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 // Used to protect dlsym() calls
162 static pthread_mutex_t dl_mutex;
163
164 // Declarations
165 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time);
166
167 // utility functions
168
169 static int SR_initialize();
170
171 julong os::available_memory() {
172 return Linux::available_memory();
173 }
174
175 julong os::Linux::available_memory() {
176 // values in struct sysinfo are "unsigned long"
177 struct sysinfo si;
178 sysinfo(&si);
179
180 return (julong)si.freeram * si.mem_unit;
181 }
182
183 julong os::physical_memory() {
184 return Linux::physical_memory();
185 }
186
187 // Return true if user is running as root.
188
189 bool os::have_special_privileges() {
190 static bool init = false;
191 static bool privileges = false;
192 if (!init) {
193 privileges = (getuid() != geteuid()) || (getgid() != getegid());
194 init = true;
195 }
196 return privileges;
197 }
198
199
200 #ifndef SYS_gettid
201 // i386: 224, ia64: 1105, amd64: 186, sparc 143
202 #ifdef __ia64__
203 #define SYS_gettid 1105
204 #elif __i386__
205 #define SYS_gettid 224
206 #elif __amd64__
207 #define SYS_gettid 186
208 #elif __sparc__
209 #define SYS_gettid 143
210 #else
211 #error define gettid for the arch
212 #endif
213 #endif
214
215 // Cpu architecture string
216 #if defined(ZERO)
217 static char cpu_arch[] = ZERO_LIBARCH;
218 #elif defined(IA64)
219 static char cpu_arch[] = "ia64";
220 #elif defined(IA32)
221 static char cpu_arch[] = "i386";
222 #elif defined(AMD64)
223 static char cpu_arch[] = "amd64";
224 #elif defined(ARM)
225 static char cpu_arch[] = "arm";
226 #elif defined(PPC32)
227 static char cpu_arch[] = "ppc";
228 #elif defined(PPC64)
229 static char cpu_arch[] = "ppc64";
230 #elif defined(SPARC)
231 #ifdef _LP64
232 static char cpu_arch[] = "sparcv9";
233 #else
234 static char cpu_arch[] = "sparc";
235 #endif
236 #elif defined(AARCH64)
237 static char cpu_arch[] = "aarch64";
238 #else
239 #error Add appropriate cpu_arch setting
240 #endif
241
242
243 // pid_t gettid()
244 //
245 // Returns the kernel thread id of the currently running thread. Kernel
246 // thread id is used to access /proc.
247 //
248 // (Note that getpid() on LinuxThreads returns kernel thread id too; but
249 // on NPTL, it returns the same pid for all threads, as required by POSIX.)
250 //
251 pid_t os::Linux::gettid() {
252 int rslt = syscall(SYS_gettid);
253 if (rslt == -1) {
254 // old kernel, no NPTL support
255 return getpid();
256 } else {
257 return (pid_t)rslt;
258 }
259 }
260
261 // Most versions of linux have a bug where the number of processors are
262 // determined by looking at the /proc file system. In a chroot environment,
263 // the system call returns 1. This causes the VM to act as if it is
264 // a single processor and elide locking (see is_MP() call).
265 static bool unsafe_chroot_detected = false;
266 static const char *unstable_chroot_error = "/proc file system not found.\n"
267 "Java may be unstable running multithreaded in a chroot "
268 "environment on Linux when /proc filesystem is not mounted.";
269
270 void os::Linux::initialize_system_info() {
271 set_processor_count(sysconf(_SC_NPROCESSORS_CONF));
272 if (processor_count() == 1) {
273 pid_t pid = os::Linux::gettid();
274 char fname[32];
275 jio_snprintf(fname, sizeof(fname), "/proc/%d", pid);
276 FILE *fp = fopen(fname, "r");
277 if (fp == NULL) {
278 unsafe_chroot_detected = true;
279 } else {
280 fclose(fp);
281 }
282 }
283 _physical_memory = (julong)sysconf(_SC_PHYS_PAGES) * (julong)sysconf(_SC_PAGESIZE);
284 assert(processor_count() > 0, "linux error");
285 }
286
287 void os::init_system_properties_values() {
288 // The next steps are taken in the product version:
289 //
290 // Obtain the JAVA_HOME value from the location of libjvm.so.
291 // This library should be located at:
292 // <JAVA_HOME>/jre/lib/<arch>/{client|server}/libjvm.so.
293 //
294 // If "/jre/lib/" appears at the right place in the path, then we
295 // assume libjvm.so is installed in a JDK and we use this path.
296 //
297 // Otherwise exit with message: "Could not create the Java virtual machine."
298 //
299 // The following extra steps are taken in the debugging version:
300 //
301 // If "/jre/lib/" does NOT appear at the right place in the path
302 // instead of exit check for $JAVA_HOME environment variable.
303 //
304 // If it is defined and we are able to locate $JAVA_HOME/jre/lib/<arch>,
305 // then we append a fake suffix "hotspot/libjvm.so" to this path so
306 // it looks like libjvm.so is installed there
307 // <JAVA_HOME>/jre/lib/<arch>/hotspot/libjvm.so.
308 //
309 // Otherwise exit.
310 //
311 // Important note: if the location of libjvm.so changes this
312 // code needs to be changed accordingly.
313
314 // See ld(1):
315 // The linker uses the following search paths to locate required
316 // shared libraries:
317 // 1: ...
318 // ...
319 // 7: The default directories, normally /lib and /usr/lib.
320 #if defined(AMD64) || defined(_LP64) && (defined(SPARC) || defined(PPC) || defined(S390))
321 #define DEFAULT_LIBPATH "/usr/lib64:/lib64:/lib:/usr/lib"
322 #else
323 #define DEFAULT_LIBPATH "/lib:/usr/lib"
324 #endif
325
326 // Base path of extensions installed on the system.
327 #define SYS_EXT_DIR "/usr/java/packages"
328 #define EXTENSIONS_DIR "/lib/ext"
329
330 // Buffer that fits several sprintfs.
331 // Note that the space for the colon and the trailing null are provided
332 // by the nulls included by the sizeof operator.
333 const size_t bufsize =
334 MAX2((size_t)MAXPATHLEN, // For dll_dir & friends.
335 (size_t)MAXPATHLEN + sizeof(EXTENSIONS_DIR) + sizeof(SYS_EXT_DIR) + sizeof(EXTENSIONS_DIR)); // extensions dir
336 char *buf = (char *)NEW_C_HEAP_ARRAY(char, bufsize, mtInternal);
337
338 // sysclasspath, java_home, dll_dir
339 {
340 char *pslash;
341 os::jvm_path(buf, bufsize);
342
343 // Found the full path to libjvm.so.
344 // Now cut the path to <java_home>/jre if we can.
345 pslash = strrchr(buf, '/');
346 if (pslash != NULL) {
347 *pslash = '\0'; // Get rid of /libjvm.so.
348 }
349 pslash = strrchr(buf, '/');
350 if (pslash != NULL) {
351 *pslash = '\0'; // Get rid of /{client|server|hotspot}.
352 }
353 Arguments::set_dll_dir(buf);
354
355 if (pslash != NULL) {
356 pslash = strrchr(buf, '/');
357 if (pslash != NULL) {
358 *pslash = '\0'; // Get rid of /<arch>.
359 pslash = strrchr(buf, '/');
360 if (pslash != NULL) {
361 *pslash = '\0'; // Get rid of /lib.
362 }
363 }
364 }
365 Arguments::set_java_home(buf);
366 set_boot_path('/', ':');
367 }
368
369 // Where to look for native libraries.
370 //
371 // Note: Due to a legacy implementation, most of the library path
372 // is set in the launcher. This was to accomodate linking restrictions
373 // on legacy Linux implementations (which are no longer supported).
374 // Eventually, all the library path setting will be done here.
375 //
376 // However, to prevent the proliferation of improperly built native
377 // libraries, the new path component /usr/java/packages is added here.
378 // Eventually, all the library path setting will be done here.
379 {
380 // Get the user setting of LD_LIBRARY_PATH, and prepended it. It
381 // should always exist (until the legacy problem cited above is
382 // addressed).
383 const char *v = ::getenv("LD_LIBRARY_PATH");
384 const char *v_colon = ":";
385 if (v == NULL) { v = ""; v_colon = ""; }
386 // That's +1 for the colon and +1 for the trailing '\0'.
387 char *ld_library_path = (char *)NEW_C_HEAP_ARRAY(char,
388 strlen(v) + 1 +
389 sizeof(SYS_EXT_DIR) + sizeof("/lib/") + strlen(cpu_arch) + sizeof(DEFAULT_LIBPATH) + 1,
390 mtInternal);
391 sprintf(ld_library_path, "%s%s" SYS_EXT_DIR "/lib/%s:" DEFAULT_LIBPATH, v, v_colon, cpu_arch);
392 Arguments::set_library_path(ld_library_path);
393 FREE_C_HEAP_ARRAY(char, ld_library_path);
394 }
395
396 // Extensions directories.
