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/vmError.hpp"
73
74 // put OS-includes here
75 # include <sys/types.h>
76 # include <sys/mman.h>
77 # include <sys/stat.h>
78 # include <sys/select.h>
79 # include <pthread.h>
80 # include <signal.h>
81 # include <errno.h>
82 # include <dlfcn.h>
83 # include <stdio.h>
84 # include <unistd.h>
85 # include <sys/resource.h>
86 # include <pthread.h>
87 # include <sys/stat.h>
88 # include <sys/time.h>
89 # include <sys/times.h>
90 # include <sys/utsname.h>
91 # include <sys/socket.h>
92 # include <sys/wait.h>
93 # include <pwd.h>
94 # include <poll.h>
95 # include <semaphore.h>
96 # include <fcntl.h>
97 # include <string.h>
98 # include <syscall.h>
99 # include <sys/sysinfo.h>
100 # include <gnu/libc-version.h>
101 # include <sys/ipc.h>
102 # include <sys/shm.h>
103 # include <link.h>
104 # include <stdint.h>
105 # include <inttypes.h>
106 # include <sys/ioctl.h>
107
108 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
109
110 // if RUSAGE_THREAD for getrusage() has not been defined, do it here. The code calling
111 // getrusage() is prepared to handle the associated failure.
112 #ifndef RUSAGE_THREAD
113 #define RUSAGE_THREAD (1) /* only the calling thread */
114 #endif
115
116 #define MAX_PATH (2 * K)
117
118 #define MAX_SECS 100000000
119
120 // for timer info max values which include all bits
121 #define ALL_64_BITS CONST64(0xFFFFFFFFFFFFFFFF)
122
123 #define LARGEPAGES_BIT (1 << 6)
124 ////////////////////////////////////////////////////////////////////////////////
125 // global variables
126 julong os::Linux::_physical_memory = 0;
127
128 address os::Linux::_initial_thread_stack_bottom = NULL;
129 uintptr_t os::Linux::_initial_thread_stack_size = 0;
130
131 int (*os::Linux::_clock_gettime)(clockid_t, struct timespec *) = NULL;
132 int (*os::Linux::_pthread_getcpuclockid)(pthread_t, clockid_t *) = NULL;
133 int (*os::Linux::_pthread_setname_np)(pthread_t, const char*) = NULL;
134 Mutex* os::Linux::_createThread_lock = NULL;
135 pthread_t os::Linux::_main_thread;
136 int os::Linux::_page_size = -1;
137 const int os::Linux::_vm_default_page_size = (8 * K);
138 bool os::Linux::_is_floating_stack = false;
139 bool os::Linux::_is_NPTL = false;
140 bool os::Linux::_supports_fast_thread_cpu_time = false;
141 const char * os::Linux::_glibc_version = NULL;
142 const char * os::Linux::_libpthread_version = NULL;
143 pthread_condattr_t os::Linux::_condattr[1];
144
145 static jlong initial_time_count=0;
146
147 static int clock_tics_per_sec = 100;
148
149 // For diagnostics to print a message once. see run_periodic_checks
150 static sigset_t check_signal_done;
151 static bool check_signals = true;
152
153 static pid_t _initial_pid = 0;
154
155 // Signal number used to suspend/resume a thread
156
157 // do not use any signal number less than SIGSEGV, see 4355769
158 static int SR_signum = SIGUSR2;
159 sigset_t SR_sigset;
160
161 // Declarations
162 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
163
164 // utility functions
165
166 static int SR_initialize();
167
168 julong os::available_memory() {
169 return Linux::available_memory();
170 }
171
172 julong os::Linux::available_memory() {
173 // values in struct sysinfo are "unsigned long"
174 struct sysinfo si;
175 sysinfo(&si);
176
177 return (julong)si.freeram * si.mem_unit;
178 }
179
180 julong os::physical_memory() {
181 return Linux::physical_memory();
182 }
183
184 // Return true if user is running as root.
185
186 bool os::have_special_privileges() {
187 static bool init = false;
188 static bool privileges = false;
189 if (!init) {
190 privileges = (getuid() != geteuid()) || (getgid() != getegid());
191 init = true;
192 }
193 return privileges;
194 }
195
196
197 #ifndef SYS_gettid
198 // i386: 224, ia64: 1105, amd64: 186, sparc 143
199 #ifdef __ia64__
200 #define SYS_gettid 1105
201 #else
202 #ifdef __i386__
203 #define SYS_gettid 224
204 #else
205 #ifdef __amd64__
206 #define SYS_gettid 186
207 #else
208 #ifdef __sparc__
209 #define SYS_gettid 143
210 #else
211 #error define gettid for the arch
212 #endif
213 #endif
214 #endif
215 #endif
216 #endif
217
218 // Cpu architecture string
219 static char cpu_arch[] = HOTSPOT_LIB_ARCH;
220
221
222 // pid_t gettid()
223 //
224 // Returns the kernel thread id of the currently running thread. Kernel
225 // thread id is used to access /proc.
226 //
227 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
228 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
229 //
230 pid_t os::Linux::gettid() {
231 int rslt = syscall(SYS_gettid);
232 if (rslt == -1) {
233 // old kernel, no NPTL support
234 return getpid();
235 } else {
236 return (pid_t)rslt;
237 }
238 }
239
240 // Most versions of linux have a bug where the number of processors are
241 // determined by looking at the /proc file system. In a chroot environment,
242 // the system call returns 1. This causes the VM to act as if it is
243 // a single processor and elide locking (see is_MP() call).
244 static bool unsafe_chroot_detected = false;
245 static const char *unstable_chroot_error = "/proc file system not found.\n"
246 "Java may be unstable running multithreaded in a chroot "
247 "environment on Linux when /proc filesystem is not mounted.";
248
249 void os::Linux::initialize_system_info() {
250 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
251 if (processor_count() == 1) {
252 pid_t pid = os::Linux::gettid();
253 char fname[32];
254 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
255 FILE *fp = fopen(fname, "r");
256 if (fp == NULL) {
257 unsafe_chroot_detected = true;
258 } else {
259 fclose(fp);
260 }
261 }
262 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
263 assert(processor_count() > 0, "linux error");
264 }
265
266 void os::init_system_properties_values() {
267 // The next steps are taken in the product version:
268 //
269 // Obtain the JAVA_HOME value from the location of libjvm.so.
270 // This library should be located at:
271 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
272 //
273 // If "/jre/lib/" appears at the right place in the path, then we
274 // assume libjvm.so is installed in a JDK and we use this path.
275 //
276 // Otherwise exit with message: "Could not create the Java virtual machine."
277 //
278 // The following extra steps are taken in the debugging version:
279 //
280 // If "/jre/lib/" does NOT appear at the right place in the path
281 // instead of exit check for $JAVA_HOME environment variable.
282 //
283 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
284 // then we append a fake suffix "hotspot/libjvm.so" to this path so
285 // it looks like libjvm.so is installed there
286 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
287 //
288 // Otherwise exit.
289 //
290 // Important note: if the location of libjvm.so changes this
291 // code needs to be changed accordingly.
292
293 // See ld(1):
294 // The linker uses the following search paths to locate required
295 // shared libraries:
296 // 1: ...
297 // ...
298 // 7: The default directories, normally /lib and /usr/lib.
299 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
300 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
301 #else
302 #define DEFAULT_LIBPATH "/lib:/usr/lib"
303 #endif
304
305 // Base path of extensions installed on the system.
306 #define SYS_EXT_DIR "/usr/java/packages"
307 #define EXTENSIONS_DIR "/lib/ext"
308
309 // Buffer that fits several sprintfs.
310 // Note that the space for the colon and the trailing null are provided
311 // by the nulls included by the sizeof operator.
312 const size_t bufsize =
313 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
314 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
315 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
316
317 // sysclasspath, java_home, dll_dir
318 {
319 char *pslash;
320 os::jvm_path(buf, bufsize);
321
322 // Found the full path to libjvm.so.
323 // Now cut the path to <java_home>/jre if we can.
324 pslash = strrchr(buf, '/');
325 if (pslash != NULL) {
326 *pslash = '\0'; // Get rid of /libjvm.so.
327 }
328 pslash = strrchr(buf, '/');
329 if (pslash != NULL) {
330 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
331 }
332 Arguments::set_dll_dir(buf);
333
334 if (pslash != NULL) {
335 pslash = strrchr(buf, '/');
336 if (pslash != NULL) {
337 *pslash = '\0'; // Get rid of /<arch>.
338 pslash = strrchr(buf, '/');
339 if (pslash != NULL) {
340 *pslash = '\0'; // Get rid of /lib.
341 }
342 }
343 }
344 Arguments::set_java_home(buf);
345 set_boot_path('/', ':');
346 }
347
348 // Where to look for native libraries.
349 //
350 // Note: Due to a legacy implementation, most of the library path
351 // is set in the launcher. This was to accomodate linking restrictions
352 // on legacy Linux implementations (which are no longer supported).
353 // Eventually, all the library path setting will be done here.
354 //
355 // However, to prevent the proliferation of improperly built native
356 // libraries, the new path component /usr/java/packages is added here.
357 // Eventually, all the library path setting will be done here.
358 {
359 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
360 // should always exist (until the legacy problem cited above is
361 // addressed).
362 const char *v = ::getenv("LD_LIBRARY_PATH");
363 const char *v_colon = ":";
364 if (v == NULL) { v = ""; v_colon = ""; }
365 // That's +1 for the colon and +1 for the trailing '\0'.
366 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
367 strlen(v) + 1 +
368 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
369 mtInternal);
370 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
371 Arguments::set_library_path(ld_library_path);
372 FREE_C_HEAP_ARRAY(char, ld_library_path);
373 }
374
375 // Extensions directories.
376 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
377 Arguments::set_ext_dirs(buf);
378
379 FREE_C_HEAP_ARRAY(char, buf);
380
381 #undef DEFAULT_LIBPATH
382 #undef SYS_EXT_DIR
383 #undef EXTENSIONS_DIR
384 }
385
386 ////////////////////////////////////////////////////////////////////////////////
387 // breakpoint support
388
389 void os::breakpoint() {
390 BREAKPOINT;
391 }
392
393 extern "C" void breakpoint() {
394 // use debugger to set breakpoint here
395 }
396
397 ////////////////////////////////////////////////////////////////////////////////
398 // signal support
399
400 debug_only(static bool signal_sets_initialized = false);
401 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
402
403 bool os::Linux::is_sig_ignored(int sig) {
404 struct sigaction oact;
405 sigaction(sig, (struct sigaction*)NULL, &oact);
406 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
407 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
408 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
409 return true;
410 } else {
411 return false;
412 }
413 }
414
415 void os::Linux::signal_sets_init() {
416 // Should also have an assertion stating we are still single-threaded.
417 assert(!signal_sets_initialized, "Already initialized");
418 // Fill in signals that are necessarily unblocked for all threads in
419 // the VM. Currently, we unblock the following signals:
420 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
421 // by -Xrs (=ReduceSignalUsage));
422 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
423 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
424 // the dispositions or masks wrt these signals.
425 // Programs embedding the VM that want to use the above signals for their
426 // own purposes must, at this time, use the "-Xrs" option to prevent
427 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
428 // (See bug 4345157, and other related bugs).
429 // In reality, though, unblocking these signals is really a nop, since
430 // these signals are not blocked by default.
431 sigemptyset(&unblocked_sigs);
432 sigemptyset(&allowdebug_blocked_sigs);
433 sigaddset(&unblocked_sigs, SIGILL);
434 sigaddset(&unblocked_sigs, SIGSEGV);
435 sigaddset(&unblocked_sigs, SIGBUS);
436 sigaddset(&unblocked_sigs, SIGFPE);
437 #if defined(PPC64)
438 sigaddset(&unblocked_sigs, SIGTRAP);
439 #endif
440 sigaddset(&unblocked_sigs, SR_signum);
441
442 if (!ReduceSignalUsage) {
443 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
444 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
445 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
446 }
447 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
448 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
449 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
450 }
451 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
452 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
453 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
454 }
455 }
456 // Fill in signals that are blocked by all but the VM thread.
457 sigemptyset(&vm_sigs);
458 if (!ReduceSignalUsage) {
459 sigaddset(&vm_sigs, BREAK_SIGNAL);
460 }
461 debug_only(signal_sets_initialized = true);
462
463 }
464
465 // These are signals that are unblocked while a thread is running Java.
466 // (For some reason, they get blocked by default.)
467 sigset_t* os::Linux::unblocked_signals() {
468 assert(signal_sets_initialized, "Not initialized");
469 return &unblocked_sigs;
470 }
471
472 // These are the signals that are blocked while a (non-VM) thread is
473 // running Java. Only the VM thread handles these signals.
474 sigset_t* os::Linux::vm_signals() {
475 assert(signal_sets_initialized, "Not initialized");
476 return &vm_sigs;
477 }
478
479 // These are signals that are blocked during cond_wait to allow debugger in
480 sigset_t* os::Linux::allowdebug_blocked_signals() {
481 assert(signal_sets_initialized, "Not initialized");
482 return &allowdebug_blocked_sigs;
483 }
484
485 void os::Linux::hotspot_sigmask(Thread* thread) {
486
487 //Save caller's signal mask before setting VM signal mask
488 sigset_t caller_sigmask;
489 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
490
491 OSThread* osthread = thread->osthread();
492 osthread->set_caller_sigmask(caller_sigmask);
493
494 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
495
496 if (!ReduceSignalUsage) {
497 if (thread->is_VM_thread()) {
498 // Only the VM thread handles BREAK_SIGNAL ...
499 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
500 } else {
501 // ... all other threads block BREAK_SIGNAL
502 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
503 }
504 }
505 }
506
507 //////////////////////////////////////////////////////////////////////////////
508 // detecting pthread library
509
510 void os::Linux::libpthread_init() {
511 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
512 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
513 // generic name for earlier versions.
514 // Define macros here so we can build HotSpot on old systems.
515 #ifndef _CS_GNU_LIBC_VERSION
516 #define _CS_GNU_LIBC_VERSION 2
517 #endif
518 #ifndef _CS_GNU_LIBPTHREAD_VERSION
519 #define _CS_GNU_LIBPTHREAD_VERSION 3
520 #endif
521
522 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
523 if (n > 0) {
524 char *str = (char *)malloc(n, mtInternal);
525 confstr(_CS_GNU_LIBC_VERSION, str, n);
526 os::Linux::set_glibc_version(str);
527 } else {
528 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
529 static char _gnu_libc_version[32];
530 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
531 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
532 os::Linux::set_glibc_version(_gnu_libc_version);
533 }
534
535 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
536 if (n > 0) {
537 char *str = (char *)malloc(n, mtInternal);
538 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
539 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
540 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
541 // is the case. LinuxThreads has a hard limit on max number of threads.
542 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
543 // On the other hand, NPTL does not have such a limit, sysconf()
544 // will return -1 and errno is not changed. Check if it is really NPTL.
545 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
546 strstr(str, "NPTL") &&
547 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
548 free(str);
549 os::Linux::set_libpthread_version("linuxthreads");
550 } else {
551 os::Linux::set_libpthread_version(str);
552 }
553 } else {
554 // glibc before 2.3.2 only has LinuxThreads.
555 os::Linux::set_libpthread_version("linuxthreads");
556 }
557
558 if (strstr(libpthread_version(), "NPTL")) {
559 os::Linux::set_is_NPTL();
560 } else {
561 os::Linux::set_is_LinuxThreads();
562 }
563
564 // LinuxThreads have two flavors: floating-stack mode, which allows variable
565 // stack size; and fixed-stack mode. NPTL is always floating-stack.
566 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
567 os::Linux::set_is_floating_stack();
568 }
569 }
570
571 /////////////////////////////////////////////////////////////////////////////
572 // thread stack
573
574 // Force Linux kernel to expand current thread stack. If "bottom" is close
575 // to the stack guard, caller should block all signals.
576 //
577 // MAP_GROWSDOWN:
578 // A special mmap() flag that is used to implement thread stacks. It tells
579 // kernel that the memory region should extend downwards when needed. This
580 // allows early versions of LinuxThreads to only mmap the first few pages
581 // when creating a new thread. Linux kernel will automatically expand thread
582 // stack as needed (on page faults).
583 //
584 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
585 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
586 // region, it's hard to tell if the fault is due to a legitimate stack
587 // access or because of reading/writing non-exist memory (e.g. buffer
588 // overrun). As a rule, if the fault happens below current stack pointer,
589 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
590 // application (see Linux kernel fault.c).
591 //
592 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
593 // stack overflow detection.
594 //
595 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
596 // not use this flag. However, the stack of initial thread is not created
597 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
598 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
599 // and then attach the thread to JVM.
600 //
601 // To get around the problem and allow stack banging on Linux, we need to
602 // manually expand thread stack after receiving the SIGSEGV.
603 //
604 // There are two ways to expand thread stack to address "bottom", we used
605 // both of them in JVM before 1.5:
606 // 1. adjust stack pointer first so that it is below "bottom", and then
607 // touch "bottom"
608 // 2. mmap() the page in question
609 //
610 // Now alternate signal stack is gone, it's harder to use 2. For instance,
611 // if current sp is already near the lower end of page 101, and we need to
612 // call mmap() to map page 100, it is possible that part of the mmap() frame
613 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
614 // That will destroy the mmap() frame and cause VM to crash.
615 //
616 // The following code works by adjusting sp first, then accessing the "bottom"
617 // page to force a page fault. Linux kernel will then automatically expand the
618 // stack mapping.
619 //
620 // _expand_stack_to() assumes its frame size is less than page size, which
621 // should always be true if the function is not inlined.
622
623 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
624 #define NOINLINE
625 #else
626 #define NOINLINE __attribute__ ((noinline))
627 #endif
628
629 static void _expand_stack_to(address bottom) NOINLINE;
630
631 static void _expand_stack_to(address bottom) {
632 address sp;
633 size_t size;
634 volatile char *p;
635
636 // Adjust bottom to point to the largest address within the same page, it
637 // gives us a one-page buffer if alloca() allocates slightly more memory.
638 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
639 bottom += os::Linux::page_size() - 1;
640
641 // sp might be slightly above current stack pointer; if that's the case, we
642 // will alloca() a little more space than necessary, which is OK. Don't use
643 // os::current_stack_pointer(), as its result can be slightly below current
644 // stack pointer, causing us to not alloca enough to reach "bottom".
645 sp = (address)&sp;
646
647 if (sp > bottom) {
648 size = sp - bottom;
649 p = (volatile char *)alloca(size);
650 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
651 p[0] = '\0';
652 }
653 }
654
655 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
656 assert(t!=NULL, "just checking");
657 assert(t->osthread()->expanding_stack(), "expand should be set");
658 assert(t->stack_base() != NULL, "stack_base was not initialized");
659
660 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
661 sigset_t mask_all, old_sigset;
662 sigfillset(&mask_all);
663 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
664 _expand_stack_to(addr);
665 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
666 return true;
667 }
668 return false;
669 }
670
671 //////////////////////////////////////////////////////////////////////////////
672 // create new thread
673
674 static address highest_vm_reserved_address();
675
676 // check if it's safe to start a new thread
677 static bool _thread_safety_check(Thread* thread) {
678 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
679 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
680 // Heap is mmap'ed at lower end of memory space. Thread stacks are
681 // allocated (MAP_FIXED) from high address space. Every thread stack
682 // occupies a fixed size slot (usually 2Mbytes, but user can change
683 // it to other values if they rebuild LinuxThreads).
684 //
685 // Problem with MAP_FIXED is that mmap() can still succeed even part of
686 // the memory region has already been mmap'ed. That means if we have too
687 // many threads and/or very large heap, eventually thread stack will
688 // collide with heap.
689 //
690 // Here we try to prevent heap/stack collision by comparing current
691 // stack bottom with the highest address that has been mmap'ed by JVM
692 // plus a safety margin for memory maps created by native code.
693 //
694 // This feature can be disabled by setting ThreadSafetyMargin to 0
695 //
696 if (ThreadSafetyMargin > 0) {
697 address stack_bottom = os::current_stack_base() - os::current_stack_size();
698
699 // not safe if our stack extends below the safety margin
700 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
701 } else {
702 return true;
703 }
704 } else {
705 // Floating stack LinuxThreads or NPTL:
706 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
707 // there's not enough space left, pthread_create() will fail. If we come
708 // here, that means enough space has been reserved for stack.
709 return true;
710 }
711 }
712
713 // Thread start routine for all newly created threads
714 static void *java_start(Thread *thread) {
715 // Try to randomize the cache line index of hot stack frames.
716 // This helps when threads of the same stack traces evict each other's
717 // cache lines. The threads can be either from the same JVM instance, or
718 // from different JVM instances. The benefit is especially true for
719 // processors with hyperthreading technology.
720 static int counter = 0;
721 int pid = os::current_process_id();
722 alloca(((pid ^ counter++) & 7) * 128);
723
724 ThreadLocalStorage::set_thread(thread);
725
726 OSThread* osthread = thread->osthread();
727 Monitor* sync = osthread->startThread_lock();
728
729 // non floating stack LinuxThreads needs extra check, see above
730 if (!_thread_safety_check(thread)) {
731 // notify parent thread
732 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
733 osthread->set_state(ZOMBIE);
734 sync->notify_all();
735 return NULL;
736 }
737
738 // thread_id is kernel thread id (similar to Solaris LWP id)
739 osthread->set_thread_id(os::Linux::gettid());
740
741 if (UseNUMA) {
742 int lgrp_id = os::numa_get_group_id();
743 if (lgrp_id != -1) {
744 thread->set_lgrp_id(lgrp_id);
745 }
746 }
747 // initialize signal mask for this thread
748 os::Linux::hotspot_sigmask(thread);
749
750 // initialize floating point control register
751 os::Linux::init_thread_fpu_state();
752
753 // handshaking with parent thread
754 {
755 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
756
757 // notify parent thread
758 osthread->set_state(INITIALIZED);
759 sync->notify_all();
760
761 // wait until os::start_thread()
762 while (osthread->get_state() == INITIALIZED) {
763 sync->wait(Mutex::_no_safepoint_check_flag);
764 }
765 }
766
767 // call one more level start routine
768 thread->run();
769
770 return 0;
771 }
772
773 bool os::create_thread(Thread* thread, ThreadType thr_type,
774 size_t stack_size) {
775 assert(thread->osthread() == NULL, "caller responsible");
776
777 // Allocate the OSThread object
778 OSThread* osthread = new OSThread(NULL, NULL);
779 if (osthread == NULL) {
780 return false;
781 }
782
783 // set the correct thread state
784 osthread->set_thread_type(thr_type);
785
786 // Initial state is ALLOCATED but not INITIALIZED
787 osthread->set_state(ALLOCATED);
788
789 thread->set_osthread(osthread);
790
791 // init thread attributes
792 pthread_attr_t attr;
793 pthread_attr_init(&attr);
794 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
795
796 // stack size
797 if (os::Linux::supports_variable_stack_size()) {
798 // calculate stack size if it's not specified by caller
799 if (stack_size == 0) {
800 stack_size = os::Linux::default_stack_size(thr_type);
801
802 switch (thr_type) {
803 case os::java_thread:
804 // Java threads use ThreadStackSize which default value can be
805 // changed with the flag -Xss
806 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
807 stack_size = JavaThread::stack_size_at_create();
808 break;
809 case os::compiler_thread:
810 if (CompilerThreadStackSize > 0) {
811 stack_size = (size_t)(CompilerThreadStackSize * K);
812 break;
813 } // else fall through:
814 // use VMThreadStackSize if CompilerThreadStackSize is not defined
815 case os::vm_thread:
816 case os::pgc_thread:
817 case os::cgc_thread:
818 case os::watcher_thread:
819 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
820 break;
821 }
822 }
823
824 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
825 pthread_attr_setstacksize(&attr, stack_size);
826 } else {
827 // let pthread_create() pick the default value.
