGV and BUS if overridden could potentially prevent
4507 // generation of hs*.log in the event of a crash, debugging
4508 // such a case can be very challenging, so we absolutely
4509 // check the following for a good measure:
4510 DO_SIGNAL_CHECK(SIGSEGV);
4511 DO_SIGNAL_CHECK(SIGILL);
4512 DO_SIGNAL_CHECK(SIGFPE);
4513 DO_SIGNAL_CHECK(SIGBUS);
4514 DO_SIGNAL_CHECK(SIGPIPE);
4515 DO_SIGNAL_CHECK(SIGXFSZ);
4516 #if defined(PPC64)
4517 DO_SIGNAL_CHECK(SIGTRAP);
4518 #endif
4519
4520 // ReduceSignalUsage allows the user to override these handlers
4521 // see comments at the very top and jvm_solaris.h
4522 if (!ReduceSignalUsage) {
4523 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4524 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4525 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4526 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4527 }
4528
4529 DO_SIGNAL_CHECK(SR_signum);
4530 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4531 }
4532
4533 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4534
4535 static os_sigaction_t os_sigaction = NULL;
4536
4537 void os::Linux::check_signal_handler(int sig) {
4538 char buf[O_BUFLEN];
4539 address jvmHandler = NULL;
4540
4541
4542 struct sigaction act;
4543 if (os_sigaction == NULL) {
4544 // only trust the default sigaction, in case it has been interposed
4545 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4546 if (os_sigaction == NULL) return;
4547 }
4548
4549 os_sigaction(sig, (struct sigaction*)NULL, &act);
4550
4551
4552 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4553
4554 address thisHandler = (act.sa_flags & SA_SIGINFO)
4555 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4556 : CAST_FROM_FN_PTR(address, act.sa_handler);
4557
4558
4559 switch (sig) {
4560 case SIGSEGV:
4561 case SIGBUS:
4562 case SIGFPE:
4563 case SIGPIPE:
4564 case SIGILL:
4565 case SIGXFSZ:
4566 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4567 break;
4568
4569 case SHUTDOWN1_SIGNAL:
4570 case SHUTDOWN2_SIGNAL:
4571 case SHUTDOWN3_SIGNAL:
4572 case BREAK_SIGNAL:
4573 jvmHandler = (address)user_handler();
4574 break;
4575
4576 case INTERRUPT_SIGNAL:
4577 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4578 break;
4579
4580 default:
4581 if (sig == SR_signum) {
4582 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4583 } else {
4584 return;
4585 }
4586 break;
4587 }
4588
4589 if (thisHandler != jvmHandler) {
4590 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4591 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4592 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4593 // No need to check this sig any longer
4594 sigaddset(&check_signal_done, sig);
4595 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4596 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4597 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4598 exception_name(sig, buf, O_BUFLEN));
4599 }
4600 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4601 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4602 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4603 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4604 // No need to check this sig any longer
4605 sigaddset(&check_signal_done, sig);
4606 }
4607
4608 // Dump all the signal
4609 if (sigismember(&check_signal_done, sig)) {
4610 print_signal_handlers(tty, buf, O_BUFLEN);
4611 }
4612 }
4613
4614 extern void report_error(char* file_name, int line_no, char* title,
4615 char* format, ...);
4616
4617 extern bool signal_name(int signo, char* buf, size_t len);
4618
4619 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4620 if (0 < exception_code && exception_code <= SIGRTMAX) {
4621 // signal
4622 if (!signal_name(exception_code, buf, size)) {
4623 jio_snprintf(buf, size, "SIG%d", exception_code);
4624 }
4625 return buf;
4626 } else {
4627 return NULL;
4628 }
4629 }
4630
4631 // this is called _before_ the most of global arguments have been parsed
4632 void os::init(void) {
4633 char dummy; // used to get a guess on initial stack address
4634 // first_hrtime = gethrtime();
4635
4636 // With LinuxThreads the JavaMain thread pid (primordial thread)
4637 // is different than the pid of the java launcher thread.
4638 // So, on Linux, the launcher thread pid is passed to the VM
4639 // via the sun.java.launcher.pid property.
4640 // Use this property instead of getpid() if it was correctly passed.
4641 // See bug 6351349.
4642 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4643
4644 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4645
4646 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4647
4648 init_random(1234567);
4649
4650 ThreadCritical::initialize();
4651
4652 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4653 if (Linux::page_size() == -1) {
4654 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4655 strerror(errno)));
4656 }
4657 init_page_sizes((size_t) Linux::page_size());
4658
4659 Linux::initialize_system_info();
4660
4661 // main_thread points to the aboriginal thread
4662 Linux::_main_thread = pthread_self();
4663
4664 Linux::clock_init();
4665 initial_time_count = javaTimeNanos();
4666
4667 // pthread_condattr initialization for monotonic clock
4668 int status;
4669 pthread_condattr_t* _condattr = os::Linux::condAttr();
4670 if ((status = pthread_condattr_init(_condattr)) != 0) {
4671 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4672 }
4673 // Only set the clock if CLOCK_MONOTONIC is available
4674 if (os::supports_monotonic_clock()) {
4675 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4676 if (status == EINVAL) {
4677 warning("Unable to use monotonic clock with relative timed-waits" \
4678 " - changes to the time-of-day clock may have adverse affects");
4679 } else {
4680 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4681 }
4682 }
4683 }
4684 // else it defaults to CLOCK_REALTIME
4685
4686 // If the pagesize of the VM is greater than 8K determine the appropriate
4687 // number of initial guard pages. The user can change this with the
4688 // command line arguments, if needed.
4689 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4690 StackYellowPages = 1;
4691 StackRedPages = 1;
4692 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4693 }
4694
4695 // retrieve entry point for pthread_setname_np
4696 Linux::_pthread_setname_np =
4697 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4698
4699 }
4700
4701 // To install functions for atexit system call
4702 extern "C" {
4703 static void perfMemory_exit_helper() {
4704 perfMemory_exit();
4705 }
4706 }
4707
4708 // this is called _after_ the global arguments have been parsed
4709 jint os::init_2(void) {
4710 Linux::fast_thread_clock_init();
4711
4712 // Allocate a single page and mark it as readable for safepoint polling
4713 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4714 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4715
4716 os::set_polling_page(polling_page);
4717
4718 #ifndef PRODUCT
4719 if (Verbose && PrintMiscellaneous) {
4720 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4721 (intptr_t)polling_page);
4722 }
4723 #endif
4724
4725 if (!UseMembar) {
4726 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4727 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4728 os::set_memory_serialize_page(mem_serialize_page);
4729
4730 #ifndef PRODUCT
4731 if (Verbose && PrintMiscellaneous) {
4732 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4733 (intptr_t)mem_serialize_page);
4734 }
4735 #endif
4736 }
4737
4738 // initialize suspend/resume support - must do this before signal_sets_init()
4739 if (SR_initialize() != 0) {
4740 perror("SR_initialize failed");
4741 return JNI_ERR;
4742 }
4743
4744 Linux::signal_sets_init();
4745 Linux::install_signal_handlers();
4746
4747 // Check minimum allowable stack size for thread creation and to initialize
4748 // the java system classes, including StackOverflowError - depends on page
4749 // size. Add a page for compiler2 recursion in main thread.