397 sprintf(buf, "%s" EXTENSIONS_DIR ":" SYS_EXT_DIR EXTENSIONS_DIR, Arguments::get_java_home());
398 Arguments::set_ext_dirs(buf);
399
400 FREE_C_HEAP_ARRAY(char, buf);
401
402 #undef DEFAULT_LIBPATH
403 #undef SYS_EXT_DIR
404 #undef EXTENSIONS_DIR
405 }
406
407 ////////////////////////////////////////////////////////////////////////////////
408 // breakpoint support
409
410 void os::breakpoint() {
411 BREAKPOINT;
412 }
413
414 extern "C" void breakpoint() {
415 // use debugger to set breakpoint here
416 }
417
418 ////////////////////////////////////////////////////////////////////////////////
419 // signal support
420
421 debug_only(static bool signal_sets_initialized = false);
422 static sigset_t unblocked_sigs, vm_sigs, allowdebug_blocked_sigs;
423
424 bool os::Linux::is_sig_ignored(int sig) {
425 struct sigaction oact;
426 sigaction(sig, (struct sigaction*)NULL, &oact);
427 void* ohlr = oact.sa_sigaction ? CAST_FROM_FN_PTR(void*, oact.sa_sigaction)
428 : CAST_FROM_FN_PTR(void*, oact.sa_handler);
429 if (ohlr == CAST_FROM_FN_PTR(void*, SIG_IGN)) {
430 return true;
431 } else {
432 return false;
433 }
434 }
435
436 void os::Linux::signal_sets_init() {
437 // Should also have an assertion stating we are still single-threaded.
438 assert(!signal_sets_initialized, "Already initialized");
439 // Fill in signals that are necessarily unblocked for all threads in
440 // the VM. Currently, we unblock the following signals:
441 // SHUTDOWN{1,2,3}_SIGNAL: for shutdown hooks support (unless over-ridden
442 // by -Xrs (=ReduceSignalUsage));
443 // BREAK_SIGNAL which is unblocked only by the VM thread and blocked by all
444 // other threads. The "ReduceSignalUsage" boolean tells us not to alter
445 // the dispositions or masks wrt these signals.
446 // Programs embedding the VM that want to use the above signals for their
447 // own purposes must, at this time, use the "-Xrs" option to prevent
448 // interference with shutdown hooks and BREAK_SIGNAL thread dumping.
449 // (See bug 4345157, and other related bugs).
450 // In reality, though, unblocking these signals is really a nop, since
451 // these signals are not blocked by default.
452 sigemptyset(&unblocked_sigs);
453 sigemptyset(&allowdebug_blocked_sigs);
454 sigaddset(&unblocked_sigs, SIGILL);
455 sigaddset(&unblocked_sigs, SIGSEGV);
456 sigaddset(&unblocked_sigs, SIGBUS);
457 sigaddset(&unblocked_sigs, SIGFPE);
458 #if defined(PPC64)
459 sigaddset(&unblocked_sigs, SIGTRAP);
460 #endif
461 sigaddset(&unblocked_sigs, SR_signum);
462
463 if (!ReduceSignalUsage) {
464 if (!os::Linux::is_sig_ignored(SHUTDOWN1_SIGNAL)) {
465 sigaddset(&unblocked_sigs, SHUTDOWN1_SIGNAL);
466 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN1_SIGNAL);
467 }
468 if (!os::Linux::is_sig_ignored(SHUTDOWN2_SIGNAL)) {
469 sigaddset(&unblocked_sigs, SHUTDOWN2_SIGNAL);
470 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN2_SIGNAL);
471 }
472 if (!os::Linux::is_sig_ignored(SHUTDOWN3_SIGNAL)) {
473 sigaddset(&unblocked_sigs, SHUTDOWN3_SIGNAL);
474 sigaddset(&allowdebug_blocked_sigs, SHUTDOWN3_SIGNAL);
475 }
476 }
477 // Fill in signals that are blocked by all but the VM thread.
478 sigemptyset(&vm_sigs);
479 if (!ReduceSignalUsage) {
480 sigaddset(&vm_sigs, BREAK_SIGNAL);
481 }
482 debug_only(signal_sets_initialized = true);
483
484 }
485
486 // These are signals that are unblocked while a thread is running Java.
487 // (For some reason, they get blocked by default.)
488 sigset_t* os::Linux::unblocked_signals() {
489 assert(signal_sets_initialized, "Not initialized");
490 return &unblocked_sigs;
491 }
492
493 // These are the signals that are blocked while a (non-VM) thread is
494 // running Java. Only the VM thread handles these signals.
495 sigset_t* os::Linux::vm_signals() {
496 assert(signal_sets_initialized, "Not initialized");
497 return &vm_sigs;
498 }
499
500 // These are signals that are blocked during cond_wait to allow debugger in
501 sigset_t* os::Linux::allowdebug_blocked_signals() {
502 assert(signal_sets_initialized, "Not initialized");
503 return &allowdebug_blocked_sigs;
504 }
505
506 void os::Linux::hotspot_sigmask(Thread* thread) {
507
508 //Save caller's signal mask before setting VM signal mask
509 sigset_t caller_sigmask;
510 pthread_sigmask(SIG_BLOCK, NULL, &caller_sigmask);
511
512 OSThread* osthread = thread->osthread();
513 osthread->set_caller_sigmask(caller_sigmask);
514
515 pthread_sigmask(SIG_UNBLOCK, os::Linux::unblocked_signals(), NULL);
516
517 if (!ReduceSignalUsage) {
518 if (thread->is_VM_thread()) {
519 // Only the VM thread handles BREAK_SIGNAL ...
520 pthread_sigmask(SIG_UNBLOCK, vm_signals(), NULL);
521 } else {
522 // ... all other threads block BREAK_SIGNAL
523 pthread_sigmask(SIG_BLOCK, vm_signals(), NULL);
524 }
525 }
526 }
527
528 //////////////////////////////////////////////////////////////////////////////
529 // detecting pthread library
530
531 void os::Linux::libpthread_init() {
532 // Save glibc and pthread version strings. Note that _CS_GNU_LIBC_VERSION
533 // and _CS_GNU_LIBPTHREAD_VERSION are supported in glibc >= 2.3.2. Use a
534 // generic name for earlier versions.
535 // Define macros here so we can build HotSpot on old systems.
536 #ifndef _CS_GNU_LIBC_VERSION
537 #define _CS_GNU_LIBC_VERSION 2
538 #endif
539 #ifndef _CS_GNU_LIBPTHREAD_VERSION
540 #define _CS_GNU_LIBPTHREAD_VERSION 3
541 #endif
542
543 size_t n = confstr(_CS_GNU_LIBC_VERSION, NULL, 0);
544 if (n > 0) {
545 char *str = (char *)malloc(n, mtInternal);
546 confstr(_CS_GNU_LIBC_VERSION, str, n);
547 os::Linux::set_glibc_version(str);
548 } else {
549 // _CS_GNU_LIBC_VERSION is not supported, try gnu_get_libc_version()
550 static char _gnu_libc_version[32];
551 jio_snprintf(_gnu_libc_version, sizeof(_gnu_libc_version),
552 "glibc %s %s", gnu_get_libc_version(), gnu_get_libc_release());
553 os::Linux::set_glibc_version(_gnu_libc_version);
554 }
555
556 n = confstr(_CS_GNU_LIBPTHREAD_VERSION, NULL, 0);
557 if (n > 0) {
558 char *str = (char *)malloc(n, mtInternal);
559 confstr(_CS_GNU_LIBPTHREAD_VERSION, str, n);
560 // Vanilla RH-9 (glibc 2.3.2) has a bug that confstr() always tells
561 // us "NPTL-0.29" even we are running with LinuxThreads. Check if this
562 // is the case. LinuxThreads has a hard limit on max number of threads.
563 // So sysconf(_SC_THREAD_THREADS_MAX) will return a positive value.
564 // On the other hand, NPTL does not have such a limit, sysconf()
565 // will return -1 and errno is not changed. Check if it is really NPTL.
566 if (strcmp(os::Linux::glibc_version(), "glibc 2.3.2") == 0 &&
567 strstr(str, "NPTL") &&
568 sysconf(_SC_THREAD_THREADS_MAX) > 0) {
569 free(str);
570 os::Linux::set_libpthread_version("linuxthreads");
571 } else {
572 os::Linux::set_libpthread_version(str);
573 }
574 } else {
575 // glibc before 2.3.2 only has LinuxThreads.
576 os::Linux::set_libpthread_version("linuxthreads");
577 }
578
579 if (strstr(libpthread_version(), "NPTL")) {
580 os::Linux::set_is_NPTL();
581 } else {
582 os::Linux::set_is_LinuxThreads();
583 }
584
585 // LinuxThreads have two flavors: floating-stack mode, which allows variable
586 // stack size; and fixed-stack mode. NPTL is always floating-stack.
587 if (os::Linux::is_NPTL() || os::Linux::supports_variable_stack_size()) {
588 os::Linux::set_is_floating_stack();
589 }
590 }
591
592 /////////////////////////////////////////////////////////////////////////////
593 // thread stack
594
595 // Force Linux kernel to expand current thread stack. If "bottom" is close
596 // to the stack guard, caller should block all signals.
597 //
598 // MAP_GROWSDOWN:
599 // A special mmap() flag that is used to implement thread stacks. It tells
600 // kernel that the memory region should extend downwards when needed. This
601 // allows early versions of LinuxThreads to only mmap the first few pages
602 // when creating a new thread. Linux kernel will automatically expand thread
603 // stack as needed (on page faults).
604 //
605 // However, because the memory region of a MAP_GROWSDOWN stack can grow on
606 // demand, if a page fault happens outside an already mapped MAP_GROWSDOWN
607 // region, it's hard to tell if the fault is due to a legitimate stack
608 // access or because of reading/writing non-exist memory (e.g. buffer
609 // overrun). As a rule, if the fault happens below current stack pointer,
610 // Linux kernel does not expand stack, instead a SIGSEGV is sent to the
611 // application (see Linux kernel fault.c).
612 //
613 // This Linux feature can cause SIGSEGV when VM bangs thread stack for
614 // stack overflow detection.
615 //
616 // Newer version of LinuxThreads (since glibc-2.2, or, RH-7.x) and NPTL do
617 // not use this flag. However, the stack of initial thread is not created
618 // by pthread, it is still MAP_GROWSDOWN. Also it's possible (though
619 // unlikely) that user code can create a thread with MAP_GROWSDOWN stack
620 // and then attach the thread to JVM.