828 }
829
830 // glibc guard page
831 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
832
833 ThreadState state;
834
835 {
836 // Serialize thread creation if we are running with fixed stack LinuxThreads
837 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
838 if (lock) {
839 os::Linux::createThread_lock()->lock_without_safepoint_check();
840 }
841
842 pthread_t tid;
843 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
844
845 pthread_attr_destroy(&attr);
846
847 if (ret != 0) {
848 if (PrintMiscellaneous && (Verbose || WizardMode)) {
849 perror("pthread_create()");
850 }
851 // Need to clean up stuff we've allocated so far
852 thread->set_osthread(NULL);
853 delete osthread;
854 if (lock) os::Linux::createThread_lock()->unlock();
855 return false;
856 }
857
858 // Store pthread info into the OSThread
859 osthread->set_pthread_id(tid);
860
861 // Wait until child thread is either initialized or aborted
862 {
863 Monitor* sync_with_child = osthread->startThread_lock();
864 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
865 while ((state = osthread->get_state()) == ALLOCATED) {
866 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
867 }
868 }
869
870 if (lock) {
871 os::Linux::createThread_lock()->unlock();
872 }
873 }
874
875 // Aborted due to thread limit being reached
876 if (state == ZOMBIE) {
877 thread->set_osthread(NULL);
878 delete osthread;
879 return false;
880 }
881
882 // The thread is returned suspended (in state INITIALIZED),
883 // and is started higher up in the call chain
884 assert(state == INITIALIZED, "race condition");
885 return true;
886 }
887
888 /////////////////////////////////////////////////////////////////////////////
889 // attach existing thread
890
891 // bootstrap the main thread
892 bool os::create_main_thread(JavaThread* thread) {
893 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
894 return create_attached_thread(thread);
895 }
896
897 bool os::create_attached_thread(JavaThread* thread) {
898 #ifdef ASSERT
899 thread->verify_not_published();
900 #endif
901
902 // Allocate the OSThread object
903 OSThread* osthread = new OSThread(NULL, NULL);
904
905 if (osthread == NULL) {
906 return false;
907 }
908
909 // Store pthread info into the OSThread
910 osthread->set_thread_id(os::Linux::gettid());
911 osthread->set_pthread_id(::pthread_self());
912
913 // initialize floating point control register
914 os::Linux::init_thread_fpu_state();
915
916 // Initial thread state is RUNNABLE
917 osthread->set_state(RUNNABLE);
918
919 thread->set_osthread(osthread);
920
921 if (UseNUMA) {
922 int lgrp_id = os::numa_get_group_id();
923 if (lgrp_id != -1) {
924 thread->set_lgrp_id(lgrp_id);
925 }
926 }
927
928 if (os::Linux::is_initial_thread()) {
929 // If current thread is initial thread, its stack is mapped on demand,
930 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
931 // the entire stack region to avoid SEGV in stack banging.
932 // It is also useful to get around the heap-stack-gap problem on SuSE
933 // kernel (see 4821821 for details). We first expand stack to the top
934 // of yellow zone, then enable stack yellow zone (order is significant,
935 // enabling yellow zone first will crash JVM on SuSE Linux), so there
936 // is no gap between the last two virtual memory regions.
937
938 JavaThread *jt = (JavaThread *)thread;
939 address addr = jt->stack_yellow_zone_base();
940 assert(addr != NULL, "initialization problem?");
941 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
942
943 osthread->set_expanding_stack();
944 os::Linux::manually_expand_stack(jt, addr);
945 osthread->clear_expanding_stack();
946 }
947
948 // initialize signal mask for this thread
949 // and save the caller's signal mask
950 os::Linux::hotspot_sigmask(thread);
951
952 return true;
953 }
954
955 void os::pd_start_thread(Thread* thread) {
956 OSThread * osthread = thread->osthread();
957 assert(osthread->get_state() != INITIALIZED, "just checking");
958 Monitor* sync_with_child = osthread->startThread_lock();
959 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
960 sync_with_child->notify();
961 }
962
963 // Free Linux resources related to the OSThread
964 void os::free_thread(OSThread* osthread) {
965 assert(osthread != NULL, "osthread not set");
966
967 if (Thread::current()->osthread() == osthread) {
968 // Restore caller's signal mask
969 sigset_t sigmask = osthread->caller_sigmask();
970 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
971 }
972
973 delete osthread;
974 }
975
976 //////////////////////////////////////////////////////////////////////////////
977 // thread local storage
978
979 // Restore the thread pointer if the destructor is called. This is in case
980 // someone from JNI code sets up a destructor with pthread_key_create to run
981 // detachCurrentThread on thread death. Unless we restore the thread pointer we
982 // will hang or crash. When detachCurrentThread is called the key will be set
983 // to null and we will not be called again. If detachCurrentThread is never
984 // called we could loop forever depending on the pthread implementation.
985 static void restore_thread_pointer(void* p) {
986 Thread* thread = (Thread*) p;
987 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
988 }
989
990 int os::allocate_thread_local_storage() {
991 pthread_key_t key;
992 int rslt = pthread_key_create(&key, restore_thread_pointer);
993 assert(rslt == 0, "cannot allocate thread local storage");
994 return (int)key;
995 }
996
997 // Note: This is currently not used by VM, as we don't destroy TLS key
998 // on VM exit.
999 void os::free_thread_local_storage(int index) {
1000 int rslt = pthread_key_delete((pthread_key_t)index);
1001 assert(rslt == 0, "invalid index");
1002 }
1003
1004 void os::thread_local_storage_at_put(int index, void* value) {
1005 int rslt = pthread_setspecific((pthread_key_t)index, value);
1006 assert(rslt == 0, "pthread_setspecific failed");
1007 }
1008
1009 extern "C" Thread* get_thread() {
1010 return ThreadLocalStorage::thread();
1011 }
1012
1013 //////////////////////////////////////////////////////////////////////////////
1014 // initial thread
1015
1016 // Check if current thread is the initial thread, similar to Solaris thr_main.
1017 bool os::Linux::is_initial_thread(void) {
1018 char dummy;
1019 // If called before init complete, thread stack bottom will be null.
1020 // Can be called if fatal error occurs before initialization.
1021 if (initial_thread_stack_bottom() == NULL) return false;
1022 assert(initial_thread_stack_bottom() != NULL &&
1023 initial_thread_stack_size() != 0,
1024 "os::init did not locate initial thread's stack region");
1025 if ((address)&dummy >= initial_thread_stack_bottom() &&
1026 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1027 return true;
1028 } else {
1029 return false;
1030 }
1031 }
1032
1033 // Find the virtual memory area that contains addr
1034 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1035 FILE *fp = fopen("/proc/self/maps", "r");
1036 if (fp) {
1037 address low, high;
1038 while (!feof(fp)) {
1039 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1040 if (low <= addr && addr < high) {
1041 if (vma_low) *vma_low = low;
1042 if (vma_high) *vma_high = high;
1043 fclose(fp);
1044 return true;
1045 }
1046 }
1047 for (;;) {
1048 int ch = fgetc(fp);
1049 if (ch == EOF || ch == (int)'\n') break;
1050 }
1051 }
1052 fclose(fp);
1053 }
1054 return false;
1055 }
1056
1057 // Locate initial thread stack. This special handling of initial thread stack
1058 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1059 // bogus value for initial thread.
1060 void os::Linux::capture_initial_stack(size_t max_size) {
1061 // stack size is the easy part, get it from RLIMIT_STACK
1062 size_t stack_size;
1063 struct rlimit rlim;
1064 getrlimit(RLIMIT_STACK, &rlim);
1065 stack_size = rlim.rlim_cur;
1066
1067 // 6308388: a bug in ld.so will relocate its own .data section to the
1068 // lower end of primordial stack; reduce ulimit -s value a little bit
1069 // so we won't install guard page on ld.so's data section.
1070 stack_size -= 2 * page_size();
1071
1072 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1073 // 7.1, in both cases we will get 2G in return value.
1074 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1075 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1076 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1077 // in case other parts in glibc still assumes 2M max stack size.
1078 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1079 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1080 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1081 stack_size = 2 * K * K IA64_ONLY(*2);
1082 }
1083 // Try to figure out where the stack base (top) is. This is harder.
1084 //
1085 // When an application is started, glibc saves the initial stack pointer in
1086 // a global variable "__libc_stack_end", which is then used by system
1087 // libraries. __libc_stack_end should be pretty close to stack top. The
1088 // variable is available since the very early days. However, because it is
1089 // a private interface, it could disappear in the future.
1090 //
1091 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1092 // to __libc_stack_end, it is very close to stack top, but isn't the real
1093 // stack top. Note that /proc may not exist if VM is running as a chroot
1094 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1095 // /proc/<pid>/stat could change in the future (though unlikely).
1096 //
1097 // We try __libc_stack_end first. If that doesn't work, look for
1098 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1099 // as a hint, which should work well in most cases.
1100
1101 uintptr_t stack_start;
1102
1103 // try __libc_stack_end first
1104 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1105 if (p && *p) {
1106 stack_start = *p;
1107 } else {
1108 // see if we can get the start_stack field from /proc/self/stat
1109 FILE *fp;
1110 int pid;
1111 char state;
1112 int ppid;
1113 int pgrp;
1114 int session;
1115 int nr;
1116 int tpgrp;
1117 unsigned long flags;
1118 unsigned long minflt;
1119 unsigned long cminflt;
1120 unsigned long majflt;
1121 unsigned long cmajflt;
1122 unsigned long utime;
1123 unsigned long stime;
1124 long cutime;
1125 long cstime;
1126 long prio;
1127 long nice;
1128 long junk;
1129 long it_real;
1130 uintptr_t start;
1131 uintptr_t vsize;
1132 intptr_t rss;
1133 uintptr_t rsslim;
1134 uintptr_t scodes;
1135 uintptr_t ecode;
1136 int i;
1137
1138 // Figure what the primordial thread stack base is. Code is inspired
1139 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1140 // followed by command name surrounded by parentheses, state, etc.
1141 char stat[2048];
1142 int statlen;
1143
1144 fp = fopen("/proc/self/stat", "r");
1145 if (fp) {
1146 statlen = fread(stat, 1, 2047, fp);
1147 stat[statlen] = '\0';
1148 fclose(fp);
1149
1150 // Skip pid and the command string. Note that we could be dealing with
1151 // weird command names, e.g. user could decide to rename java launcher
1152 // to "java 1.4.2 :)", then the stat file would look like
1153 // 1234 (java 1.4.2 :)) R ... ...
1154 // We don't really need to know the command string, just find the last
1155 // occurrence of ")" and then start parsing from there. See bug 4726580.
1156 char * s = strrchr(stat, ')');
1157
1158 i = 0;
1159 if (s) {
1160 // Skip blank chars
1161 do { s++; } while (s && isspace(*s));
1162
1163 #define _UFM UINTX_FORMAT
1164 #define _DFM INTX_FORMAT
1165
1166 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1167 // 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
1168 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,
1169 &state, // 3 %c
1170 &ppid, // 4 %d
1171 &pgrp, // 5 %d
1172 &session, // 6 %d
1173 &nr, // 7 %d
1174 &tpgrp, // 8 %d
1175 &flags, // 9 %lu
1176 &minflt, // 10 %lu
1177 &cminflt, // 11 %lu
1178 &majflt, // 12 %lu
1179 &cmajflt, // 13 %lu
1180 &utime, // 14 %lu
1181 &stime, // 15 %lu
1182 &cutime, // 16 %ld
1183 &cstime, // 17 %ld
1184 &prio, // 18 %ld
1185 &nice, // 19 %ld
1186 &junk, // 20 %ld
1187 &it_real, // 21 %ld
1188 &start, // 22 UINTX_FORMAT
1189 &vsize, // 23 UINTX_FORMAT
1190 &rss, // 24 INTX_FORMAT
1191 &rsslim, // 25 UINTX_FORMAT
1192 &scodes, // 26 UINTX_FORMAT
1193 &ecode, // 27 UINTX_FORMAT
1194 &stack_start); // 28 UINTX_FORMAT
1195 }
1196
1197 #undef _UFM
1198 #undef _DFM
1199
1200 if (i != 28 - 2) {
1201 assert(false, "Bad conversion from /proc/self/stat");
1202 // product mode - assume we are the initial thread, good luck in the
1203 // embedded case.
1204 warning("Can't detect initial thread stack location - bad conversion");
1205 stack_start = (uintptr_t) &rlim;
1206 }
1207 } else {
1208 // For some reason we can't open /proc/self/stat (for example, running on
1209 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1210 // most cases, so don't abort:
1211 warning("Can't detect initial thread stack location - no /proc/self/stat");
1212 stack_start = (uintptr_t) &rlim;
1213 }
1214 }
1215
1216 // Now we have a pointer (stack_start) very close to the stack top, the
1217 // next thing to do is to figure out the exact location of stack top. We
1218 // can find out the virtual memory area that contains stack_start by
1219 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1220 // and its upper limit is the real stack top. (again, this would fail if
1221 // running inside chroot, because /proc may not exist.)
1222
1223 uintptr_t stack_top;
1224 address low, high;
1225 if (find_vma((address)stack_start, &low, &high)) {
1226 // success, "high" is the true stack top. (ignore "low", because initial
1227 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1228 stack_top = (uintptr_t)high;
1229 } else {
1230 // failed, likely because /proc/self/maps does not exist
1231 warning("Can't detect initial thread stack location - find_vma failed");
1232 // best effort: stack_start is normally within a few pages below the real
1233 // stack top, use it as stack top, and reduce stack size so we won't put
1234 // guard page outside stack.
1235 stack_top = stack_start;
1236 stack_size -= 16 * page_size();
1237 }
1238
1239 // stack_top could be partially down the page so align it
1240 stack_top = align_size_up(stack_top, page_size());
1241
1242 if (max_size && stack_size > max_size) {
1243 _initial_thread_stack_size = max_size;
1244 } else {
1245 _initial_thread_stack_size = stack_size;
1246 }
1247
1248 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1249 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1250 }
1251
1252 ////////////////////////////////////////////////////////////////////////////////
1253 // time support
1254
1255 // Time since start-up in seconds to a fine granularity.
1256 // Used by VMSelfDestructTimer and the MemProfiler.
1257 double os::elapsedTime() {
1258
1259 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1260 }
1261
1262 jlong os::elapsed_counter() {
1263 return javaTimeNanos() - initial_time_count;
1264 }
1265
1266 jlong os::elapsed_frequency() {
1267 return NANOSECS_PER_SEC; // nanosecond resolution
1268 }
1269
1270 bool os::supports_vtime() { return true; }
1271 bool os::enable_vtime() { return false; }
1272 bool os::vtime_enabled() { return false; }
1273
1274 double os::elapsedVTime() {
1275 struct rusage usage;
1276 int retval = getrusage(RUSAGE_THREAD, &usage);
1277 if (retval == 0) {
1278 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);
1279 } else {
1280 // better than nothing, but not much
1281 return elapsedTime();
1282 }
1283 }
1284
1285 jlong os::javaTimeMillis() {
1286 timeval time;
1287 int status = gettimeofday(&time, NULL);
1288 assert(status != -1, "linux error");
1289 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1290 }
1291
1292 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1293 timeval time;
1294 int status = gettimeofday(&time, NULL);
1295 assert(status != -1, "linux error");
1296 seconds = jlong(time.tv_sec);
1297 nanos = jlong(time.tv_usec) * 1000;
1298 }
1299
1300
1301 #ifndef CLOCK_MONOTONIC
1302 #define CLOCK_MONOTONIC (1)
1303 #endif
1304
1305 void os::Linux::clock_init() {
1306 // we do dlopen's in this particular order due to bug in linux
1307 // dynamical loader (see 6348968) leading to crash on exit
1308 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1309 if (handle == NULL) {
1310 handle = dlopen("librt.so", RTLD_LAZY);
1311 }
1312
1313 if (handle) {
1314 int (*clock_getres_func)(clockid_t, struct timespec*) =
1315 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1316 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1317 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1318 if (clock_getres_func && clock_gettime_func) {
1319 // See if monotonic clock is supported by the kernel. Note that some
1320 // early implementations simply return kernel jiffies (updated every
1321 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1322 // for nano time (though the monotonic property is still nice to have).
1323 // It's fixed in newer kernels, however clock_getres() still returns
1324 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1325 // resolution for now. Hopefully as people move to new kernels, this
1326 // won't be a problem.
1327 struct timespec res;
1328 struct timespec tp;
1329 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1330 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1331 // yes, monotonic clock is supported
1332 _clock_gettime = clock_gettime_func;
1333 return;
1334 } else {
1335 // close librt if there is no monotonic clock
1336 dlclose(handle);
1337 }
1338 }
1339 }
1340 warning("No monotonic clock was available - timed services may " \
1341 "be adversely affected if the time-of-day clock changes");
1342 }
1343
1344 #ifndef SYS_clock_getres
1345 #if defined(IA32) || defined(AMD64)
1346 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1347 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1348 #else
1349 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1350 #define sys_clock_getres(x,y) -1
1351 #endif
1352 #else
1353 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1354 #endif
1355
1356 void os::Linux::fast_thread_clock_init() {
1357 if (!UseLinuxPosixThreadCPUClocks) {
1358 return;
1359 }
1360 clockid_t clockid;
1361 struct timespec tp;
1362 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1363 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1364
1365 // Switch to using fast clocks for thread cpu time if
1366 // the sys_clock_getres() returns 0 error code.
1367 // Note, that some kernels may support the current thread
1368 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1369 // returned by the pthread_getcpuclockid().
1370 // If the fast Posix clocks are supported then the sys_clock_getres()
1371 // must return at least tp.tv_sec == 0 which means a resolution
1372 // better than 1 sec. This is extra check for reliability.
1373
1374 if (pthread_getcpuclockid_func &&
1375 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1376 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1377 _supports_fast_thread_cpu_time = true;
1378 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1379 }
1380 }
1381
1382 jlong os::javaTimeNanos() {
1383 if (os::supports_monotonic_clock()) {
1384 struct timespec tp;
1385 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1386 assert(status == 0, "gettime error");
1387 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1388 return result;
1389 } else {
1390 timeval time;
1391 int status = gettimeofday(&time, NULL);
1392 assert(status != -1, "linux error");
1393 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1394 return 1000 * usecs;
1395 }
1396 }
1397
1398 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1399 if (os::supports_monotonic_clock()) {
1400 info_ptr->max_value = ALL_64_BITS;
1401
1402 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1403 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1404 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1405 } else {
1406 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1407 info_ptr->max_value = ALL_64_BITS;
1408
1409 // gettimeofday is a real time clock so it skips
1410 info_ptr->may_skip_backward = true;
1411 info_ptr->may_skip_forward = true;
1412 }
1413
1414 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1415 }
1416
1417 // Return the real, user, and system times in seconds from an
1418 // arbitrary fixed point in the past.
1419 bool os::getTimesSecs(double* process_real_time,
1420 double* process_user_time,
1421 double* process_system_time) {
1422 struct tms ticks;
1423 clock_t real_ticks = times(&ticks);
1424
1425 if (real_ticks == (clock_t) (-1)) {
1426 return false;
1427 } else {
1428 double ticks_per_second = (double) clock_tics_per_sec;
1429 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1430 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1431 *process_real_time = ((double) real_ticks) / ticks_per_second;
1432
1433 return true;
1434 }
1435 }
1436
1437
1438 char * os::local_time_string(char *buf, size_t buflen) {
1439 struct tm t;
1440 time_t long_time;
1441 time(&long_time);
1442 localtime_r(&long_time, &t);
1443 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1444 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1445 t.tm_hour, t.tm_min, t.tm_sec);
1446 return buf;
1447 }
1448
1449 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1450 return localtime_r(clock, res);
1451 }
1452
1453 ////////////////////////////////////////////////////////////////////////////////
1454 // runtime exit support
1455
1456 // Note: os::shutdown() might be called very early during initialization, or
1457 // called from signal handler. Before adding something to os::shutdown(), make
1458 // sure it is async-safe and can handle partially initialized VM.
1459 void os::shutdown() {
1460
1461 // allow PerfMemory to attempt cleanup of any persistent resources
1462 perfMemory_exit();
1463
1464 // needs to remove object in file system
1465 AttachListener::abort();
1466
1467 // flush buffered output, finish log files
1468 ostream_abort();
1469
1470 // Check for abort hook
1471 abort_hook_t abort_hook = Arguments::abort_hook();
1472 if (abort_hook != NULL) {
1473 abort_hook();
1474 }
1475
1476 }
1477
1478 // Note: os::abort() might be called very early during initialization, or
1479 // called from signal handler. Before adding something to os::abort(), make
1480 // sure it is async-safe and can handle partially initialized VM.
1481 void os::abort(bool dump_core, void* siginfo, void* context) {
1482 os::shutdown();
1483 if (dump_core) {
1484 #ifndef PRODUCT
1485 fdStream out(defaultStream::output_fd());
1486 out.print_raw("Current thread is ");
1487 char buf[16];
1488 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1489 out.print_raw_cr(buf);
1490 out.print_raw_cr("Dumping core ...");
1491 #endif
1492 ::abort(); // dump core
1493 }
1494
1495 ::exit(1);
1496 }
1497
1498 // Die immediately, no exit hook, no abort hook, no cleanup.
1499 void os::die() {
1500 // _exit() on LinuxThreads only kills current thread
1501 ::abort();
1502 }
1503
1504
1505 // This method is a copy of JDK's sysGetLastErrorString
1506 // from src/solaris/hpi/src/system_md.c
1507
1508 size_t os::lasterror(char *buf, size_t len) {
1509 if (errno == 0) return 0;
1510
1511 const char *s = ::strerror(errno);
1512 size_t n = ::strlen(s);
1513 if (n >= len) {
1514 n = len - 1;
1515 }
1516 ::strncpy(buf, s, n);
1517 buf[n] = '\0';
1518 return n;
1519 }
1520
1521 intx os::current_thread_id() { return (intx)pthread_self(); }
1522 int os::current_process_id() {
1523
1524 // Under the old linux thread library, linux gives each thread
1525 // its own process id. Because of this each thread will return
1526 // a different pid if this method were to return the result
1527 // of getpid(2). Linux provides no api that returns the pid
1528 // of the launcher thread for the vm. This implementation
1529 // returns a unique pid, the pid of the launcher thread
1530 // that starts the vm 'process'.
1531
1532 // Under the NPTL, getpid() returns the same pid as the
1533 // launcher thread rather than a unique pid per thread.
1534 // Use gettid() if you want the old pre NPTL behaviour.
1535
1536 // if you are looking for the result of a call to getpid() that
1537 // returns a unique pid for the calling thread, then look at the
1538 // OSThread::thread_id() method in osThread_linux.hpp file
1539
1540 return (int)(_initial_pid ? _initial_pid : getpid());
1541 }
1542
1543 // DLL functions
1544
1545 const char* os::dll_file_extension() { return ".so"; }
1546
1547 // This must be hard coded because it's the system's temporary
1548 // directory not the java application's temp directory, ala java.io.tmpdir.
1549 const char* os::get_temp_directory() { return "/tmp"; }
1550
1551 static bool file_exists(const char* filename) {
1552 struct stat statbuf;
1553 if (filename == NULL || strlen(filename) == 0) {
1554 return false;
1555 }
1556 return os::stat(filename, &statbuf) == 0;
1557 }
1558
1559 bool os::dll_build_name(char* buffer, size_t buflen,
1560 const char* pname, const char* fname) {
1561 bool retval = false;
1562 // Copied from libhpi
1563 const size_t pnamelen = pname ? strlen(pname) : 0;
1564
1565 // Return error on buffer overflow.