4750 // Add in 2*BytesPerWord times page size to account for VM stack during
4751 // class initialization depending on 32 or 64 bit VM.
4752 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4753 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4754 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4755
4756 size_t threadStackSizeInBytes = ThreadStackSize * K;
4757 if (threadStackSizeInBytes != 0 &&
4758 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4759 tty->print_cr("\nThe stack size specified is too small, "
4760 "Specify at least %dk",
4761 os::Linux::min_stack_allowed/ K);
4762 return JNI_ERR;
4763 }
4764
4765 // Make the stack size a multiple of the page size so that
4766 // the yellow/red zones can be guarded.
4767 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4768 vm_page_size()));
4769
4770 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4771
4772 #if defined(IA32)
4773 workaround_expand_exec_shield_cs_limit();
4774 #endif
4775
4776 Linux::libpthread_init();
4777 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4778 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4779 Linux::glibc_version(), Linux::libpthread_version(),
4780 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4781 }
4782
4783 if (UseNUMA) {
4784 if (!Linux::libnuma_init()) {
4785 UseNUMA = false;
4786 } else {
4787 if ((Linux::numa_max_node() < 1)) {
4788 // There's only one node(they start from 0), disable NUMA.
4789 UseNUMA = false;
4790 }
4791 }
4792 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4793 // we can make the adaptive lgrp chunk resizing work. If the user specified
4794 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4795 // disable adaptive resizing.
4796 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4797 if (FLAG_IS_DEFAULT(UseNUMA)) {
4798 UseNUMA = false;
4799 } else {
4800 if (FLAG_IS_DEFAULT(UseLargePages) &&
4801 FLAG_IS_DEFAULT(UseSHM) &&
4802 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4803 UseLargePages = false;
4804 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4805 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4806 UseAdaptiveSizePolicy = false;
4807 UseAdaptiveNUMAChunkSizing = false;
4808 }
4809 }
4810 }
4811 if (!UseNUMA && ForceNUMA) {
4812 UseNUMA = true;
4813 }
4814 }
4815
4816 if (MaxFDLimit) {
4817 // set the number of file descriptors to max. print out error
4818 // if getrlimit/setrlimit fails but continue regardless.
4819 struct rlimit nbr_files;
4820 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4821 if (status != 0) {
4822 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4823 perror("os::init_2 getrlimit failed");
4824 }
4825 } else {
4826 nbr_files.rlim_cur = nbr_files.rlim_max;
4827 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4828 if (status != 0) {
4829 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4830 perror("os::init_2 setrlimit failed");
4831 }
4832 }
4833 }
4834 }
4835
4836 // Initialize lock used to serialize thread creation (see os::create_thread)
4837 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4838
4839 // at-exit methods are called in the reverse order of their registration.
4840 // atexit functions are called on return from main or as a result of a
4841 // call to exit(3C). There can be only 32 of these functions registered
4842 // and atexit() does not set errno.
4843
4844 if (PerfAllowAtExitRegistration) {
4845 // only register atexit functions if PerfAllowAtExitRegistration is set.
4846 // atexit functions can be delayed until process exit time, which
4847 // can be problematic for embedded VM situations. Embedded VMs should
4848 // call DestroyJavaVM() to assure that VM resources are released.
4849
4850 // note: perfMemory_exit_helper atexit function may be removed in
4851 // the future if the appropriate cleanup code can be added to the
4852 // VM_Exit VMOperation's doit method.
4853 if (atexit(perfMemory_exit_helper) != 0) {
4854 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4855 }
4856 }
4857
4858 // initialize thread priority policy
4859 prio_init();
4860
4861 return JNI_OK;
4862 }
4863
4864 // Mark the polling page as unreadable
4865 void os::make_polling_page_unreadable(void) {
4866 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4867 fatal("Could not disable polling page");
4868 }
4869 }
4870
4871 // Mark the polling page as readable
4872 void os::make_polling_page_readable(void) {
4873 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4874 fatal("Could not enable polling page");
4875 }
4876 }
4877
4878 int os::active_processor_count() {
4879 // Linux doesn't yet have a (official) notion of processor sets,
4880 // so just return the number of online processors.
4881 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4882 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4883 return online_cpus;
4884 }
4885
4886 void os::set_native_thread_name(const char *name) {
4887 if (Linux::_pthread_setname_np) {
4888 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4889 snprintf(buf, sizeof(buf), "%s", name);
4890 buf[sizeof(buf) - 1] = '\0';
4891 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4892 // ERANGE should not happen; all other errors should just be ignored.
4893 assert(rc != ERANGE, "pthread_setname_np failed");
4894 }
4895 }
4896
4897 bool os::distribute_processes(uint length, uint* distribution) {
4898 // Not yet implemented.
4899 return false;
4900 }
4901
4902 bool os::bind_to_processor(uint processor_id) {
4903 // Not yet implemented.
4904 return false;
4905 }
4906
4907 ///
4908
4909 void os::SuspendedThreadTask::internal_do_task() {
4910 if (do_suspend(_thread->osthread())) {
4911 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4912 do_task(context);
4913 do_resume(_thread->osthread());
4914 }
4915 }
4916
4917 class PcFetcher : public os::SuspendedThreadTask {
4918 public:
4919 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4920 ExtendedPC result();
4921 protected:
4922 void do_task(const os::SuspendedThreadTaskContext& context);
4923 private:
4924 ExtendedPC _epc;
4925 };
4926
4927 ExtendedPC PcFetcher::result() {
4928 guarantee(is_done(), "task is not done yet.");
4929 return _epc;
4930 }
4931
4932 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4933 Thread* thread = context.thread();
4934 OSThread* osthread = thread->osthread();
4935 if (osthread->ucontext() != NULL) {
4936 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4937 } else {
4938 // NULL context is unexpected, double-check this is the VMThread
4939 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4940 }
4941 }
4942
4943 // Suspends the target using the signal mechanism and then grabs the PC before
4944 // resuming the target. Used by the flat-profiler only
4945 ExtendedPC os::get_thread_pc(Thread* thread) {
4946 // Make sure that it is called by the watcher for the VMThread
4947 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4948 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4949
4950 PcFetcher fetcher(thread);
4951 fetcher.run();
4952 return fetcher.result();
4953 }
4954
4955 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4956 pthread_mutex_t *_mutex,
4957 const struct timespec *_abstime) {
4958 if (is_NPTL()) {
4959 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4960 } else {
4961 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4962 // word back to default 64bit precision if condvar is signaled. Java
4963 // wants 53bit precision. Save and restore current value.