621 //
622 // To get around the problem and allow stack banging on Linux, we need to
623 // manually expand thread stack after receiving the SIGSEGV.
624 //
625 // There are two ways to expand thread stack to address "bottom", we used
626 // both of them in JVM before 1.5:
627 // 1. adjust stack pointer first so that it is below "bottom", and then
628 // touch "bottom"
629 // 2. mmap() the page in question
630 //
631 // Now alternate signal stack is gone, it's harder to use 2. For instance,
632 // if current sp is already near the lower end of page 101, and we need to
633 // call mmap() to map page 100, it is possible that part of the mmap() frame
634 // will be placed in page 100. When page 100 is mapped, it is zero-filled.
635 // That will destroy the mmap() frame and cause VM to crash.
636 //
637 // The following code works by adjusting sp first, then accessing the "bottom"
638 // page to force a page fault. Linux kernel will then automatically expand the
639 // stack mapping.
640 //
641 // _expand_stack_to() assumes its frame size is less than page size, which
642 // should always be true if the function is not inlined.
643
644 #if __GNUC__ < 3 // gcc 2.x does not support noinline attribute
645 #define NOINLINE
646 #else
647 #define NOINLINE __attribute__ ((noinline))
648 #endif
649
650 static void _expand_stack_to(address bottom) NOINLINE;
651
652 static void _expand_stack_to(address bottom) {
653 address sp;
654 size_t size;
655 volatile char *p;
656
657 // Adjust bottom to point to the largest address within the same page, it
658 // gives us a one-page buffer if alloca() allocates slightly more memory.
659 bottom = (address)align_size_down((uintptr_t)bottom, os::Linux::page_size());
660 bottom += os::Linux::page_size() - 1;
661
662 // sp might be slightly above current stack pointer; if that's the case, we
663 // will alloca() a little more space than necessary, which is OK. Don't use
664 // os::current_stack_pointer(), as its result can be slightly below current
665 // stack pointer, causing us to not alloca enough to reach "bottom".
666 sp = (address)&sp;
667
668 if (sp > bottom) {
669 size = sp - bottom;
670 p = (volatile char *)alloca(size);
671 assert(p != NULL && p <= (volatile char *)bottom, "alloca problem?");
672 p[0] = '\0';
673 }
674 }
675
676 bool os::Linux::manually_expand_stack(JavaThread * t, address addr) {
677 assert(t!=NULL, "just checking");
678 assert(t->osthread()->expanding_stack(), "expand should be set");
679 assert(t->stack_base() != NULL, "stack_base was not initialized");
680
681 if (addr < t->stack_base() && addr >= t->stack_yellow_zone_base()) {
682 sigset_t mask_all, old_sigset;
683 sigfillset(&mask_all);
684 pthread_sigmask(SIG_SETMASK, &mask_all, &old_sigset);
685 _expand_stack_to(addr);
686 pthread_sigmask(SIG_SETMASK, &old_sigset, NULL);
687 return true;
688 }
689 return false;
690 }
691
692 //////////////////////////////////////////////////////////////////////////////
693 // create new thread
694
695 static address highest_vm_reserved_address();
696
697 // check if it's safe to start a new thread
698 static bool _thread_safety_check(Thread* thread) {
699 if (os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack()) {
700 // Fixed stack LinuxThreads (SuSE Linux/x86, and some versions of Redhat)
701 // Heap is mmap'ed at lower end of memory space. Thread stacks are
702 // allocated (MAP_FIXED) from high address space. Every thread stack
703 // occupies a fixed size slot (usually 2Mbytes, but user can change
704 // it to other values if they rebuild LinuxThreads).
705 //
706 // Problem with MAP_FIXED is that mmap() can still succeed even part of
707 // the memory region has already been mmap'ed. That means if we have too
708 // many threads and/or very large heap, eventually thread stack will
709 // collide with heap.
710 //
711 // Here we try to prevent heap/stack collision by comparing current
712 // stack bottom with the highest address that has been mmap'ed by JVM
713 // plus a safety margin for memory maps created by native code.
714 //
715 // This feature can be disabled by setting ThreadSafetyMargin to 0
716 //
717 if (ThreadSafetyMargin > 0) {
718 address stack_bottom = os::current_stack_base() - os::current_stack_size();
719
720 // not safe if our stack extends below the safety margin
721 return stack_bottom - ThreadSafetyMargin >= highest_vm_reserved_address();
722 } else {
723 return true;
724 }
725 } else {
726 // Floating stack LinuxThreads or NPTL:
727 // Unlike fixed stack LinuxThreads, thread stacks are not MAP_FIXED. When
728 // there's not enough space left, pthread_create() will fail. If we come
729 // here, that means enough space has been reserved for stack.
730 return true;
731 }
732 }
733
734 // Thread start routine for all newly created threads
735 static void *java_start(Thread *thread) {
736 // Try to randomize the cache line index of hot stack frames.
737 // This helps when threads of the same stack traces evict each other's
738 // cache lines. The threads can be either from the same JVM instance, or
739 // from different JVM instances. The benefit is especially true for
740 // processors with hyperthreading technology.
741 static int counter = 0;
742 int pid = os::current_process_id();
743 alloca(((pid ^ counter++) & 7) * 128);
744
745 ThreadLocalStorage::set_thread(thread);
746
747 OSThread* osthread = thread->osthread();
748 Monitor* sync = osthread->startThread_lock();
749
750 // non floating stack LinuxThreads needs extra check, see above
751 if (!_thread_safety_check(thread)) {
752 // notify parent thread
753 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
754 osthread->set_state(ZOMBIE);
755 sync->notify_all();
756 return NULL;
757 }
758
759 // thread_id is kernel thread id (similar to Solaris LWP id)
760 osthread->set_thread_id(os::Linux::gettid());
761
762 if (UseNUMA) {
763 int lgrp_id = os::numa_get_group_id();
764 if (lgrp_id != -1) {
765 thread->set_lgrp_id(lgrp_id);
766 }
767 }
768 // initialize signal mask for this thread
769 os::Linux::hotspot_sigmask(thread);
770
771 // initialize floating point control register
772 os::Linux::init_thread_fpu_state();
773
774 // handshaking with parent thread
775 {
776 MutexLockerEx ml(sync, Mutex::_no_safepoint_check_flag);
777
778 // notify parent thread
779 osthread->set_state(INITIALIZED);
780 sync->notify_all();
781
782 // wait until os::start_thread()
783 while (osthread->get_state() == INITIALIZED) {
784 sync->wait(Mutex::_no_safepoint_check_flag);
785 }
786 }
787
788 // call one more level start routine
789 thread->run();
790
791 return 0;
792 }
793
794 bool os::create_thread(Thread* thread, ThreadType thr_type,
795 size_t stack_size) {
796 assert(thread->osthread() == NULL, "caller responsible");
797
798 // Allocate the OSThread object
799 OSThread* osthread = new OSThread(NULL, NULL);
800 if (osthread == NULL) {
801 return false;
802 }
803
804 // set the correct thread state
805 osthread->set_thread_type(thr_type);
806
807 // Initial state is ALLOCATED but not INITIALIZED
808 osthread->set_state(ALLOCATED);
809
810 thread->set_osthread(osthread);
811
812 // init thread attributes
813 pthread_attr_t attr;
814 pthread_attr_init(&attr);
815 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
816
817 // stack size
818 if (os::Linux::supports_variable_stack_size()) {
819 // calculate stack size if it's not specified by caller
820 if (stack_size == 0) {
821 stack_size = os::Linux::default_stack_size(thr_type);
822
823 switch (thr_type) {
824 case os::java_thread:
825 // Java threads use ThreadStackSize which default value can be
826 // changed with the flag -Xss
827 assert(JavaThread::stack_size_at_create() > 0, "this should be set");
828 stack_size = JavaThread::stack_size_at_create();
829 break;
830 case os::compiler_thread:
831 if (CompilerThreadStackSize > 0) {
832 stack_size = (size_t)(CompilerThreadStackSize * K);
833 break;
834 } // else fall through:
835 // use VMThreadStackSize if CompilerThreadStackSize is not defined
836 case os::vm_thread:
837 case os::pgc_thread:
838 case os::cgc_thread:
839 case os::watcher_thread:
840 if (VMThreadStackSize > 0) stack_size = (size_t)(VMThreadStackSize * K);
841 break;
842 }
843 }
844
845 stack_size = MAX2(stack_size, os::Linux::min_stack_allowed);
846 pthread_attr_setstacksize(&attr, stack_size);
847 } else {
848 // let pthread_create() pick the default value.