1566 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1567 return retval;
1568 }
1569
1570 if (pnamelen == 0) {
1571 snprintf(buffer, buflen, "lib%s.so", fname);
1572 retval = true;
1573 } else if (strchr(pname, *os::path_separator()) != NULL) {
1574 int n;
1575 char** pelements = split_path(pname, &n);
1576 if (pelements == NULL) {
1577 return false;
1578 }
1579 for (int i = 0; i < n; i++) {
1580 // Really shouldn't be NULL, but check can't hurt
1581 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1582 continue; // skip the empty path values
1583 }
1584 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1585 if (file_exists(buffer)) {
1586 retval = true;
1587 break;
1588 }
1589 }
1590 // release the storage
1591 for (int i = 0; i < n; i++) {
1592 if (pelements[i] != NULL) {
1593 FREE_C_HEAP_ARRAY(char, pelements[i]);
1594 }
1595 }
1596 if (pelements != NULL) {
1597 FREE_C_HEAP_ARRAY(char*, pelements);
1598 }
1599 } else {
1600 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1601 retval = true;
1602 }
1603 return retval;
1604 }
1605
1606 // check if addr is inside libjvm.so
1607 bool os::address_is_in_vm(address addr) {
1608 static address libjvm_base_addr;
1609 Dl_info dlinfo;
1610
1611 if (libjvm_base_addr == NULL) {
1612 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1613 libjvm_base_addr = (address)dlinfo.dli_fbase;
1614 }
1615 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1616 }
1617
1618 if (dladdr((void *)addr, &dlinfo) != 0) {
1619 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1620 }
1621
1622 return false;
1623 }
1624
1625 bool os::dll_address_to_function_name(address addr, char *buf,
1626 int buflen, int *offset) {
1627 // buf is not optional, but offset is optional
1628 assert(buf != NULL, "sanity check");
1629
1630 Dl_info dlinfo;
1631
1632 if (dladdr((void*)addr, &dlinfo) != 0) {
1633 // see if we have a matching symbol
1634 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1635 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1636 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1637 }
1638 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1639 return true;
1640 }
1641 // no matching symbol so try for just file info
1642 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1643 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1644 buf, buflen, offset, dlinfo.dli_fname)) {
1645 return true;
1646 }
1647 }
1648 }
1649
1650 buf[0] = '\0';
1651 if (offset != NULL) *offset = -1;
1652 return false;
1653 }
1654
1655 struct _address_to_library_name {
1656 address addr; // input : memory address
1657 size_t buflen; // size of fname
1658 char* fname; // output: library name
1659 address base; // library base addr
1660 };
1661
1662 static int address_to_library_name_callback(struct dl_phdr_info *info,
1663 size_t size, void *data) {
1664 int i;
1665 bool found = false;
1666 address libbase = NULL;
1667 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1668
1669 // iterate through all loadable segments
1670 for (i = 0; i < info->dlpi_phnum; i++) {
1671 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1672 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1673 // base address of a library is the lowest address of its loaded
1674 // segments.
1675 if (libbase == NULL || libbase > segbase) {
1676 libbase = segbase;
1677 }
1678 // see if 'addr' is within current segment
1679 if (segbase <= d->addr &&
1680 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1681 found = true;
1682 }
1683 }
1684 }
1685
1686 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1687 // so dll_address_to_library_name() can fall through to use dladdr() which
1688 // can figure out executable name from argv[0].
1689 if (found && info->dlpi_name && info->dlpi_name[0]) {
1690 d->base = libbase;
1691 if (d->fname) {
1692 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1693 }
1694 return 1;
1695 }
1696 return 0;
1697 }
1698
1699 bool os::dll_address_to_library_name(address addr, char* buf,
1700 int buflen, int* offset) {
1701 // buf is not optional, but offset is optional
1702 assert(buf != NULL, "sanity check");
1703
1704 Dl_info dlinfo;
1705 struct _address_to_library_name data;
1706
1707 // There is a bug in old glibc dladdr() implementation that it could resolve
1708 // to wrong library name if the .so file has a base address != NULL. Here
1709 // we iterate through the program headers of all loaded libraries to find
1710 // out which library 'addr' really belongs to. This workaround can be
1711 // removed once the minimum requirement for glibc is moved to 2.3.x.
1712 data.addr = addr;
1713 data.fname = buf;
1714 data.buflen = buflen;
1715 data.base = NULL;
1716 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1717
1718 if (rslt) {
1719 // buf already contains library name
1720 if (offset) *offset = addr - data.base;
1721 return true;
1722 }
1723 if (dladdr((void*)addr, &dlinfo) != 0) {
1724 if (dlinfo.dli_fname != NULL) {
1725 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1726 }
1727 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1728 *offset = addr - (address)dlinfo.dli_fbase;
1729 }
1730 return true;
1731 }
1732
1733 buf[0] = '\0';
1734 if (offset) *offset = -1;
1735 return false;
1736 }
1737
1738 // Loads .dll/.so and
1739 // in case of error it checks if .dll/.so was built for the
1740 // same architecture as Hotspot is running on
1741
1742
1743 // Remember the stack's state. The Linux dynamic linker will change
1744 // the stack to 'executable' at most once, so we must safepoint only once.
1745 bool os::Linux::_stack_is_executable = false;
1746
1747 // VM operation that loads a library. This is necessary if stack protection
1748 // of the Java stacks can be lost during loading the library. If we
1749 // do not stop the Java threads, they can stack overflow before the stacks
1750 // are protected again.
1751 class VM_LinuxDllLoad: public VM_Operation {
1752 private:
1753 const char *_filename;
1754 char *_ebuf;
1755 int _ebuflen;
1756 void *_lib;
1757 public:
1758 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1759 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1760 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1761 void doit() {
1762 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1763 os::Linux::_stack_is_executable = true;
1764 }
1765 void* loaded_library() { return _lib; }
1766 };
1767
1768 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1769 void * result = NULL;
1770 bool load_attempted = false;
1771
1772 // Check whether the library to load might change execution rights
1773 // of the stack. If they are changed, the protection of the stack
1774 // guard pages will be lost. We need a safepoint to fix this.
1775 //
1776 // See Linux man page execstack(8) for more info.
1777 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1778 ElfFile ef(filename);
1779 if (!ef.specifies_noexecstack()) {
1780 if (!is_init_completed()) {
1781 os::Linux::_stack_is_executable = true;
1782 // This is OK - No Java threads have been created yet, and hence no
1783 // stack guard pages to fix.
1784 //
1785 // This should happen only when you are building JDK7 using a very
1786 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1787 //
1788 // Dynamic loader will make all stacks executable after
1789 // this function returns, and will not do that again.
1790 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1791 } else {
1792 warning("You have loaded library %s which might have disabled stack guard. "
1793 "The VM will try to fix the stack guard now.\n"
1794 "It's highly recommended that you fix the library with "
1795 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1796 filename);
1797
1798 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1799 JavaThread *jt = JavaThread::current();
1800 if (jt->thread_state() != _thread_in_native) {
1801 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1802 // that requires ExecStack. Cannot enter safe point. Let's give up.
1803 warning("Unable to fix stack guard. Giving up.");
1804 } else {
1805 if (!LoadExecStackDllInVMThread) {
1806 // This is for the case where the DLL has an static
1807 // constructor function that executes JNI code. We cannot
1808 // load such DLLs in the VMThread.
1809 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1810 }
1811
1812 ThreadInVMfromNative tiv(jt);
1813 debug_only(VMNativeEntryWrapper vew;)
1814
1815 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1816 VMThread::execute(&op);
1817 if (LoadExecStackDllInVMThread) {
1818 result = op.loaded_library();
1819 }
1820 load_attempted = true;
1821 }
1822 }
1823 }
1824 }
1825
1826 if (!load_attempted) {
1827 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1828 }
1829
1830 if (result != NULL) {
1831 // Successful loading
1832 return result;
1833 }
1834
1835 Elf32_Ehdr elf_head;
1836 int diag_msg_max_length=ebuflen-strlen(ebuf);
1837 char* diag_msg_buf=ebuf+strlen(ebuf);
1838
1839 if (diag_msg_max_length==0) {
1840 // No more space in ebuf for additional diagnostics message
1841 return NULL;
1842 }
1843
1844
1845 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1846
1847 if (file_descriptor < 0) {
1848 // Can't open library, report dlerror() message
1849 return NULL;
1850 }
1851
1852 bool failed_to_read_elf_head=
1853 (sizeof(elf_head)!=
1854 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1855
1856 ::close(file_descriptor);
1857 if (failed_to_read_elf_head) {
1858 // file i/o error - report dlerror() msg
1859 return NULL;
1860 }
1861
1862 typedef struct {
1863 Elf32_Half code; // Actual value as defined in elf.h
1864 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1865 char elf_class; // 32 or 64 bit
1866 char endianess; // MSB or LSB
1867 char* name; // String representation
1868 } arch_t;
1869
1870 #ifndef EM_486
1871 #define EM_486 6 /* Intel 80486 */
1872 #endif
1873 #ifndef EM_AARCH64
1874 #define EM_AARCH64 183 /* ARM AARCH64 */
1875 #endif
1876
1877 static const arch_t arch_array[]={
1878 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1879 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1880 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1881 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1882 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1883 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1884 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1885 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1886 #if defined(VM_LITTLE_ENDIAN)
1887 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1888 #else
1889 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1890 #endif
1891 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1892 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1893 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1894 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1895 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1896 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1897 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1898 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1899 };
1900
1901 #if (defined IA32)
1902 static Elf32_Half running_arch_code=EM_386;
1903 #elif (defined AMD64)
1904 static Elf32_Half running_arch_code=EM_X86_64;
1905 #elif (defined IA64)
1906 static Elf32_Half running_arch_code=EM_IA_64;
1907 #elif (defined __sparc) && (defined _LP64)
1908 static Elf32_Half running_arch_code=EM_SPARCV9;
1909 #elif (defined __sparc) && (!defined _LP64)
1910 static Elf32_Half running_arch_code=EM_SPARC;
1911 #elif (defined __powerpc64__)
1912 static Elf32_Half running_arch_code=EM_PPC64;
1913 #elif (defined __powerpc__)
1914 static Elf32_Half running_arch_code=EM_PPC;
1915 #elif (defined ARM)
1916 static Elf32_Half running_arch_code=EM_ARM;
1917 #elif (defined S390)
1918 static Elf32_Half running_arch_code=EM_S390;
1919 #elif (defined ALPHA)
1920 static Elf32_Half running_arch_code=EM_ALPHA;
1921 #elif (defined MIPSEL)
1922 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1923 #elif (defined PARISC)
1924 static Elf32_Half running_arch_code=EM_PARISC;
1925 #elif (defined MIPS)
1926 static Elf32_Half running_arch_code=EM_MIPS;
1927 #elif (defined M68K)
1928 static Elf32_Half running_arch_code=EM_68K;
1929 #elif (defined AARCH64)
1930 static Elf32_Half running_arch_code=EM_AARCH64;
1931 #else
1932 #error Method os::dll_load requires that one of following is defined:\
1933 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1934 #endif
1935
1936 // Identify compatability class for VM's architecture and library's architecture
1937 // Obtain string descriptions for architectures
1938
1939 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1940 int running_arch_index=-1;
1941
1942 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1943 if (running_arch_code == arch_array[i].code) {
1944 running_arch_index = i;
1945 }
1946 if (lib_arch.code == arch_array[i].code) {
1947 lib_arch.compat_class = arch_array[i].compat_class;
1948 lib_arch.name = arch_array[i].name;
1949 }
1950 }
1951
1952 assert(running_arch_index != -1,
1953 "Didn't find running architecture code (running_arch_code) in arch_array");
1954 if (running_arch_index == -1) {
1955 // Even though running architecture detection failed
1956 // we may still continue with reporting dlerror() message
1957 return NULL;
1958 }
1959
1960 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1961 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1962 return NULL;
1963 }
1964
1965 #ifndef S390
1966 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1967 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1968 return NULL;
1969 }
1970 #endif // !S390
1971
1972 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1973 if (lib_arch.name!=NULL) {
1974 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1975 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1976 lib_arch.name, arch_array[running_arch_index].name);
1977 } else {
1978 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1979 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
1980 lib_arch.code,
1981 arch_array[running_arch_index].name);
1982 }
1983 }
1984
1985 return NULL;
1986 }
1987
1988 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
1989 int ebuflen) {
1990 void * result = ::dlopen(filename, RTLD_LAZY);
1991 if (result == NULL) {
1992 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
1993 ebuf[ebuflen-1] = '\0';
1994 }
1995 return result;
1996 }
1997
1998 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
1999 int ebuflen) {
2000 void * result = NULL;
2001 if (LoadExecStackDllInVMThread) {
2002 result = dlopen_helper(filename, ebuf, ebuflen);
2003 }
2004
2005 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2006 // library that requires an executable stack, or which does not have this
2007 // stack attribute set, dlopen changes the stack attribute to executable. The
2008 // read protection of the guard pages gets lost.
2009 //
2010 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2011 // may have been queued at the same time.
2012
2013 if (!_stack_is_executable) {
2014 JavaThread *jt = Threads::first();
2015
2016 while (jt) {
2017 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2018 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2019 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2020 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2021 warning("Attempt to reguard stack yellow zone failed.");
2022 }
2023 }
2024 jt = jt->next();
2025 }
2026 }
2027
2028 return result;
2029 }
2030
2031 void* os::dll_lookup(void* handle, const char* name) {
2032 void* res = dlsym(handle, name);
2033 return res;
2034 }
2035
2036 void* os::get_default_process_handle() {
2037 return (void*)::dlopen(NULL, RTLD_LAZY);
2038 }
2039
2040 static bool _print_ascii_file(const char* filename, outputStream* st) {
2041 int fd = ::open(filename, O_RDONLY);
2042 if (fd == -1) {
2043 return false;
2044 }
2045
2046 char buf[32];
2047 int bytes;
2048 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2049 st->print_raw(buf, bytes);
2050 }
2051
2052 ::close(fd);
2053
2054 return true;
2055 }
2056
2057 void os::print_dll_info(outputStream *st) {
2058 st->print_cr("Dynamic libraries:");
2059
2060 char fname[32];
2061 pid_t pid = os::Linux::gettid();
2062
2063 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2064
2065 if (!_print_ascii_file(fname, st)) {
2066 st->print("Can not get library information for pid = %d\n", pid);
2067 }
2068 }
2069
2070 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2071 FILE *procmapsFile = NULL;
2072
2073 // Open the procfs maps file for the current process
2074 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2075 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2076 char line[PATH_MAX + 100];
2077
2078 // Read line by line from 'file'
2079 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2080 u8 base, top, offset, inode;
2081 char permissions[5];
2082 char device[6];
2083 char name[PATH_MAX + 1];
2084
2085 // Parse fields from line
2086 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2087
2088 // Filter by device id '00:00' so that we only get file system mapped files.
2089 if (strcmp(device, "00:00") != 0) {
2090
2091 // Call callback with the fields of interest
2092 if(callback(name, (address)base, (address)top, param)) {
2093 // Oops abort, callback aborted
2094 fclose(procmapsFile);
2095 return 1;
2096 }
2097 }
2098 }
2099 fclose(procmapsFile);
2100 }
2101 return 0;
2102 }
2103
2104 void os::print_os_info_brief(outputStream* st) {
2105 os::Linux::print_distro_info(st);
2106
2107 os::Posix::print_uname_info(st);
2108
2109 os::Linux::print_libversion_info(st);
2110
2111 }
2112
2113 void os::print_os_info(outputStream* st) {
2114 st->print("OS:");
2115
2116 os::Linux::print_distro_info(st);
2117
2118 os::Posix::print_uname_info(st);
2119
2120 // Print warning if unsafe chroot environment detected
2121 if (unsafe_chroot_detected) {
2122 st->print("WARNING!! ");
2123 st->print_cr("%s", unstable_chroot_error);
2124 }
2125
2126 os::Linux::print_libversion_info(st);
2127
2128 os::Posix::print_rlimit_info(st);
2129
2130 os::Posix::print_load_average(st);
2131
2132 os::Linux::print_full_memory_info(st);
2133 }
2134
2135 // Try to identify popular distros.
2136 // Most Linux distributions have a /etc/XXX-release file, which contains
2137 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2138 // file that also contains the OS version string. Some have more than one
2139 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2140 // /etc/redhat-release.), so the order is important.
2141 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2142 // their own specific XXX-release file as well as a redhat-release file.
2143 // Because of this the XXX-release file needs to be searched for before the
2144 // redhat-release file.
2145 // Since Red Hat has a lsb-release file that is not very descriptive the
2146 // search for redhat-release needs to be before lsb-release.
2147 // Since the lsb-release file is the new standard it needs to be searched
2148 // before the older style release files.
2149 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2150 // next to last resort. The os-release file is a new standard that contains
2151 // distribution information and the system-release file seems to be an old
2152 // standard that has been replaced by the lsb-release and os-release files.
2153 // Searching for the debian_version file is the last resort. It contains
2154 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2155 // "Debian " is printed before the contents of the debian_version file.
2156 void os::Linux::print_distro_info(outputStream* st) {
2157 if (!_print_ascii_file("/etc/oracle-release", st) &&
2158 !_print_ascii_file("/etc/mandriva-release", st) &&
2159 !_print_ascii_file("/etc/mandrake-release", st) &&
2160 !_print_ascii_file("/etc/sun-release", st) &&
2161 !_print_ascii_file("/etc/redhat-release", st) &&
2162 !_print_ascii_file("/etc/lsb-release", st) &&
2163 !_print_ascii_file("/etc/SuSE-release", st) &&
2164 !_print_ascii_file("/etc/turbolinux-release", st) &&
2165 !_print_ascii_file("/etc/gentoo-release", st) &&
2166 !_print_ascii_file("/etc/ltib-release", st) &&
2167 !_print_ascii_file("/etc/angstrom-version", st) &&
2168 !_print_ascii_file("/etc/system-release", st) &&
2169 !_print_ascii_file("/etc/os-release", st)) {
2170
2171 if (file_exists("/etc/debian_version")) {
2172 st->print("Debian ");
2173 _print_ascii_file("/etc/debian_version", st);
2174 } else {
2175 st->print("Linux");
2176 }
2177 }
2178 st->cr();
2179 }
2180
2181 void os::Linux::print_libversion_info(outputStream* st) {
2182 // libc, pthread
2183 st->print("libc:");
2184 st->print("%s ", os::Linux::glibc_version());
2185 st->print("%s ", os::Linux::libpthread_version());
2186 if (os::Linux::is_LinuxThreads()) {
2187 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2188 }
2189 st->cr();
2190 }
2191
2192 void os::Linux::print_full_memory_info(outputStream* st) {
2193 st->print("\n/proc/meminfo:\n");
2194 _print_ascii_file("/proc/meminfo", st);
2195 st->cr();
2196 }
2197
2198 void os::print_memory_info(outputStream* st) {
2199
2200 st->print("Memory:");
2201 st->print(" %dk page", os::vm_page_size()>>10);
2202
2203 // values in struct sysinfo are "unsigned long"
2204 struct sysinfo si;
2205 sysinfo(&si);
2206
2207 st->print(", physical " UINT64_FORMAT "k",
2208 os::physical_memory() >> 10);
2209 st->print("(" UINT64_FORMAT "k free)",
2210 os::available_memory() >> 10);
2211 st->print(", swap " UINT64_FORMAT "k",
2212 ((jlong)si.totalswap * si.mem_unit) >> 10);
2213 st->print("(" UINT64_FORMAT "k free)",
2214 ((jlong)si.freeswap * si.mem_unit) >> 10);
2215 st->cr();
2216 }
2217
2218 void os::pd_print_cpu_info(outputStream* st) {
2219 #if defined(IA32) || defined(AMD64)
2220 // Only print the model name if the platform provides this as a summary
2221 st->print("\n/proc/cpuinfo:\n");
2222 FILE *fp = fopen("/proc/cpuinfo", "r");
2223 if (fp) {
2224 while (!feof(fp)) {
2225 char buf[256];
2226 if (fgets(buf, sizeof(buf), fp)) {
2227 if (strstr(buf, "model name") != NULL) {
2228 st->print_raw(buf);
2229 break;
2230 }
2231 }
2232 }
2233 } else {
2234 st->print_cr(" <Not Available>");
2235 }
2236 #else
2237 // Other platforms have reasonable cpuinfo files
2238 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2239 st->print_cr(" <Not Available>");
2240 }
2241 #endif
2242 }
2243
2244 void os::print_siginfo(outputStream* st, void* siginfo) {
2245 const siginfo_t* si = (const siginfo_t*)siginfo;
2246
2247 os::Posix::print_siginfo_brief(st, si);
2248 #if INCLUDE_CDS
2249 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2250 UseSharedSpaces) {
2251 FileMapInfo* mapinfo = FileMapInfo::current_info();
2252 if (mapinfo->is_in_shared_space(si->si_addr)) {
2253 st->print("\n\nError accessing class data sharing archive." \
2254 " Mapped file inaccessible during execution, " \
2255 " possible disk/network problem.");
2256 }
2257 }
2258 #endif
2259 st->cr();
2260 }
2261
2262
2263 static void print_signal_handler(outputStream* st, int sig,
2264 char* buf, size_t buflen);
2265
2266 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2267 st->print_cr("Signal Handlers:");
2268 print_signal_handler(st, SIGSEGV, buf, buflen);
2269 print_signal_handler(st, SIGBUS , buf, buflen);
2270 print_signal_handler(st, SIGFPE , buf, buflen);
2271 print_signal_handler(st, SIGPIPE, buf, buflen);
2272 print_signal_handler(st, SIGXFSZ, buf, buflen);
2273 print_signal_handler(st, SIGILL , buf, buflen);
2274 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2275 print_signal_handler(st, SR_signum, buf, buflen);
2276 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2277 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2278 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2279 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2280 #if defined(PPC64)
2281 print_signal_handler(st, SIGTRAP, buf, buflen);
2282 #endif
2283 }
2284
2285 static char saved_jvm_path[MAXPATHLEN] = {0};
2286
2287 // Find the full path to the current module, libjvm.so
2288 void os::jvm_path(char *buf, jint buflen) {
2289 // Error checking.
2290 if (buflen < MAXPATHLEN) {
2291 assert(false, "must use a large-enough buffer");
2292 buf[0] = '\0';
2293 return;
2294 }
2295 // Lazy resolve the path to current module.
2296 if (saved_jvm_path[0] != 0) {
2297 strcpy(buf, saved_jvm_path);
2298 return;
2299 }
2300
2301 char dli_fname[MAXPATHLEN];
2302 bool ret = dll_address_to_library_name(
2303 CAST_FROM_FN_PTR(address, os::jvm_path),
2304 dli_fname, sizeof(dli_fname), NULL);
2305 assert(ret, "cannot locate libjvm");
2306 char *rp = NULL;
2307 if (ret && dli_fname[0] != '\0') {
2308 rp = realpath(dli_fname, buf);
2309 }
2310 if (rp == NULL) {
2311 return;
2312 }
2313
2314 if (Arguments::sun_java_launcher_is_altjvm()) {
2315 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2316 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2317 // If "/jre/lib/" appears at the right place in the string, then
2318 // assume we are installed in a JDK and we're done. Otherwise, check
2319 // for a JAVA_HOME environment variable and fix up the path so it
2320 // looks like libjvm.so is installed there (append a fake suffix
2321 // hotspot/libjvm.so).
2322 const char *p = buf + strlen(buf) - 1;
2323 for (int count = 0; p > buf && count < 5; ++count) {
2324 for (--p; p > buf && *p != '/'; --p)
2325 /* empty */ ;
2326 }
2327
2328 if (strncmp(p, "/jre/lib/", 9) != 0) {
2329 // Look for JAVA_HOME in the environment.
2330 char* java_home_var = ::getenv("JAVA_HOME");
2331 if (java_home_var != NULL && java_home_var[0] != 0) {
2332 char* jrelib_p;
2333 int len;
2334
2335 // Check the current module name "libjvm.so".