4964 int fpu = get_fpu_control_word();
4965 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4966 set_fpu_control_word(fpu);
4967 return status;
4968 }
4969 }
4970
4971 ////////////////////////////////////////////////////////////////////////////////
4972 // debug support
4973
4974 bool os::find(address addr, outputStream* st) {
4975 Dl_info dlinfo;
4976 memset(&dlinfo, 0, sizeof(dlinfo));
4977 if (dladdr(addr, &dlinfo) != 0) {
4978 st->print(PTR_FORMAT ": ", addr);
4979 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4980 st->print("%s+%#x", dlinfo.dli_sname,
4981 addr - (intptr_t)dlinfo.dli_saddr);
4982 } else if (dlinfo.dli_fbase != NULL) {
4983 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4984 } else {
4985 st->print("<absolute address>");
4986 }
4987 if (dlinfo.dli_fname != NULL) {
4988 st->print(" in %s", dlinfo.dli_fname);
4989 }
4990 if (dlinfo.dli_fbase != NULL) {
4991 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4992 }
4993 st->cr();
4994
4995 if (Verbose) {
4996 // decode some bytes around the PC
4997 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4998 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4999 address lowest = (address) dlinfo.dli_sname;
5000 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5001 if (begin < lowest) begin = lowest;
5002 Dl_info dlinfo2;
5003 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5004 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5005 end = (address) dlinfo2.dli_saddr;
5006 }
5007 Disassembler::decode(begin, end, st);
5008 }
5009 return true;
5010 }
5011 return false;
5012 }
5013
5014 ////////////////////////////////////////////////////////////////////////////////
5015 // misc
5016
5017 // This does not do anything on Linux. This is basically a hook for being
5018 // able to use structured exception handling (thread-local exception filters)
5019 // on, e.g., Win32.
5020 void
5021 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5022 JavaCallArguments* args, Thread* thread) {
5023 f(value, method, args, thread);
5024 }
5025
5026 void os::print_statistics() {
5027 }
5028
5029 int os::message_box(const char* title, const char* message) {
5030 int i;
5031 fdStream err(defaultStream::error_fd());
5032 for (i = 0; i < 78; i++) err.print_raw("=");
5033 err.cr();
5034 err.print_raw_cr(title);
5035 for (i = 0; i < 78; i++) err.print_raw("-");
5036 err.cr();
5037 err.print_raw_cr(message);
5038 for (i = 0; i < 78; i++) err.print_raw("=");
5039 err.cr();
5040
5041 char buf[16];
5042 // Prevent process from exiting upon "read error" without consuming all CPU
5043 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5044
5045 return buf[0] == 'y' || buf[0] == 'Y';
5046 }
5047
5048 int os::stat(const char *path, struct stat *sbuf) {
5049 char pathbuf[MAX_PATH];
5050 if (strlen(path) > MAX_PATH - 1) {
5051 errno = ENAMETOOLONG;
5052 return -1;
5053 }
5054 os::native_path(strcpy(pathbuf, path));
5055 return ::stat(pathbuf, sbuf);
5056 }
5057
5058 bool os::check_heap(bool force) {
5059 return true;
5060 }
5061
5062 // Is a (classpath) directory empty?
5063 bool os::dir_is_empty(const char* path) {
5064 DIR *dir = NULL;
5065 struct dirent *ptr;
5066
5067 dir = opendir(path);
5068 if (dir == NULL) return true;
5069
5070 // Scan the directory
5071 bool result = true;
5072 char buf[sizeof(struct dirent) + MAX_PATH];
5073 while (result && (ptr = ::readdir(dir)) != NULL) {
5074 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5075 result = false;
5076 }
5077 }
5078 closedir(dir);
5079 return result;
5080 }
5081
5082 // This code originates from JDK's sysOpen and open64_w
5083 // from src/solaris/hpi/src/system_md.c
5084
5085 int os::open(const char *path, int oflag, int mode) {
5086 if (strlen(path) > MAX_PATH - 1) {
5087 errno = ENAMETOOLONG;
5088 return -1;
5089 }
5090
5091 // All file descriptors that are opened in the Java process and not
5092 // specifically destined for a subprocess should have the close-on-exec
5093 // flag set. If we don't set it, then careless 3rd party native code
5094 // might fork and exec without closing all appropriate file descriptors
5095 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5096 // turn might:
5097 //
5098 // - cause end-of-file to fail to be detected on some file
5099 // descriptors, resulting in mysterious hangs, or
5100 //
5101 // - might cause an fopen in the subprocess to fail on a system
5102 // suffering from bug 1085341.
5103 //
5104 // (Yes, the default setting of the close-on-exec flag is a Unix
5105 // design flaw)
5106 //
5107 // See:
5108 // 1085341: 32-bit stdio routines should support file descriptors >255
5109 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5110 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5111 //
5112 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5113 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5114 // because it saves a system call and removes a small window where the flag
5115 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5116 // and we fall back to using FD_CLOEXEC (see below).
5117 #ifdef O_CLOEXEC
5118 oflag |= O_CLOEXEC;
5119 #endif
5120
5121 int fd = ::open64(path, oflag, mode);
5122 if (fd == -1) return -1;
5123
5124 //If the open succeeded, the file might still be a directory
5125 {
5126 struct stat64 buf64;
5127 int ret = ::fstat64(fd, &buf64);
5128 int st_mode = buf64.st_mode;
5129
5130 if (ret != -1) {
5131 if ((st_mode & S_IFMT) == S_IFDIR) {
5132 errno = EISDIR;
5133 ::close(fd);
5134 return -1;
5135 }
5136 } else {
5137 ::close(fd);
5138 return -1;
5139 }
5140 }
5141
5142 #ifdef FD_CLOEXEC
5143 // Validate that the use of the O_CLOEXEC flag on open above worked.
5144 // With recent kernels, we will perform this check exactly once.