849 }
850
851 // glibc guard page
852 pthread_attr_setguardsize(&attr, os::Linux::default_guard_size(thr_type));
853
854 ThreadState state;
855
856 {
857 // Serialize thread creation if we are running with fixed stack LinuxThreads
858 bool lock = os::Linux::is_LinuxThreads() && !os::Linux::is_floating_stack();
859 if (lock) {
860 os::Linux::createThread_lock()->lock_without_safepoint_check();
861 }
862
863 pthread_t tid;
864 int ret = pthread_create(&tid, &attr, (void* (*)(void*)) java_start, thread);
865
866 pthread_attr_destroy(&attr);
867
868 if (ret != 0) {
869 if (PrintMiscellaneous && (Verbose || WizardMode)) {
870 perror("pthread_create()");
871 }
872 // Need to clean up stuff we've allocated so far
873 thread->set_osthread(NULL);
874 delete osthread;
875 if (lock) os::Linux::createThread_lock()->unlock();
876 return false;
877 }
878
879 // Store pthread info into the OSThread
880 osthread->set_pthread_id(tid);
881
882 // Wait until child thread is either initialized or aborted
883 {
884 Monitor* sync_with_child = osthread->startThread_lock();
885 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
886 while ((state = osthread->get_state()) == ALLOCATED) {
887 sync_with_child->wait(Mutex::_no_safepoint_check_flag);
888 }
889 }
890
891 if (lock) {
892 os::Linux::createThread_lock()->unlock();
893 }
894 }
895
896 // Aborted due to thread limit being reached
897 if (state == ZOMBIE) {
898 thread->set_osthread(NULL);
899 delete osthread;
900 return false;
901 }
902
903 // The thread is returned suspended (in state INITIALIZED),
904 // and is started higher up in the call chain
905 assert(state == INITIALIZED, "race condition");
906 return true;
907 }
908
909 /////////////////////////////////////////////////////////////////////////////
910 // attach existing thread
911
912 // bootstrap the main thread
913 bool os::create_main_thread(JavaThread* thread) {
914 assert(os::Linux::_main_thread == pthread_self(), "should be called inside main thread");
915 return create_attached_thread(thread);
916 }
917
918 bool os::create_attached_thread(JavaThread* thread) {
919 #ifdef ASSERT
920 thread->verify_not_published();
921 #endif
922
923 // Allocate the OSThread object
924 OSThread* osthread = new OSThread(NULL, NULL);
925
926 if (osthread == NULL) {
927 return false;
928 }
929
930 // Store pthread info into the OSThread
931 osthread->set_thread_id(os::Linux::gettid());
932 osthread->set_pthread_id(::pthread_self());
933
934 // initialize floating point control register
935 os::Linux::init_thread_fpu_state();
936
937 // Initial thread state is RUNNABLE
938 osthread->set_state(RUNNABLE);
939
940 thread->set_osthread(osthread);
941
942 if (UseNUMA) {
943 int lgrp_id = os::numa_get_group_id();
944 if (lgrp_id != -1) {
945 thread->set_lgrp_id(lgrp_id);
946 }
947 }
948
949 if (os::Linux::is_initial_thread()) {
950 // If current thread is initial thread, its stack is mapped on demand,
951 // see notes about MAP_GROWSDOWN. Here we try to force kernel to map
952 // the entire stack region to avoid SEGV in stack banging.
953 // It is also useful to get around the heap-stack-gap problem on SuSE
954 // kernel (see 4821821 for details). We first expand stack to the top
955 // of yellow zone, then enable stack yellow zone (order is significant,
956 // enabling yellow zone first will crash JVM on SuSE Linux), so there
957 // is no gap between the last two virtual memory regions.
958
959 JavaThread *jt = (JavaThread *)thread;
960 address addr = jt->stack_yellow_zone_base();
961 assert(addr != NULL, "initialization problem?");
962 assert(jt->stack_available(addr) > 0, "stack guard should not be enabled");
963
964 osthread->set_expanding_stack();
965 os::Linux::manually_expand_stack(jt, addr);
966 osthread->clear_expanding_stack();
967 }
968
969 // initialize signal mask for this thread
970 // and save the caller's signal mask
971 os::Linux::hotspot_sigmask(thread);
972
973 return true;
974 }
975
976 void os::pd_start_thread(Thread* thread) {
977 OSThread * osthread = thread->osthread();
978 assert(osthread->get_state() != INITIALIZED, "just checking");
979 Monitor* sync_with_child = osthread->startThread_lock();
980 MutexLockerEx ml(sync_with_child, Mutex::_no_safepoint_check_flag);
981 sync_with_child->notify();
982 }
983
984 // Free Linux resources related to the OSThread
985 void os::free_thread(OSThread* osthread) {
986 assert(osthread != NULL, "osthread not set");
987
988 if (Thread::current()->osthread() == osthread) {
989 // Restore caller's signal mask
990 sigset_t sigmask = osthread->caller_sigmask();
991 pthread_sigmask(SIG_SETMASK, &sigmask, NULL);
992 }
993
994 delete osthread;
995 }
996
997 //////////////////////////////////////////////////////////////////////////////
998 // thread local storage
999
1000 // Restore the thread pointer if the destructor is called. This is in case
1001 // someone from JNI code sets up a destructor with pthread_key_create to run
1002 // detachCurrentThread on thread death. Unless we restore the thread pointer we
1003 // will hang or crash. When detachCurrentThread is called the key will be set
1004 // to null and we will not be called again. If detachCurrentThread is never
1005 // called we could loop forever depending on the pthread implementation.
1006 static void restore_thread_pointer(void* p) {
1007 Thread* thread = (Thread*) p;
1008 os::thread_local_storage_at_put(ThreadLocalStorage::thread_index(), thread);
1009 }
1010
1011 int os::allocate_thread_local_storage() {
1012 pthread_key_t key;
1013 int rslt = pthread_key_create(&key, restore_thread_pointer);
1014 assert(rslt == 0, "cannot allocate thread local storage");
1015 return (int)key;
1016 }
1017
1018 // Note: This is currently not used by VM, as we don't destroy TLS key
1019 // on VM exit.
1020 void os::free_thread_local_storage(int index) {
1021 int rslt = pthread_key_delete((pthread_key_t)index);
1022 assert(rslt == 0, "invalid index");
1023 }
1024
1025 void os::thread_local_storage_at_put(int index, void* value) {
1026 int rslt = pthread_setspecific((pthread_key_t)index, value);
1027 assert(rslt == 0, "pthread_setspecific failed");
1028 }
1029
1030 extern "C" Thread* get_thread() {
1031 return ThreadLocalStorage::thread();
1032 }
1033
1034 //////////////////////////////////////////////////////////////////////////////
1035 // initial thread
1036
1037 // Check if current thread is the initial thread, similar to Solaris thr_main.
1038 bool os::Linux::is_initial_thread(void) {
1039 char dummy;
1040 // If called before init complete, thread stack bottom will be null.
1041 // Can be called if fatal error occurs before initialization.
1042 if (initial_thread_stack_bottom() == NULL) return false;
1043 assert(initial_thread_stack_bottom() != NULL &&
1044 initial_thread_stack_size() != 0,
1045 "os::init did not locate initial thread's stack region");
1046 if ((address)&dummy >= initial_thread_stack_bottom() &&
1047 (address)&dummy < initial_thread_stack_bottom() + initial_thread_stack_size()) {
1048 return true;
1049 } else {
1050 return false;
1051 }
1052 }
1053
1054 // Find the virtual memory area that contains addr
1055 static bool find_vma(address addr, address* vma_low, address* vma_high) {
1056 FILE *fp = fopen("/proc/self/maps", "r");
1057 if (fp) {
1058 address low, high;
1059 while (!feof(fp)) {
1060 if (fscanf(fp, "%p-%p", &low, &high) == 2) {
1061 if (low <= addr && addr < high) {
1062 if (vma_low) *vma_low = low;
1063 if (vma_high) *vma_high = high;
1064 fclose(fp);
1065 return true;
1066 }
1067 }
1068 for (;;) {
1069 int ch = fgetc(fp);
1070 if (ch == EOF || ch == (int)'\n') break;
1071 }
1072 }
1073 fclose(fp);
1074 }
1075 return false;
1076 }
1077
1078 // Locate initial thread stack. This special handling of initial thread stack
1079 // is needed because pthread_getattr_np() on most (all?) Linux distros returns
1080 // bogus value for initial thread.
1081 void os::Linux::capture_initial_stack(size_t max_size) {
1082 // stack size is the easy part, get it from RLIMIT_STACK
1083 size_t stack_size;
1084 struct rlimit rlim;
1085 getrlimit(RLIMIT_STACK, &rlim);
1086 stack_size = rlim.rlim_cur;
1087
1088 // 6308388: a bug in ld.so will relocate its own .data section to the
1089 // lower end of primordial stack; reduce ulimit -s value a little bit
1090 // so we won't install guard page on ld.so's data section.
1091 stack_size -= 2 * page_size();
1092
1093 // 4441425: avoid crash with "unlimited" stack size on SuSE 7.1 or Redhat
1094 // 7.1, in both cases we will get 2G in return value.
1095 // 4466587: glibc 2.2.x compiled w/o "--enable-kernel=2.4.0" (RH 7.0,
1096 // SuSE 7.2, Debian) can not handle alternate signal stack correctly
1097 // for initial thread if its stack size exceeds 6M. Cap it at 2M,
1098 // in case other parts in glibc still assumes 2M max stack size.
1099 // FIXME: alt signal stack is gone, maybe we can relax this constraint?
1100 // Problem still exists RH7.2 (IA64 anyway) but 2MB is a little small
1101 if (stack_size > 2 * K * K IA64_ONLY(*2)) {
1102 stack_size = 2 * K * K IA64_ONLY(*2);
1103 }
1104 // Try to figure out where the stack base (top) is. This is harder.
1105 //
1106 // When an application is started, glibc saves the initial stack pointer in
1107 // a global variable "__libc_stack_end", which is then used by system
1108 // libraries. __libc_stack_end should be pretty close to stack top. The
1109 // variable is available since the very early days. However, because it is
1110 // a private interface, it could disappear in the future.
1111 //
1112 // Linux kernel saves start_stack information in /proc/<pid>/stat. Similar
1113 // to __libc_stack_end, it is very close to stack top, but isn't the real
1114 // stack top. Note that /proc may not exist if VM is running as a chroot
1115 // program, so reading /proc/<pid>/stat could fail. Also the contents of
1116 // /proc/<pid>/stat could change in the future (though unlikely).
1117 //
1118 // We try __libc_stack_end first. If that doesn't work, look for
1119 // /proc/<pid>/stat. If neither of them works, we use current stack pointer
1120 // as a hint, which should work well in most cases.