2336 p = strrchr(buf, '/');
2337 if (p == NULL) {
2338 return;
2339 }
2340 assert(strstr(p, "/libjvm") == p, "invalid library name");
2341
2342 rp = realpath(java_home_var, buf);
2343 if (rp == NULL) {
2344 return;
2345 }
2346
2347 // determine if this is a legacy image or modules image
2348 // modules image doesn't have "jre" subdirectory
2349 len = strlen(buf);
2350 assert(len < buflen, "Ran out of buffer room");
2351 jrelib_p = buf + len;
2352 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2353 if (0 != access(buf, F_OK)) {
2354 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2355 }
2356
2357 if (0 == access(buf, F_OK)) {
2358 // Use current module name "libjvm.so"
2359 len = strlen(buf);
2360 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2361 } else {
2362 // Go back to path of .so
2363 rp = realpath(dli_fname, buf);
2364 if (rp == NULL) {
2365 return;
2366 }
2367 }
2368 }
2369 }
2370 }
2371
2372 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2373 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2374 }
2375
2376 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2377 // no prefix required, not even "_"
2378 }
2379
2380 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2381 // no suffix required
2382 }
2383
2384 ////////////////////////////////////////////////////////////////////////////////
2385 // sun.misc.Signal support
2386
2387 static volatile jint sigint_count = 0;
2388
2389 static void UserHandler(int sig, void *siginfo, void *context) {
2390 // 4511530 - sem_post is serialized and handled by the manager thread. When
2391 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2392 // don't want to flood the manager thread with sem_post requests.
2393 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2394 return;
2395 }
2396
2397 // Ctrl-C is pressed during error reporting, likely because the error
2398 // handler fails to abort. Let VM die immediately.
2399 if (sig == SIGINT && is_error_reported()) {
2400 os::die();
2401 }
2402
2403 os::signal_notify(sig);
2404 }
2405
2406 void* os::user_handler() {
2407 return CAST_FROM_FN_PTR(void*, UserHandler);
2408 }
2409
2410 class Semaphore : public StackObj {
2411 public:
2412 Semaphore();
2413 ~Semaphore();
2414 void signal();
2415 void wait();
2416 bool trywait();
2417 bool timedwait(unsigned int sec, int nsec);
2418 private:
2419 sem_t _semaphore;
2420 };
2421
2422 Semaphore::Semaphore() {
2423 sem_init(&_semaphore, 0, 0);
2424 }
2425
2426 Semaphore::~Semaphore() {
2427 sem_destroy(&_semaphore);
2428 }
2429
2430 void Semaphore::signal() {
2431 sem_post(&_semaphore);
2432 }
2433
2434 void Semaphore::wait() {
2435 sem_wait(&_semaphore);
2436 }
2437
2438 bool Semaphore::trywait() {
2439 return sem_trywait(&_semaphore) == 0;
2440 }
2441
2442 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2443
2444 struct timespec ts;
2445 // Semaphore's are always associated with CLOCK_REALTIME
2446 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2447 // see unpackTime for discussion on overflow checking
2448 if (sec >= MAX_SECS) {
2449 ts.tv_sec += MAX_SECS;
2450 ts.tv_nsec = 0;
2451 } else {
2452 ts.tv_sec += sec;
2453 ts.tv_nsec += nsec;
2454 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2455 ts.tv_nsec -= NANOSECS_PER_SEC;
2456 ++ts.tv_sec; // note: this must be <= max_secs
2457 }
2458 }
2459
2460 while (1) {
2461 int result = sem_timedwait(&_semaphore, &ts);
2462 if (result == 0) {
2463 return true;
2464 } else if (errno == EINTR) {
2465 continue;
2466 } else if (errno == ETIMEDOUT) {
2467 return false;
2468 } else {
2469 return false;
2470 }
2471 }
2472 }
2473
2474 extern "C" {
2475 typedef void (*sa_handler_t)(int);
2476 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2477 }
2478
2479 void* os::signal(int signal_number, void* handler) {
2480 struct sigaction sigAct, oldSigAct;
2481
2482 sigfillset(&(sigAct.sa_mask));
2483 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2484 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2485
2486 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2487 // -1 means registration failed
2488 return (void *)-1;
2489 }
2490
2491 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2492 }
2493
2494 void os::signal_raise(int signal_number) {
2495 ::raise(signal_number);
2496 }
2497
2498 // The following code is moved from os.cpp for making this
2499 // code platform specific, which it is by its very nature.
2500
2501 // Will be modified when max signal is changed to be dynamic
2502 int os::sigexitnum_pd() {
2503 return NSIG;
2504 }
2505
2506 // a counter for each possible signal value
2507 static volatile jint pending_signals[NSIG+1] = { 0 };
2508
2509 // Linux(POSIX) specific hand shaking semaphore.
2510 static sem_t sig_sem;
2511 static Semaphore sr_semaphore;
2512
2513 void os::signal_init_pd() {
2514 // Initialize signal structures
2515 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2516
2517 // Initialize signal semaphore
2518 ::sem_init(&sig_sem, 0, 0);
2519 }
2520
2521 void os::signal_notify(int sig) {
2522 Atomic::inc(&pending_signals[sig]);
2523 ::sem_post(&sig_sem);
2524 }
2525
2526 static int check_pending_signals(bool wait) {
2527 Atomic::store(0, &sigint_count);
2528 for (;;) {
2529 for (int i = 0; i < NSIG + 1; i++) {
2530 jint n = pending_signals[i];
2531 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2532 return i;
2533 }
2534 }
2535 if (!wait) {
2536 return -1;
2537 }
2538 JavaThread *thread = JavaThread::current();
2539 ThreadBlockInVM tbivm(thread);
2540
2541 bool threadIsSuspended;
2542 do {
2543 thread->set_suspend_equivalent();
2544 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2545 ::sem_wait(&sig_sem);
2546
2547 // were we externally suspended while we were waiting?
2548 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2549 if (threadIsSuspended) {
2550 // The semaphore has been incremented, but while we were waiting
2551 // another thread suspended us. We don't want to continue running
2552 // while suspended because that would surprise the thread that
2553 // suspended us.
2554 ::sem_post(&sig_sem);
2555
2556 thread->java_suspend_self();
2557 }
2558 } while (threadIsSuspended);
2559 }
2560 }
2561
2562 int os::signal_lookup() {
2563 return check_pending_signals(false);
2564 }
2565
2566 int os::signal_wait() {
2567 return check_pending_signals(true);
2568 }
2569
2570 ////////////////////////////////////////////////////////////////////////////////
2571 // Virtual Memory
2572
2573 int os::vm_page_size() {
2574 // Seems redundant as all get out
2575 assert(os::Linux::page_size() != -1, "must call os::init");
2576 return os::Linux::page_size();
2577 }
2578
2579 // Solaris allocates memory by pages.
2580 int os::vm_allocation_granularity() {
2581 assert(os::Linux::page_size() != -1, "must call os::init");
2582 return os::Linux::page_size();
2583 }
2584
2585 // Rationale behind this function:
2586 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2587 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2588 // samples for JITted code. Here we create private executable mapping over the code cache
2589 // and then we can use standard (well, almost, as mapping can change) way to provide
2590 // info for the reporting script by storing timestamp and location of symbol
2591 void linux_wrap_code(char* base, size_t size) {
2592 static volatile jint cnt = 0;
2593
2594 if (!UseOprofile) {
2595 return;
2596 }
2597
2598 char buf[PATH_MAX+1];
2599 int num = Atomic::add(1, &cnt);
2600
2601 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2602 os::get_temp_directory(), os::current_process_id(), num);
2603 unlink(buf);
2604
2605 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2606
2607 if (fd != -1) {
2608 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2609 if (rv != (off_t)-1) {
2610 if (::write(fd, "", 1) == 1) {
2611 mmap(base, size,
2612 PROT_READ|PROT_WRITE|PROT_EXEC,
2613 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2614 }
2615 }
2616 ::close(fd);
2617 unlink(buf);
2618 }
2619 }
2620
2621 static bool recoverable_mmap_error(int err) {
2622 // See if the error is one we can let the caller handle. This
2623 // list of errno values comes from JBS-6843484. I can't find a
2624 // Linux man page that documents this specific set of errno
2625 // values so while this list currently matches Solaris, it may
2626 // change as we gain experience with this failure mode.
2627 switch (err) {
2628 case EBADF:
2629 case EINVAL:
2630 case ENOTSUP:
2631 // let the caller deal with these errors
2632 return true;
2633
2634 default:
2635 // Any remaining errors on this OS can cause our reserved mapping
2636 // to be lost. That can cause confusion where different data
2637 // structures think they have the same memory mapped. The worst
2638 // scenario is if both the VM and a library think they have the
2639 // same memory mapped.
2640 return false;
2641 }
2642 }
2643
2644 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2645 int err) {
2646 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2647 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2648 strerror(err), err);
2649 }
2650
2651 static void warn_fail_commit_memory(char* addr, size_t size,
2652 size_t alignment_hint, bool exec,
2653 int err) {
2654 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2655 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2656 alignment_hint, exec, strerror(err), err);
2657 }
2658
2659 // NOTE: Linux kernel does not really reserve the pages for us.
2660 // All it does is to check if there are enough free pages
2661 // left at the time of mmap(). This could be a potential
2662 // problem.
2663 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2664 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2665 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2666 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2667 if (res != (uintptr_t) MAP_FAILED) {
2668 if (UseNUMAInterleaving) {
2669 numa_make_global(addr, size);
2670 }
2671 return 0;
2672 }
2673
2674 int err = errno; // save errno from mmap() call above
2675
2676 if (!recoverable_mmap_error(err)) {
2677 warn_fail_commit_memory(addr, size, exec, err);
2678 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2679 }
2680
2681 return err;
2682 }
2683
2684 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2685 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2686 }
2687
2688 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2689 const char* mesg) {
2690 assert(mesg != NULL, "mesg must be specified");
2691 int err = os::Linux::commit_memory_impl(addr, size, exec);
2692 if (err != 0) {
2693 // the caller wants all commit errors to exit with the specified mesg:
2694 warn_fail_commit_memory(addr, size, exec, err);
2695 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2696 }
2697 }
2698
2699 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2700 #ifndef MAP_HUGETLB
2701 #define MAP_HUGETLB 0x40000
2702 #endif
2703
2704 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2705 #ifndef MADV_HUGEPAGE
2706 #define MADV_HUGEPAGE 14
2707 #endif
2708
2709 int os::Linux::commit_memory_impl(char* addr, size_t size,
2710 size_t alignment_hint, bool exec) {
2711 int err = os::Linux::commit_memory_impl(addr, size, exec);
2712 if (err == 0) {
2713 realign_memory(addr, size, alignment_hint);
2714 }
2715 return err;
2716 }
2717
2718 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2719 bool exec) {
2720 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2721 }
2722
2723 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2724 size_t alignment_hint, bool exec,
2725 const char* mesg) {
2726 assert(mesg != NULL, "mesg must be specified");
2727 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2728 if (err != 0) {
2729 // the caller wants all commit errors to exit with the specified mesg:
2730 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2731 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2732 }
2733 }
2734
2735 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2736 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2737 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2738 // be supported or the memory may already be backed by huge pages.
2739 ::madvise(addr, bytes, MADV_HUGEPAGE);
2740 }
2741 }
2742
2743 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2744 // This method works by doing an mmap over an existing mmaping and effectively discarding
2745 // the existing pages. However it won't work for SHM-based large pages that cannot be
2746 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2747 // small pages on top of the SHM segment. This method always works for small pages, so we
2748 // allow that in any case.
2749 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2750 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2751 }
2752 }
2753
2754 void os::numa_make_global(char *addr, size_t bytes) {
2755 Linux::numa_interleave_memory(addr, bytes);
2756 }
2757
2758 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2759 // bind policy to MPOL_PREFERRED for the current thread.
2760 #define USE_MPOL_PREFERRED 0
2761
2762 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2763 // To make NUMA and large pages more robust when both enabled, we need to ease
2764 // the requirements on where the memory should be allocated. MPOL_BIND is the
2765 // default policy and it will force memory to be allocated on the specified
2766 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2767 // the specified node, but will not force it. Using this policy will prevent
2768 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2769 // free large pages.
2770 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2771 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2772 }
2773
2774 bool os::numa_topology_changed() { return false; }
2775
2776 size_t os::numa_get_groups_num() {
2777 int max_node = Linux::numa_max_node();
2778 return max_node > 0 ? max_node + 1 : 1;
2779 }
2780
2781 int os::numa_get_group_id() {
2782 int cpu_id = Linux::sched_getcpu();
2783 if (cpu_id != -1) {
2784 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2785 if (lgrp_id != -1) {
2786 return lgrp_id;
2787 }
2788 }
2789 return 0;
2790 }
2791
2792 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2793 for (size_t i = 0; i < size; i++) {
2794 ids[i] = i;
2795 }
2796 return size;
2797 }
2798
2799 bool os::get_page_info(char *start, page_info* info) {
2800 return false;
2801 }
2802
2803 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2804 page_info* page_found) {
2805 return end;
2806 }
2807
2808
2809 int os::Linux::sched_getcpu_syscall(void) {
2810 unsigned int cpu;
2811 int retval = -1;
2812
2813 #if defined(IA32)
2814 #ifndef SYS_getcpu
2815 #define SYS_getcpu 318
2816 #endif
2817 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2818 #elif defined(AMD64)
2819 // Unfortunately we have to bring all these macros here from vsyscall.h
2820 // to be able to compile on old linuxes.
2821 #define __NR_vgetcpu 2
2822 #define VSYSCALL_START (-10UL << 20)
2823 #define VSYSCALL_SIZE 1024
2824 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2825 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2826 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2827 retval = vgetcpu(&cpu, NULL, NULL);
2828 #endif
2829
2830 return (retval == -1) ? retval : cpu;
2831 }
2832
2833 // Something to do with the numa-aware allocator needs these symbols
2834 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2835 extern "C" JNIEXPORT void numa_error(char *where) { }
2836 extern "C" JNIEXPORT int fork1() { return fork(); }
2837
2838
2839 // If we are running with libnuma version > 2, then we should
2840 // be trying to use symbols with versions 1.1
2841 // If we are running with earlier version, which did not have symbol versions,
2842 // we should use the base version.
2843 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2844 void *f = dlvsym(handle, name, "libnuma_1.1");
2845 if (f == NULL) {
2846 f = dlsym(handle, name);
2847 }
2848 return f;
2849 }
2850
2851 bool os::Linux::libnuma_init() {
2852 // sched_getcpu() should be in libc.
2853 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2854 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2855
2856 // If it's not, try a direct syscall.
2857 if (sched_getcpu() == -1) {
2858 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2859 (void*)&sched_getcpu_syscall));
2860 }
2861
2862 if (sched_getcpu() != -1) { // Does it work?
2863 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2864 if (handle != NULL) {
2865 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2866 libnuma_dlsym(handle, "numa_node_to_cpus")));
2867 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2868 libnuma_dlsym(handle, "numa_max_node")));
2869 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2870 libnuma_dlsym(handle, "numa_available")));
2871 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2872 libnuma_dlsym(handle, "numa_tonode_memory")));
2873 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2874 libnuma_dlsym(handle, "numa_interleave_memory")));
2875 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2876 libnuma_dlsym(handle, "numa_set_bind_policy")));
2877
2878
2879 if (numa_available() != -1) {
2880 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2881 // Create a cpu -> node mapping
2882 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2883 rebuild_cpu_to_node_map();
2884 return true;
2885 }
2886 }
2887 }
2888 return false;
2889 }
2890
2891 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2892 // The table is later used in get_node_by_cpu().
2893 void os::Linux::rebuild_cpu_to_node_map() {
2894 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2895 // in libnuma (possible values are starting from 16,
2896 // and continuing up with every other power of 2, but less
2897 // than the maximum number of CPUs supported by kernel), and
2898 // is a subject to change (in libnuma version 2 the requirements
2899 // are more reasonable) we'll just hardcode the number they use
2900 // in the library.
2901 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2902
2903 size_t cpu_num = os::active_processor_count();
2904 size_t cpu_map_size = NCPUS / BitsPerCLong;
2905 size_t cpu_map_valid_size =
2906 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2907
2908 cpu_to_node()->clear();
2909 cpu_to_node()->at_grow(cpu_num - 1);
2910 size_t node_num = numa_get_groups_num();
2911
2912 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2913 for (size_t i = 0; i < node_num; i++) {
2914 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2915 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2916 if (cpu_map[j] != 0) {
2917 for (size_t k = 0; k < BitsPerCLong; k++) {
2918 if (cpu_map[j] & (1UL << k)) {
2919 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2920 }
2921 }
2922 }
2923 }
2924 }
2925 }
2926 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2927 }
2928
2929 int os::Linux::get_node_by_cpu(int cpu_id) {
2930 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2931 return cpu_to_node()->at(cpu_id);
2932 }
2933 return -1;
2934 }
2935
2936 GrowableArray<int>* os::Linux::_cpu_to_node;
2937 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2938 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2939 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2940 os::Linux::numa_available_func_t os::Linux::_numa_available;
2941 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2942 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2943 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2944 unsigned long* os::Linux::_numa_all_nodes;
2945
2946 bool os::pd_uncommit_memory(char* addr, size_t size) {
2947 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2948 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2949 return res != (uintptr_t) MAP_FAILED;
2950 }
2951
2952 static address get_stack_commited_bottom(address bottom, size_t size) {
2953 address nbot = bottom;
2954 address ntop = bottom + size;
2955
2956 size_t page_sz = os::vm_page_size();
2957 unsigned pages = size / page_sz;
2958
2959 unsigned char vec[1];
2960 unsigned imin = 1, imax = pages + 1, imid;
2961 int mincore_return_value = 0;
2962
2963 assert(imin <= imax, "Unexpected page size");
2964
2965 while (imin < imax) {
2966 imid = (imax + imin) / 2;
2967 nbot = ntop - (imid * page_sz);
2968
2969 // Use a trick with mincore to check whether the page is mapped or not.
2970 // mincore sets vec to 1 if page resides in memory and to 0 if page
2971 // is swapped output but if page we are asking for is unmapped
2972 // it returns -1,ENOMEM
2973 mincore_return_value = mincore(nbot, page_sz, vec);
2974
2975 if (mincore_return_value == -1) {
2976 // Page is not mapped go up
2977 // to find first mapped page
2978 if (errno != EAGAIN) {
2979 assert(errno == ENOMEM, "Unexpected mincore errno");
2980 imax = imid;
2981 }
2982 } else {
2983 // Page is mapped go down
2984 // to find first not mapped page
2985 imin = imid + 1;
2986 }
2987 }
2988
2989 nbot = nbot + page_sz;
2990
2991 // Adjust stack bottom one page up if last checked page is not mapped
2992 if (mincore_return_value == -1) {
2993 nbot = nbot + page_sz;
2994 }
2995
2996 return nbot;
2997 }
2998
2999
3000 // Linux uses a growable mapping for the stack, and if the mapping for
3001 // the stack guard pages is not removed when we detach a thread the
3002 // stack cannot grow beyond the pages where the stack guard was
3003 // mapped. If at some point later in the process the stack expands to
3004 // that point, the Linux kernel cannot expand the stack any further
3005 // because the guard pages are in the way, and a segfault occurs.
3006 //
3007 // However, it's essential not to split the stack region by unmapping
3008 // a region (leaving a hole) that's already part of the stack mapping,
3009 // so if the stack mapping has already grown beyond the guard pages at
3010 // the time we create them, we have to truncate the stack mapping.
3011 // So, we need to know the extent of the stack mapping when
3012 // create_stack_guard_pages() is called.
3013
3014 // We only need this for stacks that are growable: at the time of
3015 // writing thread stacks don't use growable mappings (i.e. those
3016 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3017 // only applies to the main thread.
3018
3019 // If the (growable) stack mapping already extends beyond the point
3020 // where we're going to put our guard pages, truncate the mapping at
3021 // that point by munmap()ping it. This ensures that when we later
3022 // munmap() the guard pages we don't leave a hole in the stack
3023 // mapping. This only affects the main/initial thread
3024
3025 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3026 if (os::Linux::is_initial_thread()) {
3027 // As we manually grow stack up to bottom inside create_attached_thread(),
3028 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3029 // we don't need to do anything special.
3030 // Check it first, before calling heavy function.
3031 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3032 unsigned char vec[1];
3033
3034 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3035 // Fallback to slow path on all errors, including EAGAIN
3036 stack_extent = (uintptr_t) get_stack_commited_bottom(
3037 os::Linux::initial_thread_stack_bottom(),
3038 (size_t)addr - stack_extent);
3039 }
3040
3041 if (stack_extent < (uintptr_t)addr) {
3042 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3043 }
3044 }
3045
3046 return os::commit_memory(addr, size, !ExecMem);
3047 }
3048
3049 // If this is a growable mapping, remove the guard pages entirely by
3050 // munmap()ping them. If not, just call uncommit_memory(). This only
3051 // affects the main/initial thread, but guard against future OS changes
3052 // It's safe to always unmap guard pages for initial thread because we
3053 // always place it right after end of the mapped region
3054
3055 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3056 uintptr_t stack_extent, stack_base;
3057
3058 if (os::Linux::is_initial_thread()) {
3059 return ::munmap(addr, size) == 0;
3060 }
3061
3062 return os::uncommit_memory(addr, size);
3063 }
3064
3065 static address _highest_vm_reserved_address = NULL;
3066
3067 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3068 // at 'requested_addr'. If there are existing memory mappings at the same
3069 // location, however, they will be overwritten. If 'fixed' is false,
3070 // 'requested_addr' is only treated as a hint, the return value may or
3071 // may not start from the requested address. Unlike Linux mmap(), this
3072 // function returns NULL to indicate failure.
3073 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3074 char * addr;
3075 int flags;
3076
3077 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3078 if (fixed) {
3079 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3080 flags |= MAP_FIXED;
3081 }
3082
3083 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3084 // touch an uncommitted page. Otherwise, the read/write might
3085 // succeed if we have enough swap space to back the physical page.
3086 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3087 flags, -1, 0);
3088
3089 if (addr != MAP_FAILED) {
3090 // anon_mmap() should only get called during VM initialization,
3091 // don't need lock (actually we can skip locking even it can be called
3092 // from multiple threads, because _highest_vm_reserved_address is just a
3093 // hint about the upper limit of non-stack memory regions.)
3094 if ((address)addr + bytes > _highest_vm_reserved_address) {
3095 _highest_vm_reserved_address = (address)addr + bytes;
3096 }
3097 }
3098
3099 return addr == MAP_FAILED ? NULL : addr;
3100 }
3101
3102 // Don't update _highest_vm_reserved_address, because there might be memory
3103 // regions above addr + size. If so, releasing a memory region only creates
3104 // a hole in the address space, it doesn't help prevent heap-stack collision.
3105 //
3106 static int anon_munmap(char * addr, size_t size) {
3107 return ::munmap(addr, size) == 0;
3108 }
3109
3110 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3111 size_t alignment_hint) {
3112 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3113 }
3114
3115 bool os::pd_release_memory(char* addr, size_t size) {
3116 return anon_munmap(addr, size);
3117 }
3118
3119 static address highest_vm_reserved_address() {
3120 return _highest_vm_reserved_address;
3121 }
3122
3123 static bool linux_mprotect(char* addr, size_t size, int prot) {
3124 // Linux wants the mprotect address argument to be page aligned.
3125 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3126
3127 // According to SUSv3, mprotect() should only be used with mappings
3128 // established by mmap(), and mmap() always maps whole pages. Unaligned
3129 // 'addr' likely indicates problem in the VM (e.g. trying to change
3130 // protection of malloc'ed or statically allocated memory). Check the
3131 // caller if you hit this assert.