5145 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5146 if (!O_CLOEXEC_is_known_to_work) {
5147 int flags = ::fcntl(fd, F_GETFD);
5148 if (flags != -1) {
5149 if ((flags & FD_CLOEXEC) != 0)
5150 O_CLOEXEC_is_known_to_work = 1;
5151 else
5152 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5153 }
5154 }
5155 #endif
5156
5157 return fd;
5158 }
5159
5160
5161 // create binary file, rewriting existing file if required
5162 int os::create_binary_file(const char* path, bool rewrite_existing) {
5163 int oflags = O_WRONLY | O_CREAT;
5164 if (!rewrite_existing) {
5165 oflags |= O_EXCL;
5166 }
5167 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5168 }
5169
5170 // return current position of file pointer
5171 jlong os::current_file_offset(int fd) {
5172 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5173 }
5174
5175 // move file pointer to the specified offset
5176 jlong os::seek_to_file_offset(int fd, jlong offset) {
5177 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5178 }
5179
5180 // This code originates from JDK's sysAvailable
5181 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5182
5183 int os::available(int fd, jlong *bytes) {
5184 jlong cur, end;
5185 int mode;
5186 struct stat64 buf64;
5187
5188 if (::fstat64(fd, &buf64) >= 0) {
5189 mode = buf64.st_mode;
5190 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5191 // XXX: is the following call interruptible? If so, this might
5192 // need to go through the INTERRUPT_IO() wrapper as for other
5193 // blocking, interruptible calls in this file.
5194 int n;
5195 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5196 *bytes = n;
5197 return 1;
5198 }
5199 }
5200 }
5201 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5202 return 0;
5203 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5204 return 0;
5205 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5206 return 0;
5207 }
5208 *bytes = end - cur;
5209 return 1;
5210 }
5211
5212 // Map a block of memory.
5213 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5214 char *addr, size_t bytes, bool read_only,
5215 bool allow_exec) {
5216 int prot;
5217 int flags = MAP_PRIVATE;
5218
5219 if (read_only) {
5220 prot = PROT_READ;
5221 } else {
5222 prot = PROT_READ | PROT_WRITE;
5223 }
5224
5225 if (allow_exec) {
5226 prot |= PROT_EXEC;
5227 }
5228
5229 if (addr != NULL) {
5230 flags |= MAP_FIXED;
5231 }
5232
5233 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5234 fd, file_offset);
5235 if (mapped_address == MAP_FAILED) {
5236 return NULL;
5237 }
5238 return mapped_address;
5239 }
5240
5241
5242 // Remap a block of memory.
5243 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5244 char *addr, size_t bytes, bool read_only,
5245 bool allow_exec) {
5246 // same as map_memory() on this OS
5247 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5248 allow_exec);
5249 }
5250
5251
5252 // Unmap a block of memory.
5253 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5254 return munmap(addr, bytes) == 0;
5255 }
5256
5257 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5258
5259 static clockid_t thread_cpu_clockid(Thread* thread) {
5260 pthread_t tid = thread->osthread()->pthread_id();
5261 clockid_t clockid;
5262
5263 // Get thread clockid
5264 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5265 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5266 return clockid;
5267 }
5268
5269 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5270 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5271 // of a thread.
5272 //
5273 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5274 // the fast estimate available on the platform.
5275
5276 jlong os::current_thread_cpu_time() {
5277 if (os::Linux::supports_fast_thread_cpu_time()) {
5278 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5279 } else {
5280 // return user + sys since the cost is the same
5281 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5282 }
5283 }
5284
5285 jlong os::thread_cpu_time(Thread* thread) {
5286 // consistent with what current_thread_cpu_time() returns
5287 if (os::Linux::supports_fast_thread_cpu_time()) {
5288 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5289 } else {
5290 return slow_thread_cpu_time(thread, true /* user + sys */);
5291 }
5292 }
5293
5294 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5295 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5296 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5297 } else {
5298 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5299 }
5300 }
5301
5302 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5303 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5304 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5305 } else {
5306 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5307 }
5308 }
5309
5310 // -1 on error.
5311 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5312 pid_t tid = thread->osthread()->thread_id();
5313 char *s;
5314 char stat[2048];
5315 int statlen;
5316 char proc_name[64];
5317 int count;
5318 long sys_time, user_time;
5319 char cdummy;
5320 int idummy;
5321 long ldummy;
5322 FILE *fp;
5323
5324 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5325 fp = fopen(proc_name, "r");
5326 if (fp == NULL) return -1;
5327 statlen = fread(stat, 1, 2047, fp);
5328 stat[statlen] = '\0';
5329 fclose(fp);
5330
5331 // Skip pid and the command string. Note that we could be dealing with
5332 // weird command names, e.g. user could decide to rename java launcher
5333 // to "java 1.4.2 :)", then the stat file would look like
5334 // 1234 (java 1.4.2 :)) R ... ...
5335 // We don't really need to know the command string, just find the last
5336 // occurrence of ")" and then start parsing from there. See bug 4726580.
5337 s = strrchr(stat, ')');
5338 if (s == NULL) return -1;
5339
5340 // Skip blank chars
5341 do { s++; } while (s && isspace(*s));
5342
5343 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5344 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5345 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5346 &user_time, &sys_time);
5347 if (count != 13) return -1;
5348 if (user_sys_cpu_time) {
5349 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5350 } else {
5351 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5352 }
5353 }
5354
5355 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5356 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5357 info_ptr->may_skip_backward = false; // elapsed time not wall time
5358 info_ptr->may_skip_forward = false; // elapsed time not wall time
5359 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5360 }
5361
5362 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5363 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5364 info_ptr->may_skip_backward = false; // elapsed time not wall time
5365 info_ptr->may_skip_forward = false; // elapsed time not wall time
5366 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5367 }
5368
5369 bool os::is_thread_cpu_time_supported() {
5370 return true;
5371 }
5372
5373 // System loadavg support. Returns -1 if load average cannot be obtained.
5374 // Linux doesn't yet have a (official) notion of processor sets,
5375 // so just return the system wide load average.
5376 int os::loadavg(double loadavg[], int nelem) {
5377 return ::getloadavg(loadavg, nelem);
5378 }
5379
5380 void os::pause() {
5381 char filename[MAX_PATH];
5382 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5383 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5384 } else {
5385 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5386 }
5387
5388 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5389 if (fd != -1) {
5390 struct stat buf;
5391 ::close(fd);
5392 while (::stat(filename, &buf) == 0) {
5393 (void)::poll(NULL, 0, 100);
5394 }
5395 } else {
5396 jio_fprintf(stderr,
5397 "Could not open pause file '%s', continuing immediately.\n", filename);
5398 }
5399 }
5400
5401
5402 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5403 // comment paragraphs are worth repeating here:
5404 //
5405 // Assumption:
5406 // Only one parker can exist on an event, which is why we allocate
5407 // them per-thread. Multiple unparkers can coexist.
5408 //
5409 // _Event serves as a restricted-range semaphore.
5410 // -1 : thread is blocked, i.e. there is a waiter
5411 // 0 : neutral: thread is running or ready,
5412 // could have been signaled after a wait started
5413 // 1 : signaled - thread is running or ready
5414 //
5415 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5416 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5417 // For specifics regarding the bug see GLIBC BUGID 261237 :
5418 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5419 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5420 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5421 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5422 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5423 // and monitorenter when we're using 1-0 locking. All those operations may result in
5424 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5425 // of libpthread avoids the problem, but isn't practical.