1121
1122 uintptr_t stack_start;
1123
1124 // try __libc_stack_end first
1125 uintptr_t *p = (uintptr_t *)dlsym(RTLD_DEFAULT, "__libc_stack_end");
1126 if (p && *p) {
1127 stack_start = *p;
1128 } else {
1129 // see if we can get the start_stack field from /proc/self/stat
1130 FILE *fp;
1131 int pid;
1132 char state;
1133 int ppid;
1134 int pgrp;
1135 int session;
1136 int nr;
1137 int tpgrp;
1138 unsigned long flags;
1139 unsigned long minflt;
1140 unsigned long cminflt;
1141 unsigned long majflt;
1142 unsigned long cmajflt;
1143 unsigned long utime;
1144 unsigned long stime;
1145 long cutime;
1146 long cstime;
1147 long prio;
1148 long nice;
1149 long junk;
1150 long it_real;
1151 uintptr_t start;
1152 uintptr_t vsize;
1153 intptr_t rss;
1154 uintptr_t rsslim;
1155 uintptr_t scodes;
1156 uintptr_t ecode;
1157 int i;
1158
1159 // Figure what the primordial thread stack base is. Code is inspired
1160 // by email from Hans Boehm. /proc/self/stat begins with current pid,
1161 // followed by command name surrounded by parentheses, state, etc.
1162 char stat[2048];
1163 int statlen;
1164
1165 fp = fopen("/proc/self/stat", "r");
1166 if (fp) {
1167 statlen = fread(stat, 1, 2047, fp);
1168 stat[statlen] = '\0';
1169 fclose(fp);
1170
1171 // Skip pid and the command string. Note that we could be dealing with
1172 // weird command names, e.g. user could decide to rename java launcher
1173 // to "java 1.4.2 :)", then the stat file would look like
1174 // 1234 (java 1.4.2 :)) R ... ...
1175 // We don't really need to know the command string, just find the last
1176 // occurrence of ")" and then start parsing from there. See bug 4726580.
1177 char * s = strrchr(stat, ')');
1178
1179 i = 0;
1180 if (s) {
1181 // Skip blank chars
1182 do { s++; } while (s && isspace(*s));
1183
1184 #define _UFM UINTX_FORMAT
1185 #define _DFM INTX_FORMAT
1186
1187 // 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2
1188 // 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
1189 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,
1190 &state, // 3 %c
1191 &ppid, // 4 %d
1192 &pgrp, // 5 %d
1193 &session, // 6 %d
1194 &nr, // 7 %d
1195 &tpgrp, // 8 %d
1196 &flags, // 9 %lu
1197 &minflt, // 10 %lu
1198 &cminflt, // 11 %lu
1199 &majflt, // 12 %lu
1200 &cmajflt, // 13 %lu
1201 &utime, // 14 %lu
1202 &stime, // 15 %lu
1203 &cutime, // 16 %ld
1204 &cstime, // 17 %ld
1205 &prio, // 18 %ld
1206 &nice, // 19 %ld
1207 &junk, // 20 %ld
1208 &it_real, // 21 %ld
1209 &start, // 22 UINTX_FORMAT
1210 &vsize, // 23 UINTX_FORMAT
1211 &rss, // 24 INTX_FORMAT
1212 &rsslim, // 25 UINTX_FORMAT
1213 &scodes, // 26 UINTX_FORMAT
1214 &ecode, // 27 UINTX_FORMAT
1215 &stack_start); // 28 UINTX_FORMAT
1216 }
1217
1218 #undef _UFM
1219 #undef _DFM
1220
1221 if (i != 28 - 2) {
1222 assert(false, "Bad conversion from /proc/self/stat");
1223 // product mode - assume we are the initial thread, good luck in the
1224 // embedded case.
1225 warning("Can't detect initial thread stack location - bad conversion");
1226 stack_start = (uintptr_t) &rlim;
1227 }
1228 } else {
1229 // For some reason we can't open /proc/self/stat (for example, running on
1230 // FreeBSD with a Linux emulator, or inside chroot), this should work for
1231 // most cases, so don't abort:
1232 warning("Can't detect initial thread stack location - no /proc/self/stat");
1233 stack_start = (uintptr_t) &rlim;
1234 }
1235 }
1236
1237 // Now we have a pointer (stack_start) very close to the stack top, the
1238 // next thing to do is to figure out the exact location of stack top. We
1239 // can find out the virtual memory area that contains stack_start by
1240 // reading /proc/self/maps, it should be the last vma in /proc/self/maps,
1241 // and its upper limit is the real stack top. (again, this would fail if
1242 // running inside chroot, because /proc may not exist.)
1243
1244 uintptr_t stack_top;
1245 address low, high;
1246 if (find_vma((address)stack_start, &low, &high)) {
1247 // success, "high" is the true stack top. (ignore "low", because initial
1248 // thread stack grows on demand, its real bottom is high - RLIMIT_STACK.)
1249 stack_top = (uintptr_t)high;
1250 } else {
1251 // failed, likely because /proc/self/maps does not exist
1252 warning("Can't detect initial thread stack location - find_vma failed");
1253 // best effort: stack_start is normally within a few pages below the real
1254 // stack top, use it as stack top, and reduce stack size so we won't put
1255 // guard page outside stack.
1256 stack_top = stack_start;
1257 stack_size -= 16 * page_size();
1258 }
1259
1260 // stack_top could be partially down the page so align it
1261 stack_top = align_size_up(stack_top, page_size());
1262
1263 if (max_size && stack_size > max_size) {
1264 _initial_thread_stack_size = max_size;
1265 } else {
1266 _initial_thread_stack_size = stack_size;
1267 }
1268
1269 _initial_thread_stack_size = align_size_down(_initial_thread_stack_size, page_size());
1270 _initial_thread_stack_bottom = (address)stack_top - _initial_thread_stack_size;
1271 }
1272
1273 ////////////////////////////////////////////////////////////////////////////////
1274 // time support
1275
1276 // Time since start-up in seconds to a fine granularity.
1277 // Used by VMSelfDestructTimer and the MemProfiler.
1278 double os::elapsedTime() {
1279
1280 return ((double)os::elapsed_counter()) / os::elapsed_frequency(); // nanosecond resolution
1281 }
1282
1283 jlong os::elapsed_counter() {
1284 return javaTimeNanos() - initial_time_count;
1285 }
1286
1287 jlong os::elapsed_frequency() {
1288 return NANOSECS_PER_SEC; // nanosecond resolution
1289 }
1290
1291 bool os::supports_vtime() { return true; }
1292 bool os::enable_vtime() { return false; }
1293 bool os::vtime_enabled() { return false; }
1294
1295 double os::elapsedVTime() {
1296 struct rusage usage;
1297 int retval = getrusage(RUSAGE_THREAD, &usage);
1298 if (retval == 0) {
1299 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);
1300 } else {
1301 // better than nothing, but not much
1302 return elapsedTime();
1303 }
1304 }
1305
1306 jlong os::javaTimeMillis() {
1307 timeval time;
1308 int status = gettimeofday(&time, NULL);
1309 assert(status != -1, "linux error");
1310 return jlong(time.tv_sec) * 1000 + jlong(time.tv_usec / 1000);
1311 }
1312
1313 void os::javaTimeSystemUTC(jlong &seconds, jlong &nanos) {
1314 timeval time;
1315 int status = gettimeofday(&time, NULL);
1316 assert(status != -1, "linux error");
1317 seconds = jlong(time.tv_sec);
1318 nanos = jlong(time.tv_usec) * 1000;
1319 }
1320
1321
1322 #ifndef CLOCK_MONOTONIC
1323 #define CLOCK_MONOTONIC (1)
1324 #endif
1325
1326 void os::Linux::clock_init() {
1327 // we do dlopen's in this particular order due to bug in linux
1328 // dynamical loader (see 6348968) leading to crash on exit
1329 void* handle = dlopen("librt.so.1", RTLD_LAZY);
1330 if (handle == NULL) {
1331 handle = dlopen("librt.so", RTLD_LAZY);
1332 }
1333
1334 if (handle) {
1335 int (*clock_getres_func)(clockid_t, struct timespec*) =
1336 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_getres");
1337 int (*clock_gettime_func)(clockid_t, struct timespec*) =
1338 (int(*)(clockid_t, struct timespec*))dlsym(handle, "clock_gettime");
1339 if (clock_getres_func && clock_gettime_func) {
1340 // See if monotonic clock is supported by the kernel. Note that some
1341 // early implementations simply return kernel jiffies (updated every
1342 // 1/100 or 1/1000 second). It would be bad to use such a low res clock
1343 // for nano time (though the monotonic property is still nice to have).
1344 // It's fixed in newer kernels, however clock_getres() still returns
1345 // 1/HZ. We check if clock_getres() works, but will ignore its reported
1346 // resolution for now. Hopefully as people move to new kernels, this
1347 // won't be a problem.