3132 assert(addr == bottom, "sanity check");
3133
3134 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3135 return ::mprotect(bottom, size, prot) == 0;
3136 }
3137
3138 // Set protections specified
3139 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3140 bool is_committed) {
3141 unsigned int p = 0;
3142 switch (prot) {
3143 case MEM_PROT_NONE: p = PROT_NONE; break;
3144 case MEM_PROT_READ: p = PROT_READ; break;
3145 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3146 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3147 default:
3148 ShouldNotReachHere();
3149 }
3150 // is_committed is unused.
3151 return linux_mprotect(addr, bytes, p);
3152 }
3153
3154 bool os::guard_memory(char* addr, size_t size) {
3155 return linux_mprotect(addr, size, PROT_NONE);
3156 }
3157
3158 bool os::unguard_memory(char* addr, size_t size) {
3159 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3160 }
3161
3162 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3163 size_t page_size) {
3164 bool result = false;
3165 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3166 MAP_ANONYMOUS|MAP_PRIVATE,
3167 -1, 0);
3168 if (p != MAP_FAILED) {
3169 void *aligned_p = align_ptr_up(p, page_size);
3170
3171 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3172
3173 munmap(p, page_size * 2);
3174 }
3175
3176 if (warn && !result) {
3177 warning("TransparentHugePages is not supported by the operating system.");
3178 }
3179
3180 return result;
3181 }
3182
3183 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3184 bool result = false;
3185 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3186 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3187 -1, 0);
3188
3189 if (p != MAP_FAILED) {
3190 // We don't know if this really is a huge page or not.
3191 FILE *fp = fopen("/proc/self/maps", "r");
3192 if (fp) {
3193 while (!feof(fp)) {
3194 char chars[257];
3195 long x = 0;
3196 if (fgets(chars, sizeof(chars), fp)) {
3197 if (sscanf(chars, "%lx-%*x", &x) == 1
3198 && x == (long)p) {
3199 if (strstr (chars, "hugepage")) {
3200 result = true;
3201 break;
3202 }
3203 }
3204 }
3205 }
3206 fclose(fp);
3207 }
3208 munmap(p, page_size);
3209 }
3210
3211 if (warn && !result) {
3212 warning("HugeTLBFS is not supported by the operating system.");
3213 }
3214
3215 return result;
3216 }
3217
3218 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3219 //
3220 // From the coredump_filter documentation:
3221 //
3222 // - (bit 0) anonymous private memory
3223 // - (bit 1) anonymous shared memory
3224 // - (bit 2) file-backed private memory
3225 // - (bit 3) file-backed shared memory
3226 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3227 // effective only if the bit 2 is cleared)
3228 // - (bit 5) hugetlb private memory
3229 // - (bit 6) hugetlb shared memory
3230 //
3231 static void set_coredump_filter(void) {
3232 FILE *f;
3233 long cdm;
3234
3235 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3236 return;
3237 }
3238
3239 if (fscanf(f, "%lx", &cdm) != 1) {
3240 fclose(f);
3241 return;
3242 }
3243
3244 rewind(f);
3245
3246 if ((cdm & LARGEPAGES_BIT) == 0) {
3247 cdm |= LARGEPAGES_BIT;
3248 fprintf(f, "%#lx", cdm);
3249 }
3250
3251 fclose(f);
3252 }
3253
3254 // Large page support
3255
3256 static size_t _large_page_size = 0;
3257
3258 size_t os::Linux::find_large_page_size() {
3259 size_t large_page_size = 0;
3260
3261 // large_page_size on Linux is used to round up heap size. x86 uses either
3262 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3263 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3264 // page as large as 256M.
3265 //
3266 // Here we try to figure out page size by parsing /proc/meminfo and looking
3267 // for a line with the following format:
3268 // Hugepagesize: 2048 kB
3269 //
3270 // If we can't determine the value (e.g. /proc is not mounted, or the text
3271 // format has been changed), we'll use the largest page size supported by
3272 // the processor.
3273
3274 #ifndef ZERO
3275 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3276 ARM32_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3277 #endif // ZERO
3278
3279 FILE *fp = fopen("/proc/meminfo", "r");
3280 if (fp) {
3281 while (!feof(fp)) {
3282 int x = 0;
3283 char buf[16];
3284 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3285 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3286 large_page_size = x * K;
3287 break;
3288 }
3289 } else {
3290 // skip to next line
3291 for (;;) {
3292 int ch = fgetc(fp);
3293 if (ch == EOF || ch == (int)'\n') break;
3294 }
3295 }
3296 }
3297 fclose(fp);
3298 }
3299
3300 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3301 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3302 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3303 proper_unit_for_byte_size(large_page_size));
3304 }
3305
3306 return large_page_size;
3307 }
3308
3309 size_t os::Linux::setup_large_page_size() {
3310 _large_page_size = Linux::find_large_page_size();
3311 const size_t default_page_size = (size_t)Linux::page_size();
3312 if (_large_page_size > default_page_size) {
3313 _page_sizes[0] = _large_page_size;
3314 _page_sizes[1] = default_page_size;
3315 _page_sizes[2] = 0;
3316 }
3317
3318 return _large_page_size;
3319 }
3320
3321 bool os::Linux::setup_large_page_type(size_t page_size) {
3322 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3323 FLAG_IS_DEFAULT(UseSHM) &&
3324 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3325
3326 // The type of large pages has not been specified by the user.
3327
3328 // Try UseHugeTLBFS and then UseSHM.
3329 UseHugeTLBFS = UseSHM = true;
3330
3331 // Don't try UseTransparentHugePages since there are known
3332 // performance issues with it turned on. This might change in the future.
3333 UseTransparentHugePages = false;
3334 }
3335
3336 if (UseTransparentHugePages) {
3337 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3338 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3339 UseHugeTLBFS = false;
3340 UseSHM = false;
3341 return true;
3342 }
3343 UseTransparentHugePages = false;
3344 }
3345
3346 if (UseHugeTLBFS) {
3347 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3348 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3349 UseSHM = false;
3350 return true;
3351 }
3352 UseHugeTLBFS = false;
3353 }
3354
3355 return UseSHM;
3356 }
3357
3358 void os::large_page_init() {
3359 if (!UseLargePages &&
3360 !UseTransparentHugePages &&
3361 !UseHugeTLBFS &&
3362 !UseSHM) {
3363 // Not using large pages.
3364 return;
3365 }
3366
3367 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3368 // The user explicitly turned off large pages.
3369 // Ignore the rest of the large pages flags.
3370 UseTransparentHugePages = false;
3371 UseHugeTLBFS = false;
3372 UseSHM = false;
3373 return;
3374 }
3375
3376 size_t large_page_size = Linux::setup_large_page_size();
3377 UseLargePages = Linux::setup_large_page_type(large_page_size);
3378
3379 set_coredump_filter();
3380 }
3381
3382 #ifndef SHM_HUGETLB
3383 #define SHM_HUGETLB 04000
3384 #endif
3385
3386 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3387 char* req_addr, bool exec) {
3388 // "exec" is passed in but not used. Creating the shared image for
3389 // the code cache doesn't have an SHM_X executable permission to check.
3390 assert(UseLargePages && UseSHM, "only for SHM large pages");
3391 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3392
3393 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3394 return NULL; // Fallback to small pages.
3395 }
3396
3397 key_t key = IPC_PRIVATE;
3398 char *addr;
3399
3400 bool warn_on_failure = UseLargePages &&
3401 (!FLAG_IS_DEFAULT(UseLargePages) ||
3402 !FLAG_IS_DEFAULT(UseSHM) ||
3403 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3404 char msg[128];
3405
3406 // Create a large shared memory region to attach to based on size.
3407 // Currently, size is the total size of the heap
3408 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3409 if (shmid == -1) {
3410 // Possible reasons for shmget failure:
3411 // 1. shmmax is too small for Java heap.
3412 // > check shmmax value: cat /proc/sys/kernel/shmmax
3413 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3414 // 2. not enough large page memory.
3415 // > check available large pages: cat /proc/meminfo
3416 // > increase amount of large pages:
3417 // echo new_value > /proc/sys/vm/nr_hugepages
3418 // Note 1: different Linux may use different name for this property,
3419 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3420 // Note 2: it's possible there's enough physical memory available but
3421 // they are so fragmented after a long run that they can't
3422 // coalesce into large pages. Try to reserve large pages when
3423 // the system is still "fresh".
3424 if (warn_on_failure) {
3425 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3426 warning("%s", msg);
3427 }
3428 return NULL;
3429 }
3430
3431 // attach to the region
3432 addr = (char*)shmat(shmid, req_addr, 0);
3433 int err = errno;
3434
3435 // Remove shmid. If shmat() is successful, the actual shared memory segment
3436 // will be deleted when it's detached by shmdt() or when the process
3437 // terminates. If shmat() is not successful this will remove the shared
3438 // segment immediately.
3439 shmctl(shmid, IPC_RMID, NULL);
3440
3441 if ((intptr_t)addr == -1) {
3442 if (warn_on_failure) {
3443 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3444 warning("%s", msg);
3445 }
3446 return NULL;
3447 }
3448
3449 return addr;
3450 }
3451
3452 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3453 int error) {
3454 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3455
3456 bool warn_on_failure = UseLargePages &&
3457 (!FLAG_IS_DEFAULT(UseLargePages) ||
3458 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3459 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3460
3461 if (warn_on_failure) {
3462 char msg[128];
3463 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3464 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3465 warning("%s", msg);
3466 }
3467 }
3468
3469 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3470 char* req_addr,
3471 bool exec) {
3472 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3473 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3474 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3475
3476 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3477 char* addr = (char*)::mmap(req_addr, bytes, prot,
3478 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3479 -1, 0);
3480
3481 if (addr == MAP_FAILED) {
3482 warn_on_large_pages_failure(req_addr, bytes, errno);
3483 return NULL;
3484 }
3485
3486 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3487
3488 return addr;
3489 }
3490
3491 // Helper for os::Linux::reserve_memory_special_huge_tlbfs_mixed().
3492 // Allocate (using mmap, NO_RESERVE, with small pages) at either a given request address
3493 // (req_addr != NULL) or with a given alignment.
3494 // - bytes shall be a multiple of alignment.
3495 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3496 // - alignment sets the alignment at which memory shall be allocated.
3497 // It must be a multiple of allocation granularity.
3498 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3499 // req_addr or NULL.
3500 static char* anon_mmap_aligned(size_t bytes, size_t alignment, char* req_addr) {
3501
3502 size_t extra_size = bytes;
3503 if (req_addr == NULL && alignment > 0) {
3504 extra_size += alignment;
3505 }
3506
3507 char* start = (char*) ::mmap(req_addr, extra_size, PROT_NONE,
3508 MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
3509 -1, 0);
3510 if (start == MAP_FAILED) {
3511 start = NULL;
3512 } else {
3513 if (req_addr != NULL) {
3514 if (start != req_addr) {
3515 ::munmap(start, extra_size);
3516 start = NULL;
3517 }
3518 } else {
3519 char* const start_aligned = (char*) align_ptr_up(start, alignment);
3520 char* const end_aligned = start_aligned + bytes;
3521 char* const end = start + extra_size;
3522 if (start_aligned > start) {
3523 ::munmap(start, start_aligned - start);
3524 }
3525 if (end_aligned < end) {
3526 ::munmap(end_aligned, end - end_aligned);
3527 }
3528 start = start_aligned;
3529 }
3530 }
3531 return start;
3532
3533 }
3534
3535 // Reserve memory using mmap(MAP_HUGETLB).
3536 // - bytes shall be a multiple of alignment.
3537 // - req_addr can be NULL. If not NULL, it must be a multiple of alignment.
3538 // - alignment sets the alignment at which memory shall be allocated.
3539 // It must be a multiple of allocation granularity.
3540 // Returns address of memory or NULL. If req_addr was not NULL, will only return
3541 // req_addr or NULL.
3542 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3543 size_t alignment,
3544 char* req_addr,
3545 bool exec) {
3546 size_t large_page_size = os::large_page_size();
3547 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3548
3549 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3550 assert(is_size_aligned(bytes, alignment), "Must be");
3551
3552 // First reserve - but not commit - the address range in small pages.
3553 char* const start = anon_mmap_aligned(bytes, alignment, req_addr);
3554
3555 if (start == NULL) {
3556 return NULL;
3557 }
3558
3559 assert(is_ptr_aligned(start, alignment), "Must be");
3560
3561 char* end = start + bytes;
3562
3563 // Find the regions of the allocated chunk that can be promoted to large pages.
3564 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3565 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3566
3567 size_t lp_bytes = lp_end - lp_start;
3568
3569 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3570
3571 if (lp_bytes == 0) {
3572 // The mapped region doesn't even span the start and the end of a large page.
3573 // Fall back to allocate a non-special area.
3574 ::munmap(start, end - start);
3575 return NULL;
3576 }
3577
3578 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3579
3580 void* result;
3581
3582 // Commit small-paged leading area.
3583 if (start != lp_start) {
3584 result = ::mmap(start, lp_start - start, prot,
3585 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3586 -1, 0);
3587 if (result == MAP_FAILED) {
3588 ::munmap(lp_start, end - lp_start);
3589 return NULL;
3590 }
3591 }
3592
3593 // Commit large-paged area.
3594 result = ::mmap(lp_start, lp_bytes, prot,
3595 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3596 -1, 0);
3597 if (result == MAP_FAILED) {
3598 warn_on_large_pages_failure(lp_start, lp_bytes, errno);
3599 // If the mmap above fails, the large pages region will be unmapped and we
3600 // have regions before and after with small pages. Release these regions.
3601 //
3602 // | mapped | unmapped | mapped |
3603 // ^ ^ ^ ^
3604 // start lp_start lp_end end
3605 //
3606 ::munmap(start, lp_start - start);
3607 ::munmap(lp_end, end - lp_end);
3608 return NULL;
3609 }
3610
3611 // Commit small-paged trailing area.
3612 if (lp_end != end) {
3613 result = ::mmap(lp_end, end - lp_end, prot,
3614 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3615 -1, 0);
3616 if (result == MAP_FAILED) {
3617 ::munmap(start, lp_end - start);
3618 return NULL;
3619 }
3620 }
3621
3622 return start;
3623 }
3624
3625 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3626 size_t alignment,
3627 char* req_addr,
3628 bool exec) {
3629 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3630 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3631 assert(is_size_aligned(alignment, os::vm_allocation_granularity()), "Must be");
3632 assert(is_power_of_2(os::large_page_size()), "Must be");
3633 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3634
3635 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3636 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3637 } else {
3638 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3639 }
3640 }
3641
3642 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3643 char* req_addr, bool exec) {
3644 assert(UseLargePages, "only for large pages");
3645
3646 char* addr;
3647 if (UseSHM) {
3648 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3649 } else {
3650 assert(UseHugeTLBFS, "must be");
3651 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3652 }
3653
3654 if (addr != NULL) {
3655 if (UseNUMAInterleaving) {
3656 numa_make_global(addr, bytes);
3657 }
3658
3659 // The memory is committed
3660 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3661 }
3662
3663 return addr;
3664 }
3665
3666 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3667 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3668 return shmdt(base) == 0;
3669 }
3670
3671 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3672 return pd_release_memory(base, bytes);
3673 }
3674
3675 bool os::release_memory_special(char* base, size_t bytes) {
3676 bool res;
3677 if (MemTracker::tracking_level() > NMT_minimal) {
3678 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3679 res = os::Linux::release_memory_special_impl(base, bytes);
3680 if (res) {
3681 tkr.record((address)base, bytes);
3682 }
3683
3684 } else {
3685 res = os::Linux::release_memory_special_impl(base, bytes);
3686 }
3687 return res;
3688 }
3689
3690 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3691 assert(UseLargePages, "only for large pages");
3692 bool res;
3693
3694 if (UseSHM) {
3695 res = os::Linux::release_memory_special_shm(base, bytes);
3696 } else {
3697 assert(UseHugeTLBFS, "must be");
3698 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3699 }
3700 return res;
3701 }
3702
3703 size_t os::large_page_size() {
3704 return _large_page_size;
3705 }
3706
3707 // With SysV SHM the entire memory region must be allocated as shared
3708 // memory.
3709 // HugeTLBFS allows application to commit large page memory on demand.
3710 // However, when committing memory with HugeTLBFS fails, the region
3711 // that was supposed to be committed will lose the old reservation
3712 // and allow other threads to steal that memory region. Because of this
3713 // behavior we can't commit HugeTLBFS memory.
3714 bool os::can_commit_large_page_memory() {
3715 return UseTransparentHugePages;
3716 }
3717
3718 bool os::can_execute_large_page_memory() {
3719 return UseTransparentHugePages || UseHugeTLBFS;
3720 }
3721
3722 // Reserve memory at an arbitrary address, only if that area is
3723 // available (and not reserved for something else).
3724
3725 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3726 const int max_tries = 10;
3727 char* base[max_tries];
3728 size_t size[max_tries];
3729 const size_t gap = 0x000000;
3730
3731 // Assert only that the size is a multiple of the page size, since
3732 // that's all that mmap requires, and since that's all we really know
3733 // about at this low abstraction level. If we need higher alignment,
3734 // we can either pass an alignment to this method or verify alignment
3735 // in one of the methods further up the call chain. See bug 5044738.
3736 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3737
3738 // Repeatedly allocate blocks until the block is allocated at the
3739 // right spot. Give up after max_tries. Note that reserve_memory() will
3740 // automatically update _highest_vm_reserved_address if the call is
3741 // successful. The variable tracks the highest memory address every reserved
3742 // by JVM. It is used to detect heap-stack collision if running with
3743 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3744 // space than needed, it could confuse the collision detecting code. To
3745 // solve the problem, save current _highest_vm_reserved_address and
3746 // calculate the correct value before return.
3747 address old_highest = _highest_vm_reserved_address;
3748
3749 // Linux mmap allows caller to pass an address as hint; give it a try first,
3750 // if kernel honors the hint then we can return immediately.
3751 char * addr = anon_mmap(requested_addr, bytes, false);
3752 if (addr == requested_addr) {
3753 return requested_addr;
3754 }
3755
3756 if (addr != NULL) {
3757 // mmap() is successful but it fails to reserve at the requested address
3758 anon_munmap(addr, bytes);
3759 }
3760
3761 int i;
3762 for (i = 0; i < max_tries; ++i) {
3763 base[i] = reserve_memory(bytes);
3764
3765 if (base[i] != NULL) {
3766 // Is this the block we wanted?
3767 if (base[i] == requested_addr) {
3768 size[i] = bytes;
3769 break;
3770 }
3771
3772 // Does this overlap the block we wanted? Give back the overlapped
3773 // parts and try again.
3774
3775 ptrdiff_t top_overlap = requested_addr + (bytes + gap) - base[i];
3776 if (top_overlap >= 0 && (size_t)top_overlap < bytes) {
3777 unmap_memory(base[i], top_overlap);
3778 base[i] += top_overlap;
3779 size[i] = bytes - top_overlap;
3780 } else {
3781 ptrdiff_t bottom_overlap = base[i] + bytes - requested_addr;
3782 if (bottom_overlap >= 0 && (size_t)bottom_overlap < bytes) {
3783 unmap_memory(requested_addr, bottom_overlap);
3784 size[i] = bytes - bottom_overlap;
3785 } else {
3786 size[i] = bytes;
3787 }
3788 }
3789 }
3790 }
3791
3792 // Give back the unused reserved pieces.
3793
3794 for (int j = 0; j < i; ++j) {
3795 if (base[j] != NULL) {
3796 unmap_memory(base[j], size[j]);
3797 }
3798 }
3799
3800 if (i < max_tries) {
3801 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3802 return requested_addr;
3803 } else {
3804 _highest_vm_reserved_address = old_highest;
3805 return NULL;
3806 }
3807 }
3808
3809 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3810 return ::read(fd, buf, nBytes);
3811 }
3812
3813 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3814 return ::pread(fd, buf, nBytes, offset);
3815 }
3816
3817 // Short sleep, direct OS call.
3818 //
3819 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3820 // sched_yield(2) will actually give up the CPU:
3821 //
3822 // * Alone on this pariticular CPU, keeps running.
3823 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3824 // (pre 2.6.39).
3825 //
3826 // So calling this with 0 is an alternative.
3827 //
3828 void os::naked_short_sleep(jlong ms) {
3829 struct timespec req;
3830
3831 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3832 req.tv_sec = 0;
3833 if (ms > 0) {
3834 req.tv_nsec = (ms % 1000) * 1000000;
3835 } else {
3836 req.tv_nsec = 1;
3837 }
3838
3839 nanosleep(&req, NULL);
3840
3841 return;
3842 }
3843
3844 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3845 void os::infinite_sleep() {
3846 while (true) { // sleep forever ...
3847 ::sleep(100); // ... 100 seconds at a time
3848 }
3849 }
3850
3851 // Used to convert frequent JVM_Yield() to nops
3852 bool os::dont_yield() {
3853 return DontYieldALot;
3854 }
3855
3856 void os::naked_yield() {
3857 sched_yield();
3858 }
3859
3860 ////////////////////////////////////////////////////////////////////////////////
3861 // thread priority support
3862
3863 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3864 // only supports dynamic priority, static priority must be zero. For real-time
3865 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3866 // However, for large multi-threaded applications, SCHED_RR is not only slower
3867 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3868 // of 5 runs - Sep 2005).
3869 //
3870 // The following code actually changes the niceness of kernel-thread/LWP. It
3871 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3872 // not the entire user process, and user level threads are 1:1 mapped to kernel
3873 // threads. It has always been the case, but could change in the future. For
3874 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3875 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3876
3877 int os::java_to_os_priority[CriticalPriority + 1] = {
3878 19, // 0 Entry should never be used
3879
3880 4, // 1 MinPriority
3881 3, // 2
3882 2, // 3
3883
3884 1, // 4
3885 0, // 5 NormPriority
3886 -1, // 6
3887
3888 -2, // 7
3889 -3, // 8
3890 -4, // 9 NearMaxPriority
3891
3892 -5, // 10 MaxPriority
3893
3894 -5 // 11 CriticalPriority
3895 };
3896
3897 static int prio_init() {
3898 if (ThreadPriorityPolicy == 1) {
3899 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3900 // if effective uid is not root. Perhaps, a more elegant way of doing
3901 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3902 if (geteuid() != 0) {
3903 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3904 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3905 }
3906 ThreadPriorityPolicy = 0;
3907 }
3908 }
3909 if (UseCriticalJavaThreadPriority) {
3910 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3911 }
3912 return 0;
3913 }
3914
3915 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3916 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3917
3918 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3919 return (ret == 0) ? OS_OK : OS_ERR;
3920 }
3921
3922 OSReturn os::get_native_priority(const Thread* const thread,
3923 int *priority_ptr) {
3924 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3925 *priority_ptr = java_to_os_priority[NormPriority];
3926 return OS_OK;
3927 }
3928
3929 errno = 0;
3930 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3931 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3932 }
3933
3934 // Hint to the underlying OS that a task switch would not be good.
3935 // Void return because it's a hint and can fail.
3936 void os::hint_no_preempt() {}
3937
3938 ////////////////////////////////////////////////////////////////////////////////
3939 // suspend/resume support
3940
3941 // the low-level signal-based suspend/resume support is a remnant from the
3942 // old VM-suspension that used to be for java-suspension, safepoints etc,
3943 // within hotspot. Now there is a single use-case for this:
3944 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3945 // that runs in the watcher thread.
3946 // The remaining code is greatly simplified from the more general suspension
3947 // code that used to be used.