5426 //
5427 // Possible remedies:
5428 //
5429 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5430 // This is palliative and probabilistic, however. If the thread is preempted
5431 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5432 // than the minimum period may have passed, and the abstime may be stale (in the
5433 // past) resultin in a hang. Using this technique reduces the odds of a hang
5434 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5435 //
5436 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5437 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5438 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5439 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5440 // thread.
5441 //
5442 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5443 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5444 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5445 // This also works well. In fact it avoids kernel-level scalability impediments
5446 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5447 // timers in a graceful fashion.
5448 //
5449 // 4. When the abstime value is in the past it appears that control returns
5450 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5451 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5452 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5453 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5454 // It may be possible to avoid reinitialization by checking the return
5455 // value from pthread_cond_timedwait(). In addition to reinitializing the
5456 // condvar we must establish the invariant that cond_signal() is only called
5457 // within critical sections protected by the adjunct mutex. This prevents
5458 // cond_signal() from "seeing" a condvar that's in the midst of being
5459 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5460 // desirable signal-after-unlock optimization that avoids futile context switching.
5461 //
5462 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5463 // structure when a condvar is used or initialized. cond_destroy() would
5464 // release the helper structure. Our reinitialize-after-timedwait fix
5465 // put excessive stress on malloc/free and locks protecting the c-heap.
5466 //
5467 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5468 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5469 // and only enabling the work-around for vulnerable environments.
5470
5471 // utility to compute the abstime argument to timedwait:
5472 // millis is the relative timeout time
5473 // abstime will be the absolute timeout time
5474 // TODO: replace compute_abstime() with unpackTime()
5475
5476 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5477 if (millis < 0) millis = 0;
5478
5479 jlong seconds = millis / 1000;
5480 millis %= 1000;
5481 if (seconds > 50000000) { // see man cond_timedwait(3T)
5482 seconds = 50000000;
5483 }
5484
5485 if (os::supports_monotonic_clock()) {
5486 struct timespec now;
5487 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5488 assert_status(status == 0, status, "clock_gettime");
5489 abstime->tv_sec = now.tv_sec + seconds;
5490 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5491 if (nanos >= NANOSECS_PER_SEC) {
5492 abstime->tv_sec += 1;
5493 nanos -= NANOSECS_PER_SEC;
5494 }
5495 abstime->tv_nsec = nanos;
5496 } else {
5497 struct timeval now;
5498 int status = gettimeofday(&now, NULL);
5499 assert(status == 0, "gettimeofday");
5500 abstime->tv_sec = now.tv_sec + seconds;
5501 long usec = now.tv_usec + millis * 1000;
5502 if (usec >= 1000000) {
5503 abstime->tv_sec += 1;
5504 usec -= 1000000;
5505 }
5506 abstime->tv_nsec = usec * 1000;
5507 }
5508 return abstime;
5509 }
5510
5511 void os::PlatformEvent::park() { // AKA "down()"
5512 // Transitions for _Event:
5513 // -1 => -1 : illegal
5514 // 1 => 0 : pass - return immediately
5515 // 0 => -1 : block; then set _Event to 0 before returning
5516
5517 // Invariant: Only the thread associated with the Event/PlatformEvent
5518 // may call park().
5519 // TODO: assert that _Assoc != NULL or _Assoc == Self
5520 assert(_nParked == 0, "invariant");
5521
5522 int v;
5523 for (;;) {
5524 v = _Event;
5525 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5526 }
5527 guarantee(v >= 0, "invariant");
5528 if (v == 0) {
5529 // Do this the hard way by blocking ...
5530 int status = pthread_mutex_lock(_mutex);
5531 assert_status(status == 0, status, "mutex_lock");
5532 guarantee(_nParked == 0, "invariant");
5533 ++_nParked;
5534 while (_Event < 0) {
5535 status = pthread_cond_wait(_cond, _mutex);
5536 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5537 // Treat this the same as if the wait was interrupted
5538 if (status == ETIME) { status = EINTR; }
5539 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5540 }
5541 --_nParked;
5542
5543 _Event = 0;
5544 status = pthread_mutex_unlock(_mutex);
5545 assert_status(status == 0, status, "mutex_unlock");
5546 // Paranoia to ensure our locked and lock-free paths interact
5547 // correctly with each other.
5548 OrderAccess::fence();
5549 }
5550 guarantee(_Event >= 0, "invariant");
5551 }
5552
5553 int os::PlatformEvent::park(jlong millis) {
5554 // Transitions for _Event:
5555 // -1 => -1 : illegal
5556 // 1 => 0 : pass - return immediately
5557 // 0 => -1 : block; then set _Event to 0 before returning
5558
5559 guarantee(_nParked == 0, "invariant");
5560
5561 int v;
5562 for (;;) {
5563 v = _Event;
5564 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5565 }
5566 guarantee(v >= 0, "invariant");
5567 if (v != 0) return OS_OK;
5568
5569 // We do this the hard way, by blocking the thread.
5570 // Consider enforcing a minimum timeout value.
5571 struct timespec abst;
5572 compute_abstime(&abst, millis);
5573
5574 int ret = OS_TIMEOUT;
5575 int status = pthread_mutex_lock(_mutex);
5576 assert_status(status == 0, status, "mutex_lock");
5577 guarantee(_nParked == 0, "invariant");
5578 ++_nParked;
5579
5580 // Object.wait(timo) will return because of
5581 // (a) notification
5582 // (b) timeout
5583 // (c) thread.interrupt
5584 //
5585 // Thread.interrupt and object.notify{All} both call Event::set.
5586 // That is, we treat thread.interrupt as a special case of notification.
5587 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5588 // We assume all ETIME returns are valid.
5589 //
5590 // TODO: properly differentiate simultaneous notify+interrupt.
5591 // In that case, we should propagate the notify to another waiter.
5592
5593 while (_Event < 0) {
5594 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5595 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5596 pthread_cond_destroy(_cond);
5597 pthread_cond_init(_cond, os::Linux::condAttr());
5598 }
5599 assert_status(status == 0 || status == EINTR ||
5600 status == ETIME || status == ETIMEDOUT,
5601 status, "cond_timedwait");
5602 if (!FilterSpuriousWakeups) break; // previous semantics
5603 if (status == ETIME || status == ETIMEDOUT) break;
5604 // We consume and ignore EINTR and spurious wakeups.
5605 }
5606 --_nParked;
5607 if (_Event >= 0) {
5608 ret = OS_OK;
5609 }
5610 _Event = 0;
5611 status = pthread_mutex_unlock(_mutex);
5612 assert_status(status == 0, status, "mutex_unlock");
5613 assert(_nParked == 0, "invariant");
5614 // Paranoia to ensure our locked and lock-free paths interact
5615 // correctly with each other.