1348 struct timespec res;
1349 struct timespec tp;
1350 if (clock_getres_func (CLOCK_MONOTONIC, &res) == 0 &&
1351 clock_gettime_func(CLOCK_MONOTONIC, &tp) == 0) {
1352 // yes, monotonic clock is supported
1353 _clock_gettime = clock_gettime_func;
1354 return;
1355 } else {
1356 // close librt if there is no monotonic clock
1357 dlclose(handle);
1358 }
1359 }
1360 }
1361 warning("No monotonic clock was available - timed services may " \
1362 "be adversely affected if the time-of-day clock changes");
1363 }
1364
1365 #ifndef SYS_clock_getres
1366 #if defined(IA32) || defined(AMD64)
1367 #define SYS_clock_getres IA32_ONLY(266) AMD64_ONLY(229)
1368 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1369 #else
1370 #warning "SYS_clock_getres not defined for this platform, disabling fast_thread_cpu_time"
1371 #define sys_clock_getres(x,y) -1
1372 #endif
1373 #else
1374 #define sys_clock_getres(x,y) ::syscall(SYS_clock_getres, x, y)
1375 #endif
1376
1377 void os::Linux::fast_thread_clock_init() {
1378 if (!UseLinuxPosixThreadCPUClocks) {
1379 return;
1380 }
1381 clockid_t clockid;
1382 struct timespec tp;
1383 int (*pthread_getcpuclockid_func)(pthread_t, clockid_t *) =
1384 (int(*)(pthread_t, clockid_t *)) dlsym(RTLD_DEFAULT, "pthread_getcpuclockid");
1385
1386 // Switch to using fast clocks for thread cpu time if
1387 // the sys_clock_getres() returns 0 error code.
1388 // Note, that some kernels may support the current thread
1389 // clock (CLOCK_THREAD_CPUTIME_ID) but not the clocks
1390 // returned by the pthread_getcpuclockid().
1391 // If the fast Posix clocks are supported then the sys_clock_getres()
1392 // must return at least tp.tv_sec == 0 which means a resolution
1393 // better than 1 sec. This is extra check for reliability.
1394
1395 if (pthread_getcpuclockid_func &&
1396 pthread_getcpuclockid_func(_main_thread, &clockid) == 0 &&
1397 sys_clock_getres(clockid, &tp) == 0 && tp.tv_sec == 0) {
1398 _supports_fast_thread_cpu_time = true;
1399 _pthread_getcpuclockid = pthread_getcpuclockid_func;
1400 }
1401 }
1402
1403 jlong os::javaTimeNanos() {
1404 if (os::supports_monotonic_clock()) {
1405 struct timespec tp;
1406 int status = Linux::clock_gettime(CLOCK_MONOTONIC, &tp);
1407 assert(status == 0, "gettime error");
1408 jlong result = jlong(tp.tv_sec) * (1000 * 1000 * 1000) + jlong(tp.tv_nsec);
1409 return result;
1410 } else {
1411 timeval time;
1412 int status = gettimeofday(&time, NULL);
1413 assert(status != -1, "linux error");
1414 jlong usecs = jlong(time.tv_sec) * (1000 * 1000) + jlong(time.tv_usec);
1415 return 1000 * usecs;
1416 }
1417 }
1418
1419 void os::javaTimeNanos_info(jvmtiTimerInfo *info_ptr) {
1420 if (os::supports_monotonic_clock()) {
1421 info_ptr->max_value = ALL_64_BITS;
1422
1423 // CLOCK_MONOTONIC - amount of time since some arbitrary point in the past
1424 info_ptr->may_skip_backward = false; // not subject to resetting or drifting
1425 info_ptr->may_skip_forward = false; // not subject to resetting or drifting
1426 } else {
1427 // gettimeofday - based on time in seconds since the Epoch thus does not wrap
1428 info_ptr->max_value = ALL_64_BITS;
1429
1430 // gettimeofday is a real time clock so it skips
1431 info_ptr->may_skip_backward = true;
1432 info_ptr->may_skip_forward = true;
1433 }
1434
1435 info_ptr->kind = JVMTI_TIMER_ELAPSED; // elapsed not CPU time
1436 }
1437
1438 // Return the real, user, and system times in seconds from an
1439 // arbitrary fixed point in the past.
1440 bool os::getTimesSecs(double* process_real_time,
1441 double* process_user_time,
1442 double* process_system_time) {
1443 struct tms ticks;
1444 clock_t real_ticks = times(&ticks);
1445
1446 if (real_ticks == (clock_t) (-1)) {
1447 return false;
1448 } else {
1449 double ticks_per_second = (double) clock_tics_per_sec;
1450 *process_user_time = ((double) ticks.tms_utime) / ticks_per_second;
1451 *process_system_time = ((double) ticks.tms_stime) / ticks_per_second;
1452 *process_real_time = ((double) real_ticks) / ticks_per_second;
1453
1454 return true;
1455 }
1456 }
1457
1458
1459 char * os::local_time_string(char *buf, size_t buflen) {
1460 struct tm t;
1461 time_t long_time;
1462 time(&long_time);
1463 localtime_r(&long_time, &t);
1464 jio_snprintf(buf, buflen, "%d-%02d-%02d %02d:%02d:%02d",
1465 t.tm_year + 1900, t.tm_mon + 1, t.tm_mday,
1466 t.tm_hour, t.tm_min, t.tm_sec);
1467 return buf;
1468 }
1469
1470 struct tm* os::localtime_pd(const time_t* clock, struct tm* res) {
1471 return localtime_r(clock, res);
1472 }
1473
1474 ////////////////////////////////////////////////////////////////////////////////
1475 // runtime exit support
1476
1477 // Note: os::shutdown() might be called very early during initialization, or
1478 // called from signal handler. Before adding something to os::shutdown(), make
1479 // sure it is async-safe and can handle partially initialized VM.
1480 void os::shutdown() {
1481
1482 // allow PerfMemory to attempt cleanup of any persistent resources
1483 perfMemory_exit();
1484
1485 // needs to remove object in file system
1486 AttachListener::abort();
1487
1488 // flush buffered output, finish log files
1489 ostream_abort();
1490
1491 // Check for abort hook
1492 abort_hook_t abort_hook = Arguments::abort_hook();
1493 if (abort_hook != NULL) {
1494 abort_hook();
1495 }
1496
1497 }
1498
1499 // Note: os::abort() might be called very early during initialization, or
1500 // called from signal handler. Before adding something to os::abort(), make
1501 // sure it is async-safe and can handle partially initialized VM.
1502 void os::abort(bool dump_core) {
1503 os::shutdown();
1504 if (dump_core) {
1505 #ifndef PRODUCT
1506 fdStream out(defaultStream::output_fd());
1507 out.print_raw("Current thread is ");
1508 char buf[16];
1509 jio_snprintf(buf, sizeof(buf), UINTX_FORMAT, os::current_thread_id());
1510 out.print_raw_cr(buf);
1511 out.print_raw_cr("Dumping core ...");
1512 #endif
1513 ::abort(); // dump core
1514 }
1515
1516 ::exit(1);
1517 }
1518
1519 // Die immediately, no exit hook, no abort hook, no cleanup.
1520 void os::die() {
1521 // _exit() on LinuxThreads only kills current thread
1522 ::abort();
1523 }
1524
1525
1526 // This method is a copy of JDK's sysGetLastErrorString
1527 // from src/solaris/hpi/src/system_md.c
1528
1529 size_t os::lasterror(char *buf, size_t len) {
1530 if (errno == 0) return 0;
1531
1532 const char *s = ::strerror(errno);
1533 size_t n = ::strlen(s);
1534 if (n >= len) {
1535 n = len - 1;
1536 }
1537 ::strncpy(buf, s, n);
1538 buf[n] = '\0';
1539 return n;
1540 }
1541
1542 intx os::current_thread_id() { return (intx)pthread_self(); }
1543 int os::current_process_id() {
1544
1545 // Under the old linux thread library, linux gives each thread
1546 // its own process id. Because of this each thread will return
1547 // a different pid if this method were to return the result
1548 // of getpid(2). Linux provides no api that returns the pid
1549 // of the launcher thread for the vm. This implementation
1550 // returns a unique pid, the pid of the launcher thread
1551 // that starts the vm 'process'.
1552
1553 // Under the NPTL, getpid() returns the same pid as the
1554 // launcher thread rather than a unique pid per thread.
1555 // Use gettid() if you want the old pre NPTL behaviour.
1556
1557 // if you are looking for the result of a call to getpid() that
1558 // returns a unique pid for the calling thread, then look at the
1559 // OSThread::thread_id() method in osThread_linux.hpp file
1560
1561 return (int)(_initial_pid ? _initial_pid : getpid());
1562 }
1563
1564 // DLL functions
1565
1566 const char* os::dll_file_extension() { return ".so"; }
1567
1568 // This must be hard coded because it's the system's temporary
1569 // directory not the java application's temp directory, ala java.io.tmpdir.
1570 const char* os::get_temp_directory() { return "/tmp"; }
1571
1572 static bool file_exists(const char* filename) {
1573 struct stat statbuf;
1574 if (filename == NULL || strlen(filename) == 0) {
1575 return false;
1576 }
1577 return os::stat(filename, &statbuf) == 0;
1578 }
1579
1580 bool os::dll_build_name(char* buffer, size_t buflen,
1581 const char* pname, const char* fname) {
1582 bool retval = false;
1583 // Copied from libhpi
1584 const size_t pnamelen = pname ? strlen(pname) : 0;
1585
1586 // Return error on buffer overflow.