3948 //
3949 // The protocol is quite simple:
3950 // - suspend:
3951 // - sends a signal to the target thread
3952 // - polls the suspend state of the osthread using a yield loop
3953 // - target thread signal handler (SR_handler) sets suspend state
3954 // and blocks in sigsuspend until continued
3955 // - resume:
3956 // - sets target osthread state to continue
3957 // - sends signal to end the sigsuspend loop in the SR_handler
3958 //
3959 // Note that the SR_lock plays no role in this suspend/resume protocol.
3960
3961 static void resume_clear_context(OSThread *osthread) {
3962 osthread->set_ucontext(NULL);
3963 osthread->set_siginfo(NULL);
3964 }
3965
3966 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3967 ucontext_t* context) {
3968 osthread->set_ucontext(context);
3969 osthread->set_siginfo(siginfo);
3970 }
3971
3972 // Handler function invoked when a thread's execution is suspended or
3973 // resumed. We have to be careful that only async-safe functions are
3974 // called here (Note: most pthread functions are not async safe and
3975 // should be avoided.)
3976 //
3977 // Note: sigwait() is a more natural fit than sigsuspend() from an
3978 // interface point of view, but sigwait() prevents the signal hander
3979 // from being run. libpthread would get very confused by not having
3980 // its signal handlers run and prevents sigwait()'s use with the
3981 // mutex granting granting signal.
3982 //
3983 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3984 //
3985 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3986 // Save and restore errno to avoid confusing native code with EINTR
3987 // after sigsuspend.
3988 int old_errno = errno;
3989
3990 Thread* thread = Thread::current();
3991 OSThread* osthread = thread->osthread();
3992 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3993
3994 os::SuspendResume::State current = osthread->sr.state();
3995 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3996 suspend_save_context(osthread, siginfo, context);
3997
3998 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3999 os::SuspendResume::State state = osthread->sr.suspended();
4000 if (state == os::SuspendResume::SR_SUSPENDED) {
4001 sigset_t suspend_set; // signals for sigsuspend()
4002
4003 // get current set of blocked signals and unblock resume signal
4004 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
4005 sigdelset(&suspend_set, SR_signum);
4006
4007 sr_semaphore.signal();
4008 // wait here until we are resumed
4009 while (1) {
4010 sigsuspend(&suspend_set);
4011
4012 os::SuspendResume::State result = osthread->sr.running();
4013 if (result == os::SuspendResume::SR_RUNNING) {
4014 sr_semaphore.signal();
4015 break;
4016 }
4017 }
4018
4019 } else if (state == os::SuspendResume::SR_RUNNING) {
4020 // request was cancelled, continue
4021 } else {
4022 ShouldNotReachHere();
4023 }
4024
4025 resume_clear_context(osthread);
4026 } else if (current == os::SuspendResume::SR_RUNNING) {
4027 // request was cancelled, continue
4028 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
4029 // ignore
4030 } else {
4031 // ignore
4032 }
4033
4034 errno = old_errno;
4035 }
4036
4037
4038 static int SR_initialize() {
4039 struct sigaction act;
4040 char *s;
4041 // Get signal number to use for suspend/resume
4042 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4043 int sig = ::strtol(s, 0, 10);
4044 if (sig > 0 || sig < _NSIG) {
4045 SR_signum = sig;
4046 }
4047 }
4048
4049 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4050 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4051
4052 sigemptyset(&SR_sigset);
4053 sigaddset(&SR_sigset, SR_signum);
4054
4055 // Set up signal handler for suspend/resume
4056 act.sa_flags = SA_RESTART|SA_SIGINFO;
4057 act.sa_handler = (void (*)(int)) SR_handler;
4058
4059 // SR_signum is blocked by default.
4060 // 4528190 - We also need to block pthread restart signal (32 on all
4061 // supported Linux platforms). Note that LinuxThreads need to block
4062 // this signal for all threads to work properly. So we don't have
4063 // to use hard-coded signal number when setting up the mask.
4064 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4065
4066 if (sigaction(SR_signum, &act, 0) == -1) {
4067 return -1;
4068 }
4069
4070 // Save signal flag
4071 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4072 return 0;
4073 }
4074
4075 static int sr_notify(OSThread* osthread) {
4076 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4077 assert_status(status == 0, status, "pthread_kill");
4078 return status;
4079 }
4080
4081 // "Randomly" selected value for how long we want to spin
4082 // before bailing out on suspending a thread, also how often
4083 // we send a signal to a thread we want to resume
4084 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4085 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4086
4087 // returns true on success and false on error - really an error is fatal
4088 // but this seems the normal response to library errors
4089 static bool do_suspend(OSThread* osthread) {
4090 assert(osthread->sr.is_running(), "thread should be running");
4091 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4092
4093 // mark as suspended and send signal
4094 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4095 // failed to switch, state wasn't running?
4096 ShouldNotReachHere();
4097 return false;
4098 }
4099
4100 if (sr_notify(osthread) != 0) {
4101 ShouldNotReachHere();
4102 }
4103
4104 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4105 while (true) {
4106 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4107 break;
4108 } else {
4109 // timeout
4110 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4111 if (cancelled == os::SuspendResume::SR_RUNNING) {
4112 return false;
4113 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4114 // make sure that we consume the signal on the semaphore as well
4115 sr_semaphore.wait();
4116 break;
4117 } else {
4118 ShouldNotReachHere();
4119 return false;
4120 }
4121 }
4122 }
4123
4124 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4125 return true;
4126 }
4127
4128 static void do_resume(OSThread* osthread) {
4129 assert(osthread->sr.is_suspended(), "thread should be suspended");
4130 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4131
4132 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4133 // failed to switch to WAKEUP_REQUEST
4134 ShouldNotReachHere();
4135 return;
4136 }
4137
4138 while (true) {
4139 if (sr_notify(osthread) == 0) {
4140 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4141 if (osthread->sr.is_running()) {
4142 return;
4143 }
4144 }
4145 } else {
4146 ShouldNotReachHere();
4147 }
4148 }
4149
4150 guarantee(osthread->sr.is_running(), "Must be running!");
4151 }
4152
4153 ///////////////////////////////////////////////////////////////////////////////////
4154 // signal handling (except suspend/resume)
4155
4156 // This routine may be used by user applications as a "hook" to catch signals.
4157 // The user-defined signal handler must pass unrecognized signals to this
4158 // routine, and if it returns true (non-zero), then the signal handler must
4159 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4160 // routine will never retun false (zero), but instead will execute a VM panic
4161 // routine kill the process.
4162 //
4163 // If this routine returns false, it is OK to call it again. This allows
4164 // the user-defined signal handler to perform checks either before or after
4165 // the VM performs its own checks. Naturally, the user code would be making
4166 // a serious error if it tried to handle an exception (such as a null check
4167 // or breakpoint) that the VM was generating for its own correct operation.
4168 //
4169 // This routine may recognize any of the following kinds of signals:
4170 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4171 // It should be consulted by handlers for any of those signals.
4172 //
4173 // The caller of this routine must pass in the three arguments supplied
4174 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4175 // field of the structure passed to sigaction(). This routine assumes that
4176 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4177 //
4178 // Note that the VM will print warnings if it detects conflicting signal
4179 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4180 //
4181 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4182 siginfo_t* siginfo,
4183 void* ucontext,
4184 int abort_if_unrecognized);
4185
4186 void signalHandler(int sig, siginfo_t* info, void* uc) {
4187 assert(info != NULL && uc != NULL, "it must be old kernel");
4188 int orig_errno = errno; // Preserve errno value over signal handler.
4189 JVM_handle_linux_signal(sig, info, uc, true);
4190 errno = orig_errno;
4191 }
4192
4193
4194 // This boolean allows users to forward their own non-matching signals
4195 // to JVM_handle_linux_signal, harmlessly.
4196 bool os::Linux::signal_handlers_are_installed = false;
4197
4198 // For signal-chaining
4199 struct sigaction os::Linux::sigact[MAXSIGNUM];
4200 unsigned int os::Linux::sigs = 0;
4201 bool os::Linux::libjsig_is_loaded = false;
4202 typedef struct sigaction *(*get_signal_t)(int);
4203 get_signal_t os::Linux::get_signal_action = NULL;
4204
4205 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4206 struct sigaction *actp = NULL;
4207
4208 if (libjsig_is_loaded) {
4209 // Retrieve the old signal handler from libjsig
4210 actp = (*get_signal_action)(sig);
4211 }
4212 if (actp == NULL) {
4213 // Retrieve the preinstalled signal handler from jvm
4214 actp = get_preinstalled_handler(sig);
4215 }
4216
4217 return actp;
4218 }
4219
4220 static bool call_chained_handler(struct sigaction *actp, int sig,
4221 siginfo_t *siginfo, void *context) {
4222 // Call the old signal handler
4223 if (actp->sa_handler == SIG_DFL) {
4224 // It's more reasonable to let jvm treat it as an unexpected exception
4225 // instead of taking the default action.
4226 return false;
4227 } else if (actp->sa_handler != SIG_IGN) {
4228 if ((actp->sa_flags & SA_NODEFER) == 0) {
4229 // automaticlly block the signal
4230 sigaddset(&(actp->sa_mask), sig);
4231 }
4232
4233 sa_handler_t hand;
4234 sa_sigaction_t sa;
4235 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4236 // retrieve the chained handler
4237 if (siginfo_flag_set) {
4238 sa = actp->sa_sigaction;
4239 } else {
4240 hand = actp->sa_handler;
4241 }
4242
4243 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4244 actp->sa_handler = SIG_DFL;
4245 }
4246
4247 // try to honor the signal mask
4248 sigset_t oset;
4249 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4250
4251 // call into the chained handler
4252 if (siginfo_flag_set) {
4253 (*sa)(sig, siginfo, context);
4254 } else {
4255 (*hand)(sig);
4256 }
4257
4258 // restore the signal mask
4259 pthread_sigmask(SIG_SETMASK, &oset, 0);
4260 }
4261 // Tell jvm's signal handler the signal is taken care of.
4262 return true;
4263 }
4264
4265 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4266 bool chained = false;
4267 // signal-chaining
4268 if (UseSignalChaining) {
4269 struct sigaction *actp = get_chained_signal_action(sig);
4270 if (actp != NULL) {
4271 chained = call_chained_handler(actp, sig, siginfo, context);
4272 }
4273 }
4274 return chained;
4275 }
4276
4277 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4278 if ((((unsigned int)1 << sig) & sigs) != 0) {
4279 return &sigact[sig];
4280 }
4281 return NULL;
4282 }
4283
4284 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4285 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4286 sigact[sig] = oldAct;
4287 sigs |= (unsigned int)1 << sig;
4288 }
4289
4290 // for diagnostic
4291 int os::Linux::sigflags[MAXSIGNUM];
4292
4293 int os::Linux::get_our_sigflags(int sig) {
4294 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4295 return sigflags[sig];
4296 }
4297
4298 void os::Linux::set_our_sigflags(int sig, int flags) {
4299 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4300 sigflags[sig] = flags;
4301 }
4302
4303 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4304 // Check for overwrite.
4305 struct sigaction oldAct;
4306 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4307
4308 void* oldhand = oldAct.sa_sigaction
4309 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4310 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4311 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4312 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4313 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4314 if (AllowUserSignalHandlers || !set_installed) {
4315 // Do not overwrite; user takes responsibility to forward to us.
4316 return;
4317 } else if (UseSignalChaining) {
4318 // save the old handler in jvm
4319 save_preinstalled_handler(sig, oldAct);
4320 // libjsig also interposes the sigaction() call below and saves the
4321 // old sigaction on it own.
4322 } else {
4323 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4324 "%#lx for signal %d.", (long)oldhand, sig));
4325 }
4326 }
4327
4328 struct sigaction sigAct;
4329 sigfillset(&(sigAct.sa_mask));
4330 sigAct.sa_handler = SIG_DFL;
4331 if (!set_installed) {
4332 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4333 } else {
4334 sigAct.sa_sigaction = signalHandler;
4335 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4336 }
4337 // Save flags, which are set by ours
4338 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4339 sigflags[sig] = sigAct.sa_flags;
4340
4341 int ret = sigaction(sig, &sigAct, &oldAct);
4342 assert(ret == 0, "check");
4343
4344 void* oldhand2 = oldAct.sa_sigaction
4345 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4346 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4347 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4348 }
4349
4350 // install signal handlers for signals that HotSpot needs to
4351 // handle in order to support Java-level exception handling.
4352
4353 void os::Linux::install_signal_handlers() {
4354 if (!signal_handlers_are_installed) {
4355 signal_handlers_are_installed = true;
4356
4357 // signal-chaining
4358 typedef void (*signal_setting_t)();
4359 signal_setting_t begin_signal_setting = NULL;
4360 signal_setting_t end_signal_setting = NULL;
4361 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4362 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4363 if (begin_signal_setting != NULL) {
4364 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4365 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4366 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4367 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4368 libjsig_is_loaded = true;
4369 assert(UseSignalChaining, "should enable signal-chaining");
4370 }
4371 if (libjsig_is_loaded) {
4372 // Tell libjsig jvm is setting signal handlers
4373 (*begin_signal_setting)();
4374 }
4375
4376 set_signal_handler(SIGSEGV, true);
4377 set_signal_handler(SIGPIPE, true);
4378 set_signal_handler(SIGBUS, true);
4379 set_signal_handler(SIGILL, true);
4380 set_signal_handler(SIGFPE, true);
4381 #if defined(PPC64)
4382 set_signal_handler(SIGTRAP, true);
4383 #endif
4384 set_signal_handler(SIGXFSZ, true);
4385
4386 if (libjsig_is_loaded) {
4387 // Tell libjsig jvm finishes setting signal handlers
4388 (*end_signal_setting)();
4389 }
4390
4391 // We don't activate signal checker if libjsig is in place, we trust ourselves
4392 // and if UserSignalHandler is installed all bets are off.
4393 // Log that signal checking is off only if -verbose:jni is specified.
4394 if (CheckJNICalls) {
4395 if (libjsig_is_loaded) {
4396 if (PrintJNIResolving) {
4397 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4398 }
4399 check_signals = false;
4400 }
4401 if (AllowUserSignalHandlers) {
4402 if (PrintJNIResolving) {
4403 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4404 }
4405 check_signals = false;
4406 }
4407 }
4408 }
4409 }
4410
4411 // This is the fastest way to get thread cpu time on Linux.
4412 // Returns cpu time (user+sys) for any thread, not only for current.
4413 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4414 // It might work on 2.6.10+ with a special kernel/glibc patch.
4415 // For reference, please, see IEEE Std 1003.1-2004:
4416 // http://www.unix.org/single_unix_specification
4417
4418 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4419 struct timespec tp;
4420 int rc = os::Linux::clock_gettime(clockid, &tp);
4421 assert(rc == 0, "clock_gettime is expected to return 0 code");
4422
4423 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4424 }
4425
4426 /////
4427 // glibc on Linux platform uses non-documented flag
4428 // to indicate, that some special sort of signal
4429 // trampoline is used.
4430 // We will never set this flag, and we should
4431 // ignore this flag in our diagnostic
4432 #ifdef SIGNIFICANT_SIGNAL_MASK
4433 #undef SIGNIFICANT_SIGNAL_MASK
4434 #endif
4435 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4436
4437 static const char* get_signal_handler_name(address handler,
4438 char* buf, int buflen) {
4439 int offset;
4440 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4441 if (found) {
4442 // skip directory names
4443 const char *p1, *p2;
4444 p1 = buf;
4445 size_t len = strlen(os::file_separator());
4446 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4447 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4448 } else {
4449 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4450 }
4451 return buf;
4452 }
4453
4454 static void print_signal_handler(outputStream* st, int sig,
4455 char* buf, size_t buflen) {
4456 struct sigaction sa;
4457
4458 sigaction(sig, NULL, &sa);
4459
4460 // See comment for SIGNIFICANT_SIGNAL_MASK define
4461 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4462
4463 st->print("%s: ", os::exception_name(sig, buf, buflen));
4464
4465 address handler = (sa.sa_flags & SA_SIGINFO)
4466 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4467 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4468
4469 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4470 st->print("SIG_DFL");
4471 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4472 st->print("SIG_IGN");
4473 } else {
4474 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4475 }
4476
4477 st->print(", sa_mask[0]=");
4478 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4479
4480 address rh = VMError::get_resetted_sighandler(sig);
4481 // May be, handler was resetted by VMError?
4482 if (rh != NULL) {
4483 handler = rh;
4484 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4485 }
4486
4487 st->print(", sa_flags=");
4488 os::Posix::print_sa_flags(st, sa.sa_flags);
4489
4490 // Check: is it our handler?
4491 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4492 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4493 // It is our signal handler
4494 // check for flags, reset system-used one!
4495 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4496 st->print(
4497 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4498 os::Linux::get_our_sigflags(sig));
4499 }
4500 }
4501 st->cr();
4502 }
4503
4504
4505 #define DO_SIGNAL_CHECK(sig) \
4506 do { \
4507 if (!sigismember(&check_signal_done, sig)) { \
4508 os::Linux::check_signal_handler(sig); \
4509 } \
4510 } while (0)
4511
4512 // This method is a periodic task to check for misbehaving JNI applications
4513 // under CheckJNI, we can add any periodic checks here
4514
4515 void os::run_periodic_checks() {
4516 if (check_signals == false) return;
4517
4518 // SEGV and BUS if overridden could potentially prevent
4519 // generation of hs*.log in the event of a crash, debugging
4520 // such a case can be very challenging, so we absolutely
4521 // check the following for a good measure:
4522 DO_SIGNAL_CHECK(SIGSEGV);
4523 DO_SIGNAL_CHECK(SIGILL);
4524 DO_SIGNAL_CHECK(SIGFPE);
4525 DO_SIGNAL_CHECK(SIGBUS);
4526 DO_SIGNAL_CHECK(SIGPIPE);
4527 DO_SIGNAL_CHECK(SIGXFSZ);
4528 #if defined(PPC64)
4529 DO_SIGNAL_CHECK(SIGTRAP);
4530 #endif
4531
4532 // ReduceSignalUsage allows the user to override these handlers
4533 // see comments at the very top and jvm_solaris.h
4534 if (!ReduceSignalUsage) {
4535 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4536 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4537 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4538 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4539 }
4540
4541 DO_SIGNAL_CHECK(SR_signum);
4542 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4543 }
4544
4545 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4546
4547 static os_sigaction_t os_sigaction = NULL;
4548
4549 void os::Linux::check_signal_handler(int sig) {
4550 char buf[O_BUFLEN];
4551 address jvmHandler = NULL;
4552
4553
4554 struct sigaction act;
4555 if (os_sigaction == NULL) {
4556 // only trust the default sigaction, in case it has been interposed
4557 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4558 if (os_sigaction == NULL) return;
4559 }
4560
4561 os_sigaction(sig, (struct sigaction*)NULL, &act);
4562
4563
4564 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4565
4566 address thisHandler = (act.sa_flags & SA_SIGINFO)
4567 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4568 : CAST_FROM_FN_PTR(address, act.sa_handler);
4569
4570
4571 switch (sig) {
4572 case SIGSEGV:
4573 case SIGBUS:
4574 case SIGFPE:
4575 case SIGPIPE:
4576 case SIGILL:
4577 case SIGXFSZ:
4578 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4579 break;
4580
4581 case SHUTDOWN1_SIGNAL:
4582 case SHUTDOWN2_SIGNAL:
4583 case SHUTDOWN3_SIGNAL:
4584 case BREAK_SIGNAL:
4585 jvmHandler = (address)user_handler();
4586 break;
4587
4588 case INTERRUPT_SIGNAL:
4589 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4590 break;
4591
4592 default:
4593 if (sig == SR_signum) {
4594 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4595 } else {
4596 return;
4597 }
4598 break;
4599 }
4600
4601 if (thisHandler != jvmHandler) {
4602 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4603 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4604 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4605 // No need to check this sig any longer
4606 sigaddset(&check_signal_done, sig);
4607 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4608 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4609 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4610 exception_name(sig, buf, O_BUFLEN));
4611 }
4612 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4613 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4614 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4615 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4616 // No need to check this sig any longer
4617 sigaddset(&check_signal_done, sig);
4618 }
4619
4620 // Dump all the signal
4621 if (sigismember(&check_signal_done, sig)) {
4622 print_signal_handlers(tty, buf, O_BUFLEN);
4623 }
4624 }
4625
4626 extern void report_error(char* file_name, int line_no, char* title,
4627 char* format, ...);
4628
4629 extern bool signal_name(int signo, char* buf, size_t len);
4630
4631 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4632 if (0 < exception_code && exception_code <= SIGRTMAX) {
4633 // signal
4634 if (!signal_name(exception_code, buf, size)) {
4635 jio_snprintf(buf, size, "SIG%d", exception_code);
4636 }
4637 return buf;
4638 } else {
4639 return NULL;
4640 }
4641 }
4642
4643 // this is called _before_ the most of global arguments have been parsed
4644 void os::init(void) {
4645 char dummy; // used to get a guess on initial stack address
4646 // first_hrtime = gethrtime();
4647
4648 // With LinuxThreads the JavaMain thread pid (primordial thread)
4649 // is different than the pid of the java launcher thread.
4650 // So, on Linux, the launcher thread pid is passed to the VM
4651 // via the sun.java.launcher.pid property.
4652 // Use this property instead of getpid() if it was correctly passed.
4653 // See bug 6351349.
4654 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4655
4656 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4657
4658 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4659
4660 init_random(1234567);
4661
4662 ThreadCritical::initialize();
4663
4664 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4665 if (Linux::page_size() == -1) {
4666 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4667 strerror(errno)));
4668 }
4669 init_page_sizes((size_t) Linux::page_size());
4670
4671 Linux::initialize_system_info();
4672
4673 // main_thread points to the aboriginal thread
4674 Linux::_main_thread = pthread_self();
4675
4676 Linux::clock_init();
4677 initial_time_count = javaTimeNanos();
4678
4679 // pthread_condattr initialization for monotonic clock
4680 int status;
4681 pthread_condattr_t* _condattr = os::Linux::condAttr();
4682 if ((status = pthread_condattr_init(_condattr)) != 0) {
4683 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4684 }
4685 // Only set the clock if CLOCK_MONOTONIC is available
4686 if (os::supports_monotonic_clock()) {
4687 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4688 if (status == EINVAL) {
4689 warning("Unable to use monotonic clock with relative timed-waits" \
4690 " - changes to the time-of-day clock may have adverse affects");
4691 } else {
4692 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4693 }
4694 }
4695 }
4696 // else it defaults to CLOCK_REALTIME
4697
4698 // If the pagesize of the VM is greater than 8K determine the appropriate
4699 // number of initial guard pages. The user can change this with the
4700 // command line arguments, if needed.
4701 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4702 StackYellowPages = 1;
4703 StackRedPages = 1;
4704 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4705 }
4706
4707 // retrieve entry point for pthread_setname_np
4708 Linux::_pthread_setname_np =
4709 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4710
4711 }
4712
4713 // To install functions for atexit system call
4714 extern "C" {
4715 static void perfMemory_exit_helper() {
4716 perfMemory_exit();
4717 }
4718 }
4719
4720 // this is called _after_ the global arguments have been parsed
4721 jint os::init_2(void) {
4722 Linux::fast_thread_clock_init();
4723
4724 // Allocate a single page and mark it as readable for safepoint polling
4725 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4726 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4727
4728 os::set_polling_page(polling_page);
4729
4730 #ifndef PRODUCT
4731 if (Verbose && PrintMiscellaneous) {
4732 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4733 (intptr_t)polling_page);
4734 }
4735 #endif
4736
4737 if (!UseMembar) {
4738 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4739 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4740 os::set_memory_serialize_page(mem_serialize_page);
4741
4742 #ifndef PRODUCT
4743 if (Verbose && PrintMiscellaneous) {
4744 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4745 (intptr_t)mem_serialize_page);
4746 }
4747 #endif
4748 }
4749
4750 // initialize suspend/resume support - must do this before signal_sets_init()
4751 if (SR_initialize() != 0) {
4752 perror("SR_initialize failed");
4753 return JNI_ERR;
4754 }
4755
4756 Linux::signal_sets_init();
4757 Linux::install_signal_handlers();
4758
4759 // Check minimum allowable stack size for thread creation and to initialize
4760 // the java system classes, including StackOverflowError - depends on page
4761 // size. Add a page for compiler2 recursion in main thread.