5616 OrderAccess::fence();
5617 return ret;
5618 }
5619
5620 void os::PlatformEvent::unpark() {
5621 // Transitions for _Event:
5622 // 0 => 1 : just return
5623 // 1 => 1 : just return
5624 // -1 => either 0 or 1; must signal target thread
5625 // That is, we can safely transition _Event from -1 to either
5626 // 0 or 1.
5627 // See also: "Semaphores in Plan 9" by Mullender & Cox
5628 //
5629 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5630 // that it will take two back-to-back park() calls for the owning
5631 // thread to block. This has the benefit of forcing a spurious return
5632 // from the first park() call after an unpark() call which will help
5633 // shake out uses of park() and unpark() without condition variables.
5634
5635 if (Atomic::xchg(1, &_Event) >= 0) return;
5636
5637 // Wait for the thread associated with the event to vacate
5638 int status = pthread_mutex_lock(_mutex);
5639 assert_status(status == 0, status, "mutex_lock");
5640 int AnyWaiters = _nParked;
5641 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5642 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5643 AnyWaiters = 0;
5644 pthread_cond_signal(_cond);
5645 }
5646 status = pthread_mutex_unlock(_mutex);
5647 assert_status(status == 0, status, "mutex_unlock");
5648 if (AnyWaiters != 0) {
5649 // Note that we signal() *after* dropping the lock for "immortal" Events.
5650 // This is safe and avoids a common class of futile wakeups. In rare
5651 // circumstances this can cause a thread to return prematurely from
5652 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5653 // will simply re-test the condition and re-park itself.
5654 // This provides particular benefit if the underlying platform does not
5655 // provide wait morphing.
5656 status = pthread_cond_signal(_cond);
5657 assert_status(status == 0, status, "cond_signal");
5658 }
5659 }
5660
5661
5662 // JSR166
5663 // -------------------------------------------------------
5664
5665 // The solaris and linux implementations of park/unpark are fairly
5666 // conservative for now, but can be improved. They currently use a
5667 // mutex/condvar pair, plus a a count.
5668 // Park decrements count if > 0, else does a condvar wait. Unpark
5669 // sets count to 1 and signals condvar. Only one thread ever waits
5670 // on the condvar. Contention seen when trying to park implies that someone
5671 // is unparking you, so don't wait. And spurious returns are fine, so there
5672 // is no need to track notifications.
5673
5674 // This code is common to linux and solaris and will be moved to a
5675 // common place in dolphin.
5676 //
5677 // The passed in time value is either a relative time in nanoseconds
5678 // or an absolute time in milliseconds. Either way it has to be unpacked
5679 // into suitable seconds and nanoseconds components and stored in the
5680 // given timespec structure.
5681 // Given time is a 64-bit value and the time_t used in the timespec is only
5682 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5683 // overflow if times way in the future are given. Further on Solaris versions
5684 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5685 // number of seconds, in abstime, is less than current_time + 100,000,000.
5686 // As it will be 28 years before "now + 100000000" will overflow we can
5687 // ignore overflow and just impose a hard-limit on seconds using the value
5688 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5689 // years from "now".
5690
5691 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5692 assert(time > 0, "convertTime");
5693 time_t max_secs = 0;
5694
5695 if (!os::supports_monotonic_clock() || isAbsolute) {
5696 struct timeval now;
5697 int status = gettimeofday(&now, NULL);
5698 assert(status == 0, "gettimeofday");
5699
5700 max_secs = now.tv_sec + MAX_SECS;
5701
5702 if (isAbsolute) {
5703 jlong secs = time / 1000;
5704 if (secs > max_secs) {
5705 absTime->tv_sec = max_secs;
5706 } else {
5707 absTime->tv_sec = secs;
5708 }
5709 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5710 } else {
5711 jlong secs = time / NANOSECS_PER_SEC;
5712 if (secs >= MAX_SECS) {
5713 absTime->tv_sec = max_secs;
5714 absTime->tv_nsec = 0;
5715 } else {
5716 absTime->tv_sec = now.tv_sec + secs;
5717 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5718 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5719 absTime->tv_nsec -= NANOSECS_PER_SEC;
5720 ++absTime->tv_sec; // note: this must be <= max_secs
5721 }
5722 }
5723 }
5724 } else {
5725 // must be relative using monotonic clock
5726 struct timespec now;
5727 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5728 assert_status(status == 0, status, "clock_gettime");
5729 max_secs = now.tv_sec + MAX_SECS;
5730 jlong secs = time / NANOSECS_PER_SEC;
5731 if (secs >= MAX_SECS) {
5732 absTime->tv_sec = max_secs;
5733 absTime->tv_nsec = 0;
5734 } else {
5735 absTime->tv_sec = now.tv_sec + secs;
5736 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5737 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5738 absTime->tv_nsec -= NANOSECS_PER_SEC;
5739 ++absTime->tv_sec; // note: this must be <= max_secs
5740 }
5741 }
5742 }
5743 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5744 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5745 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5746 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5747 }
5748
5749 void Parker::park(bool isAbsolute, jlong time) {
5750 // Ideally we'd do something useful while spinning, such
5751 // as calling unpackTime().
5752
5753 // Optional fast-path check:
5754 // Return immediately if a permit is available.
5755 // We depend on Atomic::xchg() having full barrier semantics
5756 // since we are doing a lock-free update to _counter.
5757 if (Atomic::xchg(0, &_counter) > 0) return;
5758
5759 Thread* thread = Thread::current();
5760 assert(thread->is_Java_thread(), "Must be JavaThread");
5761 JavaThread *jt = (JavaThread *)thread;
5762
5763 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5764 // Check interrupt before trying to wait
5765 if (Thread::is_interrupted(thread, false)) {
5766 return;
5767 }
5768
5769 // Next, demultiplex/decode time arguments
5770 timespec absTime;
5771 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5772 return;
5773 }
5774 if (time > 0) {
5775 unpackTime(&absTime, isAbsolute, time);
5776 }
5777
5778
5779 // Enter safepoint region
5780 // Beware of deadlocks such as 6317397.
5781 // The per-thread Parker:: mutex is a classic leaf-lock.
5782 // In particular a thread must never block on the Threads_lock while
5783 // holding the Parker:: mutex. If safepoints are pending both the
5784 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5785 ThreadBlockInVM tbivm(jt);
5786
5787 // Don't wait if cannot get lock since interference arises from
5788 // unblocking. Also. check interrupt before trying wait
5789 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5790 return;
5791 }
5792
5793 int status;
5794 if (_counter > 0) { // no wait needed
5795 _counter = 0;
5796 status = pthread_mutex_unlock(_mutex);
5797 assert(status == 0, "invariant");
5798 // Paranoia to ensure our locked and lock-free paths interact
5799 // correctly with each other and Java-level accesses.