1587 if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
1588 return retval;
1589 }
1590
1591 if (pnamelen == 0) {
1592 snprintf(buffer, buflen, "lib%s.so", fname);
1593 retval = true;
1594 } else if (strchr(pname, *os::path_separator()) != NULL) {
1595 int n;
1596 char** pelements = split_path(pname, &n);
1597 if (pelements == NULL) {
1598 return false;
1599 }
1600 for (int i = 0; i < n; i++) {
1601 // Really shouldn't be NULL, but check can't hurt
1602 if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
1603 continue; // skip the empty path values
1604 }
1605 snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
1606 if (file_exists(buffer)) {
1607 retval = true;
1608 break;
1609 }
1610 }
1611 // release the storage
1612 for (int i = 0; i < n; i++) {
1613 if (pelements[i] != NULL) {
1614 FREE_C_HEAP_ARRAY(char, pelements[i]);
1615 }
1616 }
1617 if (pelements != NULL) {
1618 FREE_C_HEAP_ARRAY(char*, pelements);
1619 }
1620 } else {
1621 snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
1622 retval = true;
1623 }
1624 return retval;
1625 }
1626
1627 // check if addr is inside libjvm.so
1628 bool os::address_is_in_vm(address addr) {
1629 static address libjvm_base_addr;
1630 Dl_info dlinfo;
1631
1632 if (libjvm_base_addr == NULL) {
1633 if (dladdr(CAST_FROM_FN_PTR(void *, os::address_is_in_vm), &dlinfo) != 0) {
1634 libjvm_base_addr = (address)dlinfo.dli_fbase;
1635 }
1636 assert(libjvm_base_addr !=NULL, "Cannot obtain base address for libjvm");
1637 }
1638
1639 if (dladdr((void *)addr, &dlinfo) != 0) {
1640 if (libjvm_base_addr == (address)dlinfo.dli_fbase) return true;
1641 }
1642
1643 return false;
1644 }
1645
1646 bool os::dll_address_to_function_name(address addr, char *buf,
1647 int buflen, int *offset) {
1648 // buf is not optional, but offset is optional
1649 assert(buf != NULL, "sanity check");
1650
1651 Dl_info dlinfo;
1652
1653 if (dladdr((void*)addr, &dlinfo) != 0) {
1654 // see if we have a matching symbol
1655 if (dlinfo.dli_saddr != NULL && dlinfo.dli_sname != NULL) {
1656 if (!Decoder::demangle(dlinfo.dli_sname, buf, buflen)) {
1657 jio_snprintf(buf, buflen, "%s", dlinfo.dli_sname);
1658 }
1659 if (offset != NULL) *offset = addr - (address)dlinfo.dli_saddr;
1660 return true;
1661 }
1662 // no matching symbol so try for just file info
1663 if (dlinfo.dli_fname != NULL && dlinfo.dli_fbase != NULL) {
1664 if (Decoder::decode((address)(addr - (address)dlinfo.dli_fbase),
1665 buf, buflen, offset, dlinfo.dli_fname)) {
1666 return true;
1667 }
1668 }
1669 }
1670
1671 buf[0] = '\0';
1672 if (offset != NULL) *offset = -1;
1673 return false;
1674 }
1675
1676 struct _address_to_library_name {
1677 address addr; // input : memory address
1678 size_t buflen; // size of fname
1679 char* fname; // output: library name
1680 address base; // library base addr
1681 };
1682
1683 static int address_to_library_name_callback(struct dl_phdr_info *info,
1684 size_t size, void *data) {
1685 int i;
1686 bool found = false;
1687 address libbase = NULL;
1688 struct _address_to_library_name * d = (struct _address_to_library_name *)data;
1689
1690 // iterate through all loadable segments
1691 for (i = 0; i < info->dlpi_phnum; i++) {
1692 address segbase = (address)(info->dlpi_addr + info->dlpi_phdr[i].p_vaddr);
1693 if (info->dlpi_phdr[i].p_type == PT_LOAD) {
1694 // base address of a library is the lowest address of its loaded
1695 // segments.
1696 if (libbase == NULL || libbase > segbase) {
1697 libbase = segbase;
1698 }
1699 // see if 'addr' is within current segment
1700 if (segbase <= d->addr &&
1701 d->addr < segbase + info->dlpi_phdr[i].p_memsz) {
1702 found = true;
1703 }
1704 }
1705 }
1706
1707 // dlpi_name is NULL or empty if the ELF file is executable, return 0
1708 // so dll_address_to_library_name() can fall through to use dladdr() which
1709 // can figure out executable name from argv[0].
1710 if (found && info->dlpi_name && info->dlpi_name[0]) {
1711 d->base = libbase;
1712 if (d->fname) {
1713 jio_snprintf(d->fname, d->buflen, "%s", info->dlpi_name);
1714 }
1715 return 1;
1716 }
1717 return 0;
1718 }
1719
1720 bool os::dll_address_to_library_name(address addr, char* buf,
1721 int buflen, int* offset) {
1722 // buf is not optional, but offset is optional
1723 assert(buf != NULL, "sanity check");
1724
1725 Dl_info dlinfo;
1726 struct _address_to_library_name data;
1727
1728 // There is a bug in old glibc dladdr() implementation that it could resolve
1729 // to wrong library name if the .so file has a base address != NULL. Here
1730 // we iterate through the program headers of all loaded libraries to find
1731 // out which library 'addr' really belongs to. This workaround can be
1732 // removed once the minimum requirement for glibc is moved to 2.3.x.
1733 data.addr = addr;
1734 data.fname = buf;
1735 data.buflen = buflen;
1736 data.base = NULL;
1737 int rslt = dl_iterate_phdr(address_to_library_name_callback, (void *)&data);
1738
1739 if (rslt) {
1740 // buf already contains library name
1741 if (offset) *offset = addr - data.base;
1742 return true;
1743 }
1744 if (dladdr((void*)addr, &dlinfo) != 0) {
1745 if (dlinfo.dli_fname != NULL) {
1746 jio_snprintf(buf, buflen, "%s", dlinfo.dli_fname);
1747 }
1748 if (dlinfo.dli_fbase != NULL && offset != NULL) {
1749 *offset = addr - (address)dlinfo.dli_fbase;
1750 }
1751 return true;
1752 }
1753
1754 buf[0] = '\0';
1755 if (offset) *offset = -1;
1756 return false;
1757 }
1758
1759 // Loads .dll/.so and
1760 // in case of error it checks if .dll/.so was built for the
1761 // same architecture as Hotspot is running on
1762
1763
1764 // Remember the stack's state. The Linux dynamic linker will change
1765 // the stack to 'executable' at most once, so we must safepoint only once.
1766 bool os::Linux::_stack_is_executable = false;
1767
1768 // VM operation that loads a library. This is necessary if stack protection
1769 // of the Java stacks can be lost during loading the library. If we
1770 // do not stop the Java threads, they can stack overflow before the stacks
1771 // are protected again.
1772 class VM_LinuxDllLoad: public VM_Operation {
1773 private:
1774 const char *_filename;
1775 char *_ebuf;
1776 int _ebuflen;
1777 void *_lib;
1778 public:
1779 VM_LinuxDllLoad(const char *fn, char *ebuf, int ebuflen) :
1780 _filename(fn), _ebuf(ebuf), _ebuflen(ebuflen), _lib(NULL) {}
1781 VMOp_Type type() const { return VMOp_LinuxDllLoad; }
1782 void doit() {
1783 _lib = os::Linux::dll_load_in_vmthread(_filename, _ebuf, _ebuflen);
1784 os::Linux::_stack_is_executable = true;
1785 }
1786 void* loaded_library() { return _lib; }
1787 };
1788
1789 void * os::dll_load(const char *filename, char *ebuf, int ebuflen) {
1790 void * result = NULL;
1791 bool load_attempted = false;
1792
1793 // Check whether the library to load might change execution rights
1794 // of the stack. If they are changed, the protection of the stack
1795 // guard pages will be lost. We need a safepoint to fix this.
1796 //
1797 // See Linux man page execstack(8) for more info.
1798 if (os::uses_stack_guard_pages() && !os::Linux::_stack_is_executable) {
1799 ElfFile ef(filename);
1800 if (!ef.specifies_noexecstack()) {
1801 if (!is_init_completed()) {
1802 os::Linux::_stack_is_executable = true;
1803 // This is OK - No Java threads have been created yet, and hence no
1804 // stack guard pages to fix.
1805 //
1806 // This should happen only when you are building JDK7 using a very
1807 // old version of JDK6 (e.g., with JPRT) and running test_gamma.
1808 //
1809 // Dynamic loader will make all stacks executable after
1810 // this function returns, and will not do that again.
1811 assert(Threads::first() == NULL, "no Java threads should exist yet.");
1812 } else {
1813 warning("You have loaded library %s which might have disabled stack guard. "
1814 "The VM will try to fix the stack guard now.\n"
1815 "It's highly recommended that you fix the library with "
1816 "'execstack -c <libfile>', or link it with '-z noexecstack'.",
1817 filename);
1818
1819 assert(Thread::current()->is_Java_thread(), "must be Java thread");
1820 JavaThread *jt = JavaThread::current();
1821 if (jt->thread_state() != _thread_in_native) {
1822 // This happens when a compiler thread tries to load a hsdis-<arch>.so file
1823 // that requires ExecStack. Cannot enter safe point. Let's give up.
1824 warning("Unable to fix stack guard. Giving up.");
1825 } else {
1826 if (!LoadExecStackDllInVMThread) {
1827 // This is for the case where the DLL has an static
1828 // constructor function that executes JNI code. We cannot
1829 // load such DLLs in the VMThread.