4762 // Add in 2*BytesPerWord times page size to account for VM stack during
4763 // class initialization depending on 32 or 64 bit VM.
4764 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4765 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4766 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4767
4768 size_t threadStackSizeInBytes = ThreadStackSize * K;
4769 if (threadStackSizeInBytes != 0 &&
4770 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4771 tty->print_cr("\nThe stack size specified is too small, "
4772 "Specify at least %dk",
4773 os::Linux::min_stack_allowed/ K);
4774 return JNI_ERR;
4775 }
4776
4777 // Make the stack size a multiple of the page size so that
4778 // the yellow/red zones can be guarded.
4779 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4780 vm_page_size()));
4781
4782 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4783
4784 #if defined(IA32)
4785 workaround_expand_exec_shield_cs_limit();
4786 #endif
4787
4788 Linux::libpthread_init();
4789 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4790 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4791 Linux::glibc_version(), Linux::libpthread_version(),
4792 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4793 }
4794
4795 if (UseNUMA) {
4796 if (!Linux::libnuma_init()) {
4797 UseNUMA = false;
4798 } else {
4799 if ((Linux::numa_max_node() < 1)) {
4800 // There's only one node(they start from 0), disable NUMA.
4801 UseNUMA = false;
4802 }
4803 }
4804 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4805 // we can make the adaptive lgrp chunk resizing work. If the user specified
4806 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4807 // disable adaptive resizing.
4808 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4809 if (FLAG_IS_DEFAULT(UseNUMA)) {
4810 UseNUMA = false;
4811 } else {
4812 if (FLAG_IS_DEFAULT(UseLargePages) &&
4813 FLAG_IS_DEFAULT(UseSHM) &&
4814 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4815 UseLargePages = false;
4816 } else if (UseAdaptiveSizePolicy || UseAdaptiveNUMAChunkSizing) {
4817 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing (-XX:-UseAdaptiveSizePolicy -XX:-UseAdaptiveNUMAChunkSizing)");
4818 UseAdaptiveSizePolicy = false;
4819 UseAdaptiveNUMAChunkSizing = false;
4820 }
4821 }
4822 }
4823 if (!UseNUMA && ForceNUMA) {
4824 UseNUMA = true;
4825 }
4826 }
4827
4828 if (MaxFDLimit) {
4829 // set the number of file descriptors to max. print out error
4830 // if getrlimit/setrlimit fails but continue regardless.
4831 struct rlimit nbr_files;
4832 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4833 if (status != 0) {
4834 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4835 perror("os::init_2 getrlimit failed");
4836 }
4837 } else {
4838 nbr_files.rlim_cur = nbr_files.rlim_max;
4839 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4840 if (status != 0) {
4841 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4842 perror("os::init_2 setrlimit failed");
4843 }
4844 }
4845 }
4846 }
4847
4848 // Initialize lock used to serialize thread creation (see os::create_thread)
4849 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4850
4851 // at-exit methods are called in the reverse order of their registration.
4852 // atexit functions are called on return from main or as a result of a
4853 // call to exit(3C). There can be only 32 of these functions registered
4854 // and atexit() does not set errno.
4855
4856 if (PerfAllowAtExitRegistration) {
4857 // only register atexit functions if PerfAllowAtExitRegistration is set.
4858 // atexit functions can be delayed until process exit time, which
4859 // can be problematic for embedded VM situations. Embedded VMs should
4860 // call DestroyJavaVM() to assure that VM resources are released.
4861
4862 // note: perfMemory_exit_helper atexit function may be removed in
4863 // the future if the appropriate cleanup code can be added to the
4864 // VM_Exit VMOperation's doit method.
4865 if (atexit(perfMemory_exit_helper) != 0) {
4866 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4867 }
4868 }
4869
4870 // initialize thread priority policy
4871 prio_init();
4872
4873 return JNI_OK;
4874 }
4875
4876 // Mark the polling page as unreadable
4877 void os::make_polling_page_unreadable(void) {
4878 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4879 fatal("Could not disable polling page");
4880 }
4881 }
4882
4883 // Mark the polling page as readable
4884 void os::make_polling_page_readable(void) {
4885 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4886 fatal("Could not enable polling page");
4887 }
4888 }
4889
4890 int os::active_processor_count() {
4891 // Linux doesn't yet have a (official) notion of processor sets,
4892 // so just return the number of online processors.
4893 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4894 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4895 return online_cpus;
4896 }
4897
4898 void os::set_native_thread_name(const char *name) {
4899 if (Linux::_pthread_setname_np) {
4900 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4901 snprintf(buf, sizeof(buf), "%s", name);
4902 buf[sizeof(buf) - 1] = '\0';
4903 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4904 // ERANGE should not happen; all other errors should just be ignored.
4905 assert(rc != ERANGE, "pthread_setname_np failed");
4906 }
4907 }
4908
4909 bool os::distribute_processes(uint length, uint* distribution) {
4910 // Not yet implemented.
4911 return false;
4912 }
4913
4914 bool os::bind_to_processor(uint processor_id) {
4915 // Not yet implemented.
4916 return false;
4917 }
4918
4919 ///
4920
4921 void os::SuspendedThreadTask::internal_do_task() {
4922 if (do_suspend(_thread->osthread())) {
4923 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4924 do_task(context);
4925 do_resume(_thread->osthread());
4926 }
4927 }
4928
4929 class PcFetcher : public os::SuspendedThreadTask {
4930 public:
4931 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4932 ExtendedPC result();
4933 protected:
4934 void do_task(const os::SuspendedThreadTaskContext& context);
4935 private:
4936 ExtendedPC _epc;
4937 };
4938
4939 ExtendedPC PcFetcher::result() {
4940 guarantee(is_done(), "task is not done yet.");
4941 return _epc;
4942 }
4943
4944 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4945 Thread* thread = context.thread();
4946 OSThread* osthread = thread->osthread();
4947 if (osthread->ucontext() != NULL) {
4948 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4949 } else {
4950 // NULL context is unexpected, double-check this is the VMThread
4951 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4952 }
4953 }
4954
4955 // Suspends the target using the signal mechanism and then grabs the PC before
4956 // resuming the target. Used by the flat-profiler only
4957 ExtendedPC os::get_thread_pc(Thread* thread) {
4958 // Make sure that it is called by the watcher for the VMThread
4959 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4960 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4961
4962 PcFetcher fetcher(thread);
4963 fetcher.run();
4964 return fetcher.result();
4965 }
4966
4967 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4968 pthread_mutex_t *_mutex,
4969 const struct timespec *_abstime) {
4970 if (is_NPTL()) {
4971 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4972 } else {
4973 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4974 // word back to default 64bit precision if condvar is signaled. Java
4975 // wants 53bit precision. Save and restore current value.
4976 int fpu = get_fpu_control_word();
4977 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4978 set_fpu_control_word(fpu);
4979 return status;
4980 }
4981 }
4982
4983 ////////////////////////////////////////////////////////////////////////////////
4984 // debug support
4985
4986 bool os::find(address addr, outputStream* st) {
4987 Dl_info dlinfo;
4988 memset(&dlinfo, 0, sizeof(dlinfo));
4989 if (dladdr(addr, &dlinfo) != 0) {
4990 st->print(PTR_FORMAT ": ", addr);
4991 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4992 st->print("%s+%#x", dlinfo.dli_sname,
4993 addr - (intptr_t)dlinfo.dli_saddr);
4994 } else if (dlinfo.dli_fbase != NULL) {
4995 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4996 } else {
4997 st->print("<absolute address>");
4998 }
4999 if (dlinfo.dli_fname != NULL) {
5000 st->print(" in %s", dlinfo.dli_fname);
5001 }
5002 if (dlinfo.dli_fbase != NULL) {
5003 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
5004 }
5005 st->cr();
5006
5007 if (Verbose) {
5008 // decode some bytes around the PC
5009 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
5010 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
5011 address lowest = (address) dlinfo.dli_sname;
5012 if (!lowest) lowest = (address) dlinfo.dli_fbase;
5013 if (begin < lowest) begin = lowest;
5014 Dl_info dlinfo2;
5015 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
5016 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
5017 end = (address) dlinfo2.dli_saddr;
5018 }
5019 Disassembler::decode(begin, end, st);
5020 }
5021 return true;
5022 }
5023 return false;
5024 }
5025
5026 ////////////////////////////////////////////////////////////////////////////////
5027 // misc
5028
5029 // This does not do anything on Linux. This is basically a hook for being
5030 // able to use structured exception handling (thread-local exception filters)
5031 // on, e.g., Win32.
5032 void
5033 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5034 JavaCallArguments* args, Thread* thread) {
5035 f(value, method, args, thread);
5036 }
5037
5038 void os::print_statistics() {
5039 }
5040
5041 int os::message_box(const char* title, const char* message) {
5042 int i;
5043 fdStream err(defaultStream::error_fd());
5044 for (i = 0; i < 78; i++) err.print_raw("=");
5045 err.cr();
5046 err.print_raw_cr(title);
5047 for (i = 0; i < 78; i++) err.print_raw("-");
5048 err.cr();
5049 err.print_raw_cr(message);
5050 for (i = 0; i < 78; i++) err.print_raw("=");
5051 err.cr();
5052
5053 char buf[16];
5054 // Prevent process from exiting upon "read error" without consuming all CPU
5055 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5056
5057 return buf[0] == 'y' || buf[0] == 'Y';
5058 }
5059
5060 int os::stat(const char *path, struct stat *sbuf) {
5061 char pathbuf[MAX_PATH];
5062 if (strlen(path) > MAX_PATH - 1) {
5063 errno = ENAMETOOLONG;
5064 return -1;
5065 }
5066 os::native_path(strcpy(pathbuf, path));
5067 return ::stat(pathbuf, sbuf);
5068 }
5069
5070 bool os::check_heap(bool force) {
5071 return true;
5072 }
5073
5074 // Is a (classpath) directory empty?
5075 bool os::dir_is_empty(const char* path) {
5076 DIR *dir = NULL;
5077 struct dirent *ptr;
5078
5079 dir = opendir(path);
5080 if (dir == NULL) return true;
5081
5082 // Scan the directory
5083 bool result = true;
5084 char buf[sizeof(struct dirent) + MAX_PATH];
5085 while (result && (ptr = ::readdir(dir)) != NULL) {
5086 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5087 result = false;
5088 }
5089 }
5090 closedir(dir);
5091 return result;
5092 }
5093
5094 // This code originates from JDK's sysOpen and open64_w
5095 // from src/solaris/hpi/src/system_md.c
5096
5097 int os::open(const char *path, int oflag, int mode) {
5098 if (strlen(path) > MAX_PATH - 1) {
5099 errno = ENAMETOOLONG;
5100 return -1;
5101 }
5102
5103 // All file descriptors that are opened in the Java process and not
5104 // specifically destined for a subprocess should have the close-on-exec
5105 // flag set. If we don't set it, then careless 3rd party native code
5106 // might fork and exec without closing all appropriate file descriptors
5107 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5108 // turn might:
5109 //
5110 // - cause end-of-file to fail to be detected on some file
5111 // descriptors, resulting in mysterious hangs, or
5112 //
5113 // - might cause an fopen in the subprocess to fail on a system
5114 // suffering from bug 1085341.
5115 //
5116 // (Yes, the default setting of the close-on-exec flag is a Unix
5117 // design flaw)
5118 //
5119 // See:
5120 // 1085341: 32-bit stdio routines should support file descriptors >255
5121 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5122 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5123 //
5124 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5125 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5126 // because it saves a system call and removes a small window where the flag
5127 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5128 // and we fall back to using FD_CLOEXEC (see below).
5129 #ifdef O_CLOEXEC
5130 oflag |= O_CLOEXEC;
5131 #endif
5132
5133 int fd = ::open64(path, oflag, mode);
5134 if (fd == -1) return -1;
5135
5136 //If the open succeeded, the file might still be a directory
5137 {
5138 struct stat64 buf64;
5139 int ret = ::fstat64(fd, &buf64);
5140 int st_mode = buf64.st_mode;
5141
5142 if (ret != -1) {
5143 if ((st_mode & S_IFMT) == S_IFDIR) {
5144 errno = EISDIR;
5145 ::close(fd);
5146 return -1;
5147 }
5148 } else {
5149 ::close(fd);
5150 return -1;
5151 }
5152 }
5153
5154 #ifdef FD_CLOEXEC
5155 // Validate that the use of the O_CLOEXEC flag on open above worked.
5156 // With recent kernels, we will perform this check exactly once.
5157 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5158 if (!O_CLOEXEC_is_known_to_work) {
5159 int flags = ::fcntl(fd, F_GETFD);
5160 if (flags != -1) {
5161 if ((flags & FD_CLOEXEC) != 0)
5162 O_CLOEXEC_is_known_to_work = 1;
5163 else
5164 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5165 }
5166 }
5167 #endif
5168
5169 return fd;
5170 }
5171
5172
5173 // create binary file, rewriting existing file if required
5174 int os::create_binary_file(const char* path, bool rewrite_existing) {
5175 int oflags = O_WRONLY | O_CREAT;
5176 if (!rewrite_existing) {
5177 oflags |= O_EXCL;
5178 }
5179 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5180 }
5181
5182 // return current position of file pointer
5183 jlong os::current_file_offset(int fd) {
5184 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5185 }
5186
5187 // move file pointer to the specified offset
5188 jlong os::seek_to_file_offset(int fd, jlong offset) {
5189 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5190 }
5191
5192 // This code originates from JDK's sysAvailable
5193 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5194
5195 int os::available(int fd, jlong *bytes) {
5196 jlong cur, end;
5197 int mode;
5198 struct stat64 buf64;
5199
5200 if (::fstat64(fd, &buf64) >= 0) {
5201 mode = buf64.st_mode;
5202 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5203 // XXX: is the following call interruptible? If so, this might
5204 // need to go through the INTERRUPT_IO() wrapper as for other
5205 // blocking, interruptible calls in this file.
5206 int n;
5207 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5208 *bytes = n;
5209 return 1;
5210 }
5211 }
5212 }
5213 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5214 return 0;
5215 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5216 return 0;
5217 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5218 return 0;
5219 }
5220 *bytes = end - cur;
5221 return 1;
5222 }
5223
5224 // Map a block of memory.
5225 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5226 char *addr, size_t bytes, bool read_only,
5227 bool allow_exec) {
5228 int prot;
5229 int flags = MAP_PRIVATE;
5230
5231 if (read_only) {
5232 prot = PROT_READ;
5233 } else {
5234 prot = PROT_READ | PROT_WRITE;
5235 }
5236
5237 if (allow_exec) {
5238 prot |= PROT_EXEC;
5239 }
5240
5241 if (addr != NULL) {
5242 flags |= MAP_FIXED;
5243 }
5244
5245 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5246 fd, file_offset);
5247 if (mapped_address == MAP_FAILED) {
5248 return NULL;
5249 }
5250 return mapped_address;
5251 }
5252
5253
5254 // Remap a block of memory.
5255 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5256 char *addr, size_t bytes, bool read_only,
5257 bool allow_exec) {
5258 // same as map_memory() on this OS
5259 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5260 allow_exec);
5261 }
5262
5263
5264 // Unmap a block of memory.
5265 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5266 return munmap(addr, bytes) == 0;
5267 }
5268
5269 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5270
5271 static clockid_t thread_cpu_clockid(Thread* thread) {
5272 pthread_t tid = thread->osthread()->pthread_id();
5273 clockid_t clockid;
5274
5275 // Get thread clockid
5276 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5277 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5278 return clockid;
5279 }
5280
5281 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5282 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5283 // of a thread.
5284 //
5285 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5286 // the fast estimate available on the platform.
5287
5288 jlong os::current_thread_cpu_time() {
5289 if (os::Linux::supports_fast_thread_cpu_time()) {
5290 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5291 } else {
5292 // return user + sys since the cost is the same
5293 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5294 }
5295 }
5296
5297 jlong os::thread_cpu_time(Thread* thread) {
5298 // consistent with what current_thread_cpu_time() returns
5299 if (os::Linux::supports_fast_thread_cpu_time()) {
5300 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5301 } else {
5302 return slow_thread_cpu_time(thread, true /* user + sys */);
5303 }
5304 }
5305
5306 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5307 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5308 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5309 } else {
5310 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5311 }
5312 }
5313
5314 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5315 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5316 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5317 } else {
5318 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5319 }
5320 }
5321
5322 // -1 on error.
5323 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5324 pid_t tid = thread->osthread()->thread_id();
5325 char *s;
5326 char stat[2048];
5327 int statlen;
5328 char proc_name[64];
5329 int count;
5330 long sys_time, user_time;
5331 char cdummy;
5332 int idummy;
5333 long ldummy;
5334 FILE *fp;
5335
5336 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5337 fp = fopen(proc_name, "r");
5338 if (fp == NULL) return -1;
5339 statlen = fread(stat, 1, 2047, fp);
5340 stat[statlen] = '\0';
5341 fclose(fp);
5342
5343 // Skip pid and the command string. Note that we could be dealing with
5344 // weird command names, e.g. user could decide to rename java launcher
5345 // to "java 1.4.2 :)", then the stat file would look like
5346 // 1234 (java 1.4.2 :)) R ... ...
5347 // We don't really need to know the command string, just find the last
5348 // occurrence of ")" and then start parsing from there. See bug 4726580.
5349 s = strrchr(stat, ')');
5350 if (s == NULL) return -1;
5351
5352 // Skip blank chars
5353 do { s++; } while (s && isspace(*s));
5354
5355 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5356 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5357 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5358 &user_time, &sys_time);
5359 if (count != 13) return -1;
5360 if (user_sys_cpu_time) {
5361 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5362 } else {
5363 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5364 }
5365 }
5366
5367 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5368 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5369 info_ptr->may_skip_backward = false; // elapsed time not wall time
5370 info_ptr->may_skip_forward = false; // elapsed time not wall time
5371 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5372 }
5373
5374 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5375 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5376 info_ptr->may_skip_backward = false; // elapsed time not wall time
5377 info_ptr->may_skip_forward = false; // elapsed time not wall time
5378 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5379 }
5380
5381 bool os::is_thread_cpu_time_supported() {
5382 return true;
5383 }
5384
5385 // System loadavg support. Returns -1 if load average cannot be obtained.
5386 // Linux doesn't yet have a (official) notion of processor sets,
5387 // so just return the system wide load average.
5388 int os::loadavg(double loadavg[], int nelem) {
5389 return ::getloadavg(loadavg, nelem);
5390 }
5391
5392 void os::pause() {
5393 char filename[MAX_PATH];
5394 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5395 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5396 } else {
5397 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5398 }
5399
5400 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5401 if (fd != -1) {
5402 struct stat buf;
5403 ::close(fd);
5404 while (::stat(filename, &buf) == 0) {
5405 (void)::poll(NULL, 0, 100);
5406 }
5407 } else {
5408 jio_fprintf(stderr,
5409 "Could not open pause file '%s', continuing immediately.\n", filename);
5410 }
5411 }
5412
5413
5414 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5415 // comment paragraphs are worth repeating here:
5416 //
5417 // Assumption:
5418 // Only one parker can exist on an event, which is why we allocate
5419 // them per-thread. Multiple unparkers can coexist.
5420 //
5421 // _Event serves as a restricted-range semaphore.
5422 // -1 : thread is blocked, i.e. there is a waiter
5423 // 0 : neutral: thread is running or ready,
5424 // could have been signaled after a wait started
5425 // 1 : signaled - thread is running or ready
5426 //
5427 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5428 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5429 // For specifics regarding the bug see GLIBC BUGID 261237 :
5430 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5431 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5432 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5433 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5434 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5435 // and monitorenter when we're using 1-0 locking. All those operations may result in
5436 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5437 // of libpthread avoids the problem, but isn't practical.
5438 //
5439 // Possible remedies:
5440 //
5441 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5442 // This is palliative and probabilistic, however. If the thread is preempted
5443 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5444 // than the minimum period may have passed, and the abstime may be stale (in the
5445 // past) resultin in a hang. Using this technique reduces the odds of a hang
5446 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5447 //
5448 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5449 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5450 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5451 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5452 // thread.
5453 //
5454 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5455 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5456 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5457 // This also works well. In fact it avoids kernel-level scalability impediments
5458 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5459 // timers in a graceful fashion.
5460 //
5461 // 4. When the abstime value is in the past it appears that control returns
5462 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5463 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5464 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5465 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5466 // It may be possible to avoid reinitialization by checking the return
5467 // value from pthread_cond_timedwait(). In addition to reinitializing the
5468 // condvar we must establish the invariant that cond_signal() is only called
5469 // within critical sections protected by the adjunct mutex. This prevents
5470 // cond_signal() from "seeing" a condvar that's in the midst of being
5471 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5472 // desirable signal-after-unlock optimization that avoids futile context switching.
5473 //
5474 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5475 // structure when a condvar is used or initialized. cond_destroy() would
5476 // release the helper structure. Our reinitialize-after-timedwait fix
5477 // put excessive stress on malloc/free and locks protecting the c-heap.
5478 //
5479 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5480 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5481 // and only enabling the work-around for vulnerable environments.
5482
5483 // utility to compute the abstime argument to timedwait:
5484 // millis is the relative timeout time
5485 // abstime will be the absolute timeout time
5486 // TODO: replace compute_abstime() with unpackTime()
5487
5488 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5489 if (millis < 0) millis = 0;
5490
5491 jlong seconds = millis / 1000;
5492 millis %= 1000;
5493 if (seconds > 50000000) { // see man cond_timedwait(3T)
5494 seconds = 50000000;
5495 }
5496
5497 if (os::supports_monotonic_clock()) {
5498 struct timespec now;
5499 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5500 assert_status(status == 0, status, "clock_gettime");
5501 abstime->tv_sec = now.tv_sec + seconds;
5502 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5503 if (nanos >= NANOSECS_PER_SEC) {
5504 abstime->tv_sec += 1;
5505 nanos -= NANOSECS_PER_SEC;
5506 }
5507 abstime->tv_nsec = nanos;
5508 } else {
5509 struct timeval now;
5510 int status = gettimeofday(&now, NULL);
5511 assert(status == 0, "gettimeofday");
5512 abstime->tv_sec = now.tv_sec + seconds;
5513 long usec = now.tv_usec + millis * 1000;
5514 if (usec >= 1000000) {
5515 abstime->tv_sec += 1;
5516 usec -= 1000000;
5517 }
5518 abstime->tv_nsec = usec * 1000;
5519 }
5520 return abstime;
5521 }
5522
5523 void os::PlatformEvent::park() { // AKA "down()"
5524 // Transitions for _Event:
5525 // -1 => -1 : illegal
5526 // 1 => 0 : pass - return immediately
5527 // 0 => -1 : block; then set _Event to 0 before returning
5528
5529 // Invariant: Only the thread associated with the Event/PlatformEvent
5530 // may call park().
5531 // TODO: assert that _Assoc != NULL or _Assoc == Self
5532 assert(_nParked == 0, "invariant");
5533
5534 int v;
5535 for (;;) {
5536 v = _Event;
5537 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5538 }
5539 guarantee(v >= 0, "invariant");
5540 if (v == 0) {
5541 // Do this the hard way by blocking ...