5800 OrderAccess::fence();
5801 return;
5802 }
5803
5804 #ifdef ASSERT
5805 // Don't catch signals while blocked; let the running threads have the signals.
5806 // (This allows a debugger to break into the running thread.)
5807 sigset_t oldsigs;
5808 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5809 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5810 #endif
5811
5812 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5813 jt->set_suspend_equivalent();
5814 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5815
5816 assert(_cur_index == -1, "invariant");
5817 if (time == 0) {
5818 _cur_index = REL_INDEX; // arbitrary choice when not timed
5819 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5820 } else {
5821 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5822 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5823 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5824 pthread_cond_destroy(&_cond[_cur_index]);
5825 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5826 }
5827 }
5828 _cur_index = -1;
5829 assert_status(status == 0 || status == EINTR ||
5830 status == ETIME || status == ETIMEDOUT,
5831 status, "cond_timedwait");
5832
5833 #ifdef ASSERT
5834 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5835 #endif
5836
5837 _counter = 0;
5838 status = pthread_mutex_unlock(_mutex);
5839 assert_status(status == 0, status, "invariant");
5840 // Paranoia to ensure our locked and lock-free paths interact
5841 // correctly with each other and Java-level accesses.
5842 OrderAccess::fence();
5843
5844 // If externally suspended while waiting, re-suspend
5845 if (jt->handle_special_suspend_equivalent_condition()) {
5846 jt->java_suspend_self();
5847 }
5848 }
5849
5850 void Parker::unpark() {
5851 int status = pthread_mutex_lock(_mutex);
5852 assert(status == 0, "invariant");
5853 const int s = _counter;
5854 _counter = 1;
5855 if (s < 1) {
5856 // thread might be parked
5857 if (_cur_index != -1) {
5858 // thread is definitely parked
5859 if (WorkAroundNPTLTimedWaitHang) {
5860 status = pthread_cond_signal(&_cond[_cur_index]);
5861 assert(status == 0, "invariant");
5862 status = pthread_mutex_unlock(_mutex);
5863 assert(status == 0, "invariant");
5864 } else {
5865 status = pthread_mutex_unlock(_mutex);
5866 assert(status == 0, "invariant");
5867 status = pthread_cond_signal(&_cond[_cur_index]);
5868 assert(status == 0, "invariant");
5869 }
5870 } else {
5871 pthread_mutex_unlock(_mutex);
5872 assert(status == 0, "invariant");
5873 }
5874 } else {
5875 pthread_mutex_unlock(_mutex);
5876 assert(status == 0, "invariant");
5877 }
5878 }
5879
5880
5881 extern char** environ;
5882
5883 // Run the specified command in a separate process. Return its exit value,
5884 // or -1 on failure (e.g. can't fork a new process).
5885 // Unlike system(), this function can be called from signal handler. It
5886 // doesn't block SIGINT et al.
5887 int os::fork_and_exec(char* cmd) {
5888 const char * argv[4] = {"sh", "-c", cmd, NULL};
5889
5890 pid_t pid = fork();
5891
5892 if (pid < 0) {
5893 // fork failed
5894 return -1;
5895
5896 } else if (pid == 0) {
5897 // child process
5898
5899 execve("/bin/sh", (char* const*)argv, environ);
5900
5901 // execve failed
5902 _exit(-1);
5903
5904 } else {
5905 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5906 // care about the actual exit code, for now.
5907
5908 int status;
5909
5910 // Wait for the child process to exit. This returns immediately if
5911 // the child has already exited. */
5912 while (waitpid(pid, &status, 0) < 0) {
5913 switch (errno) {
5914 case ECHILD: return 0;
5915 case EINTR: break;
5916 default: return -1;
5917 }
5918 }
5919
5920 if (WIFEXITED(status)) {
5921 // The child exited normally; get its exit code.
5922 return WEXITSTATUS(status);
5923 } else if (WIFSIGNALED(status)) {
5924 // The child exited because of a signal
5925 // The best value to return is 0x80 + signal number,
5926 // because that is what all Unix shells do, and because
5927 // it allows callers to distinguish between process exit and
5928 // process death by signal.
5929 return 0x80 + WTERMSIG(status);
5930 } else {
5931 // Unknown exit code; pass it through
5932 return status;
5933 }
5934 }
5935 }
5936
5937 // is_headless_jre()
5938 //
5939 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5940 // in order to report if we are running in a headless jre
5941 //
5942 // Since JDK8 xawt/libmawt.so was moved into the same directory
5943 // as libawt.so, and renamed libawt_xawt.so
5944 //
5945 bool os::is_headless_jre() {
5946 struct stat statbuf;
5947 char buf[MAXPATHLEN];
5948 char libmawtpath[MAXPATHLEN];
5949 const char *xawtstr = "/xawt/libmawt.so";
5950 const char *new_xawtstr = "/libawt_xawt.so";
5951 char *p;
5952
5953 // Get path to libjvm.so
5954 os::jvm_path(buf, sizeof(buf));
5955
5956 // Get rid of libjvm.so
5957 p = strrchr(buf, '/');
5958 if (p == NULL) {
5959 return false;
5960 } else {
5961 *p = '\0';
5962 }
5963
5964 // Get rid of client or server
5965 p = strrchr(buf, '/');
5966 if (p == NULL) {
5967 return false;
5968 } else {
5969 *p = '\0';
5970 }
5971
5972 // check xawt/libmawt.so
5973 strcpy(libmawtpath, buf);
5974 strcat(libmawtpath, xawtstr);
5975 if (::stat(libmawtpath, &statbuf) == 0) return false;
5976
5977 // check libawt_xawt.so
5978 strcpy(libmawtpath, buf);
5979 strcat(libmawtpath, new_xawtstr);
5980 if (::stat(libmawtpath, &statbuf) == 0) return false;
5981
5982 return true;
5983 }
5984
5985 // Get the default path to the core file
5986 // Returns the length of the string
5987 int os::get_core_path(char* buffer, size_t bufferSize) {
5988 /*
5989 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5990 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5991 */
5992 const int core_pattern_len = 129;
5993 char core_pattern[core_pattern_len] = {0};
5994
5995 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
5996 if (core_pattern_file != -1) {
5997 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
5998 ::close(core_pattern_file);
5999
6000 if (ret > 0) {
6001 char *last_char = core_pattern + strlen(core_pattern) - 1;
6002
6003 if (*last_char == '\n') {
6004 *last_char = '\0';
6005 }
6006 }
6007 }
6008
6009 if (strlen(core_pattern) == 0) {
6010 return -1;
6011 }
6012
6013 char *pid_pos = strstr(core_pattern, "%p");
6014 int written;
6015
6016 if (core_pattern[0] == '/') {
6017 written = jio_snprintf(buffer, bufferSize, core_pattern);
6018 } else {
6019 char cwd[PATH_MAX];
6020
6021 const char* p = get_current_directory(cwd, PATH_MAX);
6022 if (p == NULL) {
6023 return -1;
6024 }
6025
6026 if (core_pattern[0] == '|') {
6027 written = jio_snprintf(buffer, bufferSize,
6028 "\"%s\" (or dumping to %s/core.%d)",
6029 &core_pattern[1], p, current_process_id());
6030 } else {