1830 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1831 }
1832
1833 ThreadInVMfromNative tiv(jt);
1834 debug_only(VMNativeEntryWrapper vew;)
1835
1836 VM_LinuxDllLoad op(filename, ebuf, ebuflen);
1837 VMThread::execute(&op);
1838 if (LoadExecStackDllInVMThread) {
1839 result = op.loaded_library();
1840 }
1841 load_attempted = true;
1842 }
1843 }
1844 }
1845 }
1846
1847 if (!load_attempted) {
1848 result = os::Linux::dlopen_helper(filename, ebuf, ebuflen);
1849 }
1850
1851 if (result != NULL) {
1852 // Successful loading
1853 return result;
1854 }
1855
1856 Elf32_Ehdr elf_head;
1857 int diag_msg_max_length=ebuflen-strlen(ebuf);
1858 char* diag_msg_buf=ebuf+strlen(ebuf);
1859
1860 if (diag_msg_max_length==0) {
1861 // No more space in ebuf for additional diagnostics message
1862 return NULL;
1863 }
1864
1865
1866 int file_descriptor= ::open(filename, O_RDONLY | O_NONBLOCK);
1867
1868 if (file_descriptor < 0) {
1869 // Can't open library, report dlerror() message
1870 return NULL;
1871 }
1872
1873 bool failed_to_read_elf_head=
1874 (sizeof(elf_head)!=
1875 (::read(file_descriptor, &elf_head,sizeof(elf_head))));
1876
1877 ::close(file_descriptor);
1878 if (failed_to_read_elf_head) {
1879 // file i/o error - report dlerror() msg
1880 return NULL;
1881 }
1882
1883 typedef struct {
1884 Elf32_Half code; // Actual value as defined in elf.h
1885 Elf32_Half compat_class; // Compatibility of archs at VM's sense
1886 char elf_class; // 32 or 64 bit
1887 char endianess; // MSB or LSB
1888 char* name; // String representation
1889 } arch_t;
1890
1891 #ifndef EM_486
1892 #define EM_486 6 /* Intel 80486 */
1893 #endif
1894 #ifndef EM_AARCH64
1895 #define EM_AARCH64 183 /* ARM AARCH64 */
1896 #endif
1897
1898 static const arch_t arch_array[]={
1899 {EM_386, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1900 {EM_486, EM_386, ELFCLASS32, ELFDATA2LSB, (char*)"IA 32"},
1901 {EM_IA_64, EM_IA_64, ELFCLASS64, ELFDATA2LSB, (char*)"IA 64"},
1902 {EM_X86_64, EM_X86_64, ELFCLASS64, ELFDATA2LSB, (char*)"AMD 64"},
1903 {EM_SPARC, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1904 {EM_SPARC32PLUS, EM_SPARC, ELFCLASS32, ELFDATA2MSB, (char*)"Sparc 32"},
1905 {EM_SPARCV9, EM_SPARCV9, ELFCLASS64, ELFDATA2MSB, (char*)"Sparc v9 64"},
1906 {EM_PPC, EM_PPC, ELFCLASS32, ELFDATA2MSB, (char*)"Power PC 32"},
1907 #if defined(VM_LITTLE_ENDIAN)
1908 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2LSB, (char*)"Power PC 64"},
1909 #else
1910 {EM_PPC64, EM_PPC64, ELFCLASS64, ELFDATA2MSB, (char*)"Power PC 64"},
1911 #endif
1912 {EM_ARM, EM_ARM, ELFCLASS32, ELFDATA2LSB, (char*)"ARM"},
1913 {EM_S390, EM_S390, ELFCLASSNONE, ELFDATA2MSB, (char*)"IBM System/390"},
1914 {EM_ALPHA, EM_ALPHA, ELFCLASS64, ELFDATA2LSB, (char*)"Alpha"},
1915 {EM_MIPS_RS3_LE, EM_MIPS_RS3_LE, ELFCLASS32, ELFDATA2LSB, (char*)"MIPSel"},
1916 {EM_MIPS, EM_MIPS, ELFCLASS32, ELFDATA2MSB, (char*)"MIPS"},
1917 {EM_PARISC, EM_PARISC, ELFCLASS32, ELFDATA2MSB, (char*)"PARISC"},
1918 {EM_68K, EM_68K, ELFCLASS32, ELFDATA2MSB, (char*)"M68k"},
1919 {EM_AARCH64, EM_AARCH64, ELFCLASS64, ELFDATA2LSB, (char*)"AARCH64"},
1920 };
1921
1922 #if (defined IA32)
1923 static Elf32_Half running_arch_code=EM_386;
1924 #elif (defined AMD64)
1925 static Elf32_Half running_arch_code=EM_X86_64;
1926 #elif (defined IA64)
1927 static Elf32_Half running_arch_code=EM_IA_64;
1928 #elif (defined __sparc) && (defined _LP64)
1929 static Elf32_Half running_arch_code=EM_SPARCV9;
1930 #elif (defined __sparc) && (!defined _LP64)
1931 static Elf32_Half running_arch_code=EM_SPARC;
1932 #elif (defined __powerpc64__)
1933 static Elf32_Half running_arch_code=EM_PPC64;
1934 #elif (defined __powerpc__)
1935 static Elf32_Half running_arch_code=EM_PPC;
1936 #elif (defined ARM)
1937 static Elf32_Half running_arch_code=EM_ARM;
1938 #elif (defined S390)
1939 static Elf32_Half running_arch_code=EM_S390;
1940 #elif (defined ALPHA)
1941 static Elf32_Half running_arch_code=EM_ALPHA;
1942 #elif (defined MIPSEL)
1943 static Elf32_Half running_arch_code=EM_MIPS_RS3_LE;
1944 #elif (defined PARISC)
1945 static Elf32_Half running_arch_code=EM_PARISC;
1946 #elif (defined MIPS)
1947 static Elf32_Half running_arch_code=EM_MIPS;
1948 #elif (defined M68K)
1949 static Elf32_Half running_arch_code=EM_68K;
1950 #elif (defined AARCH64)
1951 static Elf32_Half running_arch_code=EM_AARCH64;
1952 #else
1953 #error Method os::dll_load requires that one of following is defined:\
1954 IA32, AMD64, IA64, __sparc, __powerpc__, ARM, S390, ALPHA, MIPS, MIPSEL, PARISC, M68K, AARCH64
1955 #endif
1956
1957 // Identify compatability class for VM's architecture and library's architecture
1958 // Obtain string descriptions for architectures
1959
1960 arch_t lib_arch={elf_head.e_machine,0,elf_head.e_ident[EI_CLASS], elf_head.e_ident[EI_DATA], NULL};
1961 int running_arch_index=-1;
1962
1963 for (unsigned int i=0; i < ARRAY_SIZE(arch_array); i++) {
1964 if (running_arch_code == arch_array[i].code) {
1965 running_arch_index = i;
1966 }
1967 if (lib_arch.code == arch_array[i].code) {
1968 lib_arch.compat_class = arch_array[i].compat_class;
1969 lib_arch.name = arch_array[i].name;
1970 }
1971 }
1972
1973 assert(running_arch_index != -1,
1974 "Didn't find running architecture code (running_arch_code) in arch_array");
1975 if (running_arch_index == -1) {
1976 // Even though running architecture detection failed
1977 // we may still continue with reporting dlerror() message
1978 return NULL;
1979 }
1980
1981 if (lib_arch.endianess != arch_array[running_arch_index].endianess) {
1982 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: endianness mismatch)");
1983 return NULL;
1984 }
1985
1986 #ifndef S390
1987 if (lib_arch.elf_class != arch_array[running_arch_index].elf_class) {
1988 ::snprintf(diag_msg_buf, diag_msg_max_length-1," (Possible cause: architecture word width mismatch)");
1989 return NULL;
1990 }
1991 #endif // !S390
1992
1993 if (lib_arch.compat_class != arch_array[running_arch_index].compat_class) {
1994 if (lib_arch.name!=NULL) {
1995 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
1996 " (Possible cause: can't load %s-bit .so on a %s-bit platform)",
1997 lib_arch.name, arch_array[running_arch_index].name);
1998 } else {
1999 ::snprintf(diag_msg_buf, diag_msg_max_length-1,
2000 " (Possible cause: can't load this .so (machine code=0x%x) on a %s-bit platform)",
2001 lib_arch.code,
2002 arch_array[running_arch_index].name);
2003 }
2004 }
2005
2006 return NULL;
2007 }
2008
2009 void * os::Linux::dlopen_helper(const char *filename, char *ebuf,
2010 int ebuflen) {
2011 void * result = ::dlopen(filename, RTLD_LAZY);
2012 if (result == NULL) {
2013 ::strncpy(ebuf, ::dlerror(), ebuflen - 1);
2014 ebuf[ebuflen-1] = '\0';
2015 }
2016 return result;
2017 }
2018
2019 void * os::Linux::dll_load_in_vmthread(const char *filename, char *ebuf,
2020 int ebuflen) {
2021 void * result = NULL;
2022 if (LoadExecStackDllInVMThread) {
2023 result = dlopen_helper(filename, ebuf, ebuflen);
2024 }
2025
2026 // Since 7019808, libjvm.so is linked with -noexecstack. If the VM loads a
2027 // library that requires an executable stack, or which does not have this
2028 // stack attribute set, dlopen changes the stack attribute to executable. The
2029 // read protection of the guard pages gets lost.
2030 //
2031 // Need to check _stack_is_executable again as multiple VM_LinuxDllLoad
2032 // may have been queued at the same time.
2033
2034 if (!_stack_is_executable) {
2035 JavaThread *jt = Threads::first();
2036
2037 while (jt) {
2038 if (!jt->stack_guard_zone_unused() && // Stack not yet fully initialized
2039 jt->stack_yellow_zone_enabled()) { // No pending stack overflow exceptions
2040 if (!os::guard_memory((char *) jt->stack_red_zone_base() - jt->stack_red_zone_size(),
2041 jt->stack_yellow_zone_size() + jt->stack_red_zone_size())) {
2042 warning("Attempt to reguard stack yellow zone failed.");
2043 }
2044 }
2045 jt = jt->next();
2046 }
2047 }
2048
2049 return result;
2050 }
2051
2052 // glibc-2.0 libdl is not MT safe. If you are building with any glibc,
2053 // chances are you might want to run the generated bits against glibc-2.0
2054 // libdl.so, so always use locking for any version of glibc.
2055 //
2056 void* os::dll_lookup(void* handle, const char* name) {
2057 pthread_mutex_lock(&dl_mutex);
2058 void* res = dlsym(handle, name);
2059 pthread_mutex