5542 int status = pthread_mutex_lock(_mutex);
5543 assert_status(status == 0, status, "mutex_lock");
5544 guarantee(_nParked == 0, "invariant");
5545 ++_nParked;
5546 while (_Event < 0) {
5547 status = pthread_cond_wait(_cond, _mutex);
5548 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5549 // Treat this the same as if the wait was interrupted
5550 if (status == ETIME) { status = EINTR; }
5551 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5552 }
5553 --_nParked;
5554
5555 _Event = 0;
5556 status = pthread_mutex_unlock(_mutex);
5557 assert_status(status == 0, status, "mutex_unlock");
5558 // Paranoia to ensure our locked and lock-free paths interact
5559 // correctly with each other.
5560 OrderAccess::fence();
5561 }
5562 guarantee(_Event >= 0, "invariant");
5563 }
5564
5565 int os::PlatformEvent::park(jlong millis) {
5566 // Transitions for _Event:
5567 // -1 => -1 : illegal
5568 // 1 => 0 : pass - return immediately
5569 // 0 => -1 : block; then set _Event to 0 before returning
5570
5571 guarantee(_nParked == 0, "invariant");
5572
5573 int v;
5574 for (;;) {
5575 v = _Event;
5576 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5577 }
5578 guarantee(v >= 0, "invariant");
5579 if (v != 0) return OS_OK;
5580
5581 // We do this the hard way, by blocking the thread.
5582 // Consider enforcing a minimum timeout value.
5583 struct timespec abst;
5584 compute_abstime(&abst, millis);
5585
5586 int ret = OS_TIMEOUT;
5587 int status = pthread_mutex_lock(_mutex);
5588 assert_status(status == 0, status, "mutex_lock");
5589 guarantee(_nParked == 0, "invariant");
5590 ++_nParked;
5591
5592 // Object.wait(timo) will return because of
5593 // (a) notification
5594 // (b) timeout
5595 // (c) thread.interrupt
5596 //
5597 // Thread.interrupt and object.notify{All} both call Event::set.
5598 // That is, we treat thread.interrupt as a special case of notification.
5599 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5600 // We assume all ETIME returns are valid.
5601 //
5602 // TODO: properly differentiate simultaneous notify+interrupt.
5603 // In that case, we should propagate the notify to another waiter.
5604
5605 while (_Event < 0) {
5606 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5607 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5608 pthread_cond_destroy(_cond);
5609 pthread_cond_init(_cond, os::Linux::condAttr());
5610 }
5611 assert_status(status == 0 || status == EINTR ||
5612 status == ETIME || status == ETIMEDOUT,
5613 status, "cond_timedwait");
5614 if (!FilterSpuriousWakeups) break; // previous semantics
5615 if (status == ETIME || status == ETIMEDOUT) break;
5616 // We consume and ignore EINTR and spurious wakeups.
5617 }
5618 --_nParked;
5619 if (_Event >= 0) {
5620 ret = OS_OK;
5621 }
5622 _Event = 0;
5623 status = pthread_mutex_unlock(_mutex);
5624 assert_status(status == 0, status, "mutex_unlock");
5625 assert(_nParked == 0, "invariant");
5626 // Paranoia to ensure our locked and lock-free paths interact
5627 // correctly with each other.
5628 OrderAccess::fence();
5629 return ret;
5630 }
5631
5632 void os::PlatformEvent::unpark() {
5633 // Transitions for _Event:
5634 // 0 => 1 : just return
5635 // 1 => 1 : just return
5636 // -1 => either 0 or 1; must signal target thread
5637 // That is, we can safely transition _Event from -1 to either
5638 // 0 or 1.
5639 // See also: "Semaphores in Plan 9" by Mullender & Cox
5640 //
5641 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5642 // that it will take two back-to-back park() calls for the owning
5643 // thread to block. This has the benefit of forcing a spurious return
5644 // from the first park() call after an unpark() call which will help
5645 // shake out uses of park() and unpark() without condition variables.
5646
5647 if (Atomic::xchg(1, &_Event) >= 0) return;
5648
5649 // Wait for the thread associated with the event to vacate
5650 int status = pthread_mutex_lock(_mutex);
5651 assert_status(status == 0, status, "mutex_lock");
5652 int AnyWaiters = _nParked;
5653 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5654 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5655 AnyWaiters = 0;
5656 pthread_cond_signal(_cond);
5657 }
5658 status = pthread_mutex_unlock(_mutex);
5659 assert_status(status == 0, status, "mutex_unlock");
5660 if (AnyWaiters != 0) {
5661 // Note that we signal() *after* dropping the lock for "immortal" Events.
5662 // This is safe and avoids a common class of futile wakeups. In rare
5663 // circumstances this can cause a thread to return prematurely from
5664 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5665 // will simply re-test the condition and re-park itself.
5666 // This provides particular benefit if the underlying platform does not
5667 // provide wait morphing.
5668 status = pthread_cond_signal(_cond);
5669 assert_status(status == 0, status, "cond_signal");
5670 }
5671 }
5672
5673
5674 // JSR166
5675 // -------------------------------------------------------
5676
5677 // The solaris and linux implementations of park/unpark are fairly
5678 // conservative for now, but can be improved. They currently use a
5679 // mutex/condvar pair, plus a a count.
5680 // Park decrements count if > 0, else does a condvar wait. Unpark
5681 // sets count to 1 and signals condvar. Only one thread ever waits
5682 // on the condvar. Contention seen when trying to park implies that someone
5683 // is unparking you, so don't wait. And spurious returns are fine, so there
5684 // is no need to track notifications.
5685
5686 // This code is common to linux and solaris and will be moved to a
5687 // common place in dolphin.
5688 //
5689 // The passed in time value is either a relative time in nanoseconds
5690 // or an absolute time in milliseconds. Either way it has to be unpacked
5691 // into suitable seconds and nanoseconds components and stored in the
5692 // given timespec structure.
5693 // Given time is a 64-bit value and the time_t used in the timespec is only
5694 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5695 // overflow if times way in the future are given. Further on Solaris versions
5696 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5697 // number of seconds, in abstime, is less than current_time + 100,000,000.
5698 // As it will be 28 years before "now + 100000000" will overflow we can
5699 // ignore overflow and just impose a hard-limit on seconds using the value
5700 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5701 // years from "now".
5702
5703 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5704 assert(time > 0, "convertTime");
5705 time_t max_secs = 0;
5706
5707 if (!os::supports_monotonic_clock() || isAbsolute) {
5708 struct timeval now;
5709 int status = gettimeofday(&now, NULL);
5710 assert(status == 0, "gettimeofday");
5711
5712 max_secs = now.tv_sec + MAX_SECS;
5713
5714 if (isAbsolute) {
5715 jlong secs = time / 1000;
5716 if (secs > max_secs) {
5717 absTime->tv_sec = max_secs;
5718 } else {
5719 absTime->tv_sec = secs;
5720 }
5721 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5722 } else {
5723 jlong secs = time / NANOSECS_PER_SEC;
5724 if (secs >= MAX_SECS) {
5725 absTime->tv_sec = max_secs;
5726 absTime->tv_nsec = 0;
5727 } else {
5728 absTime->tv_sec = now.tv_sec + secs;
5729 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5730 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5731 absTime->tv_nsec -= NANOSECS_PER_SEC;
5732 ++absTime->tv_sec; // note: this must be <= max_secs
5733 }
5734 }
5735 }
5736 } else {
5737 // must be relative using monotonic clock
5738 struct timespec now;
5739 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5740 assert_status(status == 0, status, "clock_gettime");
5741 max_secs = now.tv_sec + MAX_SECS;
5742 jlong secs = time / NANOSECS_PER_SEC;
5743 if (secs >= MAX_SECS) {
5744 absTime->tv_sec = max_secs;
5745 absTime->tv_nsec = 0;
5746 } else {
5747 absTime->tv_sec = now.tv_sec + secs;
5748 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5749 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5750 absTime->tv_nsec -= NANOSECS_PER_SEC;
5751 ++absTime->tv_sec; // note: this must be <= max_secs
5752 }
5753 }
5754 }
5755 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5756 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5757 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5758 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5759 }
5760
5761 void Parker::park(bool isAbsolute, jlong time) {
5762 // Ideally we'd do something useful while spinning, such
5763 // as calling unpackTime().
5764
5765 // Optional fast-path check:
5766 // Return immediately if a permit is available.
5767 // We depend on Atomic::xchg() having full barrier semantics
5768 // since we are doing a lock-free update to _counter.
5769 if (Atomic::xchg(0, &_counter) > 0) return;
5770
5771 Thread* thread = Thread::current();
5772 assert(thread->is_Java_thread(), "Must be JavaThread");
5773 JavaThread *jt = (JavaThread *)thread;
5774
5775 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5776 // Check interrupt before trying to wait
5777 if (Thread::is_interrupted(thread, false)) {
5778 return;
5779 }
5780
5781 // Next, demultiplex/decode time arguments
5782 timespec absTime;
5783 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5784 return;
5785 }
5786 if (time > 0) {
5787 unpackTime(&absTime, isAbsolute, time);
5788 }
5789
5790
5791 // Enter safepoint region
5792 // Beware of deadlocks such as 6317397.
5793 // The per-thread Parker:: mutex is a classic leaf-lock.
5794 // In particular a thread must never block on the Threads_lock while
5795 // holding the Parker:: mutex. If safepoints are pending both the
5796 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5797 ThreadBlockInVM tbivm(jt);
5798
5799 // Don't wait if cannot get lock since interference arises from
5800 // unblocking. Also. check interrupt before trying wait
5801 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5802 return;
5803 }
5804
5805 int status;
5806 if (_counter > 0) { // no wait needed
5807 _counter = 0;
5808 status = pthread_mutex_unlock(_mutex);
5809 assert(status == 0, "invariant");
5810 // Paranoia to ensure our locked and lock-free paths interact
5811 // correctly with each other and Java-level accesses.
5812 OrderAccess::fence();
5813 return;
5814 }
5815
5816 #ifdef ASSERT
5817 // Don't catch signals while blocked; let the running threads have the signals.
5818 // (This allows a debugger to break into the running thread.)
5819 sigset_t oldsigs;
5820 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5821 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5822 #endif
5823
5824 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5825 jt->set_suspend_equivalent();
5826 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5827
5828 assert(_cur_index == -1, "invariant");
5829 if (time == 0) {
5830 _cur_index = REL_INDEX; // arbitrary choice when not timed
5831 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5832 } else {
5833 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5834 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5835 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5836 pthread_cond_destroy(&_cond[_cur_index]);
5837 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5838 }
5839 }
5840 _cur_index = -1;
5841 assert_status(status == 0 || status == EINTR ||
5842 status == ETIME || status == ETIMEDOUT,
5843 status, "cond_timedwait");
5844
5845 #ifdef ASSERT
5846 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5847 #endif
5848
5849 _counter = 0;
5850 status = pthread_mutex_unlock(_mutex);
5851 assert_status(status == 0, status, "invariant");
5852 // Paranoia to ensure our locked and lock-free paths interact
5853 // correctly with each other and Java-level accesses.
5854 OrderAccess::fence();
5855
5856 // If externally suspended while waiting, re-suspend
5857 if (jt->handle_special_suspend_equivalent_condition()) {
5858 jt->java_suspend_self();
5859 }
5860 }
5861
5862 void Parker::unpark() {
5863 int status = pthread_mutex_lock(_mutex);
5864 assert(status == 0, "invariant");
5865 const int s = _counter;
5866 _counter = 1;
5867 if (s < 1) {
5868 // thread might be parked
5869 if (_cur_index != -1) {
5870 // thread is definitely parked
5871 if (WorkAroundNPTLTimedWaitHang) {
5872 status = pthread_cond_signal(&_cond[_cur_index]);
5873 assert(status == 0, "invariant");
5874 status = pthread_mutex_unlock(_mutex);
5875 assert(status == 0, "invariant");
5876 } else {
5877 status = pthread_mutex_unlock(_mutex);
5878 assert(status == 0, "invariant");
5879 status = pthread_cond_signal(&_cond[_cur_index]);
5880 assert(status == 0, "invariant");
5881 }
5882 } else {
5883 pthread_mutex_unlock(_mutex);
5884 assert(status == 0, "invariant");
5885 }
5886 } else {
5887 pthread_mutex_unlock(_mutex);
5888 assert(status == 0, "invariant");
5889 }
5890 }
5891
5892
5893 extern char** environ;
5894
5895 // Run the specified command in a separate process. Return its exit value,
5896 // or -1 on failure (e.g. can't fork a new process).
5897 // Unlike system(), this function can be called from signal handler. It
5898 // doesn't block SIGINT et al.
5899 int os::fork_and_exec(char* cmd) {
5900 const char * argv[4] = {"sh", "-c", cmd, NULL};
5901
5902 pid_t pid = fork();
5903
5904 if (pid < 0) {
5905 // fork failed
5906 return -1;
5907
5908 } else if (pid == 0) {
5909 // child process
5910
5911 execve("/bin/sh", (char* const*)argv, environ);
5912
5913 // execve failed
5914 _exit(-1);
5915
5916 } else {
5917 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5918 // care about the actual exit code, for now.
5919
5920 int status;
5921
5922 // Wait for the child process to exit. This returns immediately if
5923 // the child has already exited. */
5924 while (waitpid(pid, &status, 0) < 0) {
5925 switch (errno) {
5926 case ECHILD: return 0;
5927 case EINTR: break;
5928 default: return -1;
5929 }
5930 }
5931
5932 if (WIFEXITED(status)) {
5933 // The child exited normally; get its exit code.
5934 return WEXITSTATUS(status);
5935 } else if (WIFSIGNALED(status)) {
5936 // The child exited because of a signal
5937 // The best value to return is 0x80 + signal number,
5938 // because that is what all Unix shells do, and because
5939 // it allows callers to distinguish between process exit and
5940 // process death by signal.
5941 return 0x80 + WTERMSIG(status);
5942 } else {
5943 // Unknown exit code; pass it through
5944 return status;
5945 }
5946 }
5947 }
5948
5949 // is_headless_jre()
5950 //
5951 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5952 // in order to report if we are running in a headless jre
5953 //
5954 // Since JDK8 xawt/libmawt.so was moved into the same directory
5955 // as libawt.so, and renamed libawt_xawt.so
5956 //
5957 bool os::is_headless_jre() {
5958 struct stat statbuf;
5959 char buf[MAXPATHLEN];
5960 char libmawtpath[MAXPATHLEN];
5961 const char *xawtstr = "/xawt/libmawt.so";
5962 const char *new_xawtstr = "/libawt_xawt.so";
5963 char *p;
5964
5965 // Get path to libjvm.so
5966 os::jvm_path(buf, sizeof(buf));
5967
5968 // Get rid of libjvm.so
5969 p = strrchr(buf, '/');
5970 if (p == NULL) {
5971 return false;
5972 } else {
5973 *p = '\0';
5974 }
5975
5976 // Get rid of client or server
5977 p = strrchr(buf, '/');
5978 if (p == NULL) {
5979 return false;
5980 } else {
5981 *p = '\0';
5982 }
5983
5984 // check xawt/libmawt.so
5985 strcpy(libmawtpath, buf);
5986 strcat(libmawtpath, xawtstr);
5987 if (::stat(libmawtpath, &statbuf) == 0) return false;
5988
5989 // check libawt_xawt.so
5990 strcpy(libmawtpath, buf);
5991 strcat(libmawtpath, new_xawtstr);
5992 if (::stat(libmawtpath, &statbuf) == 0) return false;
5993
5994 return true;
5995 }
5996
5997 // Get the default path to the core file
5998 // Returns the length of the string
5999 int os::get_core_path(char* buffer, size_t bufferSize) {
6000 /*
6001 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
6002 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
6003 */
6004 const int core_pattern_len = 129;
6005 char core_pattern[core_pattern_len] = {0};
6006
6007 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6008 if (core_pattern_file != -1) {
6009 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6010 ::close(core_pattern_file);
6011
6012 if (ret > 0) {
6013 char *last_char = core_pattern + strlen(core_pattern) - 1;
6014
6015 if (*last_char == '\n') {
6016 *last_char = '\0';
6017 }
6018 }
6019 }
6020
6021 if (strlen(core_pattern) == 0) {
6022 return -1;
6023 }
6024
6025 char *pid_pos = strstr(core_pattern, "%p");
6026 int written;
6027
6028 if (core_pattern[0] == '/') {
6029 written = jio_snprintf(buffer, bufferSize, core_pattern);
6030 } else {
6031 char cwd[PATH_MAX];
6032
6033 const char* p = get_current_directory(cwd, PATH_MAX);
6034 if (p == NULL) {
6035 return -1;
6036 }
6037
6038 if (core_pattern[0] == '|') {
6039 written = jio_snprintf(buffer, bufferSize,
6040 "\"%s\" (or dumping to %s/core.%d)",
6041 &core_pattern[1], p, current_process_id());
6042 } else {
6043 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6044 }
6045 }
6046
6047 if (written < 0) {
6048 return -1;
6049 }
6050
6051 if (((size_t)written < bufferSize) && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6052 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6053
6054 if (core_uses_pid_file != -1) {
6055 char core_uses_pid = 0;
6056 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6057 ::close(core_uses_pid_file);
6058
6059 if (core_uses_pid == '1') {
6060 jio_snprintf(buffer + written, bufferSize - written,
6061 ".%d", current_process_id());
6062 }
6063 }
6064 }
6065
6066 return strlen(buffer);
6067 }
6068
6069 /////////////// Unit tests ///////////////
6070
6071 #ifndef PRODUCT
6072
6073 #define test_log(...) \
6074 do { \
6075 if (VerboseInternalVMTests) { \
6076 tty->print_cr(__VA_ARGS__); \
6077 tty->flush(); \
6078 } \
6079 } while (false)
6080
6081 class TestReserveMemorySpecial : AllStatic {
6082 public:
6083 static void small_page_write(void* addr, size_t size) {
6084 size_t page_size = os::vm_page_size();
6085
6086 char* end = (char*)addr + size;
6087 for (char* p = (char*)addr; p < end; p += page_size) {
6088 *p = 1;
6089 }
6090 }
6091
6092 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6093 if (!UseHugeTLBFS) {
6094 return;
6095 }
6096
6097 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6098
6099 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6100
6101 if (addr != NULL) {
6102 small_page_write(addr, size);
6103
6104 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6105 }
6106 }
6107
6108 static void test_reserve_memory_special_huge_tlbfs_only() {
6109 if (!UseHugeTLBFS) {
6110 return;
6111 }
6112
6113 size_t lp = os::large_page_size();
6114
6115 for (size_t size = lp; size <= lp * 10; size += lp) {
6116 test_reserve_memory_special_huge_tlbfs_only(size);
6117 }
6118 }
6119
6120 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6121 size_t lp = os::large_page_size();
6122 size_t ag = os::vm_allocation_granularity();
6123
6124 // sizes to test
6125 const size_t sizes[] = {
6126 lp, lp + ag, lp + lp / 2, lp * 2,
6127 lp * 2 + ag, lp * 2 - ag, lp * 2 + lp / 2,
6128 lp * 10, lp * 10 + lp / 2
6129 };
6130 const int num_sizes = sizeof(sizes) / sizeof(size_t);
6131
6132 // For each size/alignment combination, we test three scenarios:
6133 // 1) with req_addr == NULL
6134 // 2) with a non-null req_addr at which we expect to successfully allocate
6135 // 3) with a non-null req_addr which contains a pre-existing mapping, at which we
6136 // expect the allocation to either fail or to ignore req_addr
6137
6138 // Pre-allocate two areas; they shall be as large as the largest allocation
6139 // and aligned to the largest alignment we will be testing.
6140 const size_t mapping_size = sizes[num_sizes - 1] * 2;
6141 char* const mapping1 = (char*) ::mmap(NULL, mapping_size,
6142 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6143 -1, 0);
6144 assert(mapping1 != MAP_FAILED, "should work");
6145
6146 char* const mapping2 = (char*) ::mmap(NULL, mapping_size,
6147 PROT_NONE, MAP_PRIVATE|MAP_ANONYMOUS|MAP_NORESERVE,
6148 -1, 0);
6149 assert(mapping2 != MAP_FAILED, "should work");
6150
6151 // Unmap the first mapping, but leave the second mapping intact: the first
6152 // mapping will serve as a value for a "good" req_addr (case 2). The second
6153 // mapping, still intact, as "bad" req_addr (case 3).
6154 ::munmap(mapping1, mapping_size);
6155
6156 // Case 1
6157 test_log("%s, req_addr NULL:", __FUNCTION__);
6158 test_log("size align result");
6159
6160 for (int i = 0; i < num_sizes; i++) {
6161 const size_t size = sizes[i];
6162 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6163 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6164 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " -> " PTR_FORMAT " %s",
6165 size, alignment, p, (p != NULL ? "" : "(failed)"));
6166 if (p != NULL) {
6167 assert(is_ptr_aligned(p, alignment), "must be");
6168 small_page_write(p, size);
6169 os::Linux::release_memory_special_huge_tlbfs(p, size);
6170 }
6171 }
6172 }
6173
6174 // Case 2
6175 test_log("%s, req_addr non-NULL:", __FUNCTION__);
6176 test_log("size align req_addr result");
6177
6178 for (int i = 0; i < num_sizes; i++) {
6179 const size_t size = sizes[i];
6180 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6181 char* const req_addr = (char*) align_ptr_up(mapping1, alignment);
6182 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6183 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6184 size, alignment, req_addr, p,
6185 ((p != NULL ? (p == req_addr ? "(exact match)" : "") : "(failed)")));
6186 if (p != NULL) {
6187 assert(p == req_addr, "must be");
6188 small_page_write(p, size);
6189 os::Linux::release_memory_special_huge_tlbfs(p, size);
6190 }
6191 }
6192 }
6193
6194 // Case 3
6195 test_log("%s, req_addr non-NULL with preexisting mapping:", __FUNCTION__);
6196 test_log("size align req_addr result");
6197
6198 for (int i = 0; i < num_sizes; i++) {
6199 const size_t size = sizes[i];
6200 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6201 char* const req_addr = (char*) align_ptr_up(mapping2, alignment);
6202 char* p = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, req_addr, false);
6203 test_log(SIZE_FORMAT_HEX " " SIZE_FORMAT_HEX " " PTR_FORMAT " -> " PTR_FORMAT " %s",
6204 size, alignment, req_addr, p,
6205 ((p != NULL ? "" : "(failed)")));
6206 // as the area around req_addr contains already existing mappings, the API should always
6207 // return NULL (as per contract, it cannot return another address)
6208 assert(p == NULL, "must be");
6209 }
6210 }
6211
6212 ::munmap(mapping2, mapping_size);
6213
6214 }
6215
6216 static void test_reserve_memory_special_huge_tlbfs() {
6217 if (!UseHugeTLBFS) {
6218 return;
6219 }
6220
6221 test_reserve_memory_special_huge_tlbfs_only();
6222 test_reserve_memory_special_huge_tlbfs_mixed();
6223 }
6224
6225 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6226 if (!UseSHM) {
6227 return;
6228 }
6229
6230 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6231
6232 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6233
6234 if (addr != NULL) {
6235 assert(is_ptr_aligned(addr, alignment), "Check");
6236 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6237
6238 small_page_write(addr, size);
6239
6240 os::Linux::release_memory_special_shm(addr, size);
6241 }
6242 }
6243
6244 static void test_reserve_memory_special_shm() {
6245 size_t lp = os::large_page_size();
6246 size_t ag = os::vm_allocation_granularity();
6247
6248 for (size_t size = ag; size < lp * 3; size += ag) {
6249 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6250 test_reserve_memory_special_shm(size, alignment);
6251 }
6252 }
6253 }
6254
6255 static void test() {
6256 test_reserve_memory_special_huge_tlbfs();
6257 test_reserve_memory_special_shm();
6258 }
6259 };
6260
6261 void TestReserveMemorySpecial_test() {
6262 TestReserveMemorySpecial::test();
6263 }
6264
6265 #endif