6031 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6032 }
6033 }
6034
6035 if (written < 0) {
6036 return -1;
6037 }
6038
6039 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6040 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6041
6042 if (core_uses_pid_file != -1) {
6043 char core_uses_pid = 0;
6044 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6045 ::close(core_uses_pid_file);
6046
6047 if (core_uses_pid == '1') {
6048 jio_snprintf(buffer + written, bufferSize - written,
6049 ".%d", current_process_id());
6050 }
6051 }
6052 }
6053
6054 return strlen(buffer);
6055 }
6056
6057 /////////////// Unit tests ///////////////
6058
6059 #ifndef PRODUCT
6060
6061 #define test_log(...) \
6062 do { \
6063 if (VerboseInternalVMTests) { \
6064 tty->print_cr(__VA_ARGS__); \
6065 tty->flush(); \
6066 } \
6067 } while (false)
6068
6069 class TestReserveMemorySpecial : AllStatic {
6070 public:
6071 static void small_page_write(void* addr, size_t size) {
6072 size_t page_size = os::vm_page_size();
6073
6074 char* end = (char*)addr + size;
6075 for (char* p = (char*)addr; p < end; p += page_size) {
6076 *p = 1;
6077 }
6078 }
6079
6080 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6081 if (!UseHugeTLBFS) {
6082 return;
6083 }
6084
6085 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6086
6087 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6088
6089 if (addr != NULL) {
6090 small_page_write(addr, size);
6091
6092 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6093 }
6094 }
6095
6096 static void test_reserve_memory_special_huge_tlbfs_only() {
6097 if (!UseHugeTLBFS) {
6098 return;
6099 }
6100
6101 size_t lp = os::large_page_size();
6102
6103 for (size_t size = lp; size <= lp * 10; size += lp) {
6104 test_reserve_memory_special_huge_tlbfs_only(size);
6105 }
6106 }
6107
6108 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6109 size_t lp = os::large_page_size();
6110 size_t ag = os::vm_allocation_granularity();
6111
6112 // sizes to test
6113 const size_t sizes[] = {
6114 lp, lp + ag, lp + lp / 2, lp * 2,
6115 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6116 lp * 10, lp * 10 + lp / 2
6117 };
6118 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6119
6120 // For each size/alignment combination, we test three scenarios:
6121 // 1) with req_addr == NULL
6122 // 2) with a non-null req_addr at which we expect to successfully allocate
6123 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6124 // expect the allocation to either fail or to ignore req_addr
6125
6126 // Pre-allocate two areas; they shall be as large as the largest allocation
6127 // and aligned to the largest alignment we will be testing.
6128 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6129 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6130 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6131 -1, 0);
6132 assert(mapping1 != MAP_FAILED, "should work");
6133
6134 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6135 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6136 -1, 0);
6137 assert(mapping2 != MAP_FAILED, "should work");
6138
6139 // Unmap the first mapping, but leave the second mapping intact: the first
6140 // mapping will serve as a value for a "good" req_addr (case 2). The second
6141 // mapping, still intact, as "bad" req_addr (case 3).
6142 ::munmap(mapping1, mapping_size);
6143
6144 // Case 1
6145 test_log("%s, req_addr NULL:", __FUNCTION__);
6146 test_log("size align result");
6147
6148 for (int i = 0; i < num_sizes; i++) {
6149 const size_t size = sizes[i];
6150 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6151 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6152 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6153 size, alignment, p, (p != NULL ? "" : "(failed)"));
6154 if (p != NULL) {
6155 assert(is_ptr_aligned(p, alignment), "must be");
6156 small_page_write(p, size);
6157 os::Linux::release_memory_special_huge_tlbfs(p, size);
6158 }
6159 }
6160 }
6161
6162 // Case 2
6163 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6164 test_log("size align req_addr result");
6165
6166 for (int i = 0; i < num_sizes; i++) {
6167 const size_t size = sizes[i];
6168 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6169 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6170 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6171 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6172 size, alignment, req_addr, p,
6173 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6174 if (p != NULL) {
6175 assert(p == req_addr, "must be");
6176 small_page_write(p, size);
6177 os::Linux::release_memory_special_huge_tlbfs(p, size);
6178 }
6179 }
6180 }
6181
6182 // Case 3
6183 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6184 test_log("size align req_addr result");
6185
6186 for (int i = 0; i < num_sizes; i++) {
6187 const size_t size = sizes[i];
6188 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6189 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6190 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6191 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6192 size, alignment, req_addr, p,
6193 ((p != NULL ? "" : "(failed)")));
6194 // as the area around req_addr contains already existing mappings, the API should always
6195 // return NULL (as per contract, it cannot return another address)
6196 assert(p == NULL, "must be");
6197 }
6198 }
6199
6200 ::munmap(mapping2, mapping_size);
6201
6202 }
6203
6204 static void test_reserve_memory_special_huge_tlbfs() {
6205 if (!UseHugeTLBFS) {
6206 return;
6207 }
6208
6209 test_reserve_memory_special_huge_tlbfs_only();
6210 test_reserve_memory_special_huge_tlbfs_mixed();
6211 }
6212
6213 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6214 if (!UseSHM) {
6215 return;
6216 }
6217
6218 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6219
6220 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6221
6222 if (addr != NULL) {
6223 assert(is_ptr_aligned(addr, alignment), "Check");
6224 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6225
6226 small_page_write(addr, size);
6227
6228 os::Linux::release_memory_special_shm(addr, size);
6229 }
6230 }
6231
6232 static void test_reserve_memory_special_shm() {
6233 size_t lp = os::large_page_size();
6234 size_t ag = os::vm_allocation_granularity();
6235
6236 for (size_t size = ag; size < lp * 3; size += ag) {
6237 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6238 test_reserve_memory_special_shm(size, alignment);
6239 }
6240 }
6241 }
6242
6243 static void test() {
6244 test_reserve_memory_special_huge_tlbfs();
6245 test_reserve_memory_special_shm();
6246 }
6247 };
6248
6249 void TestReserveMemorySpecial_test() {
6250 TestReserveMemorySpecial::test();
6251 }
6252
6253 #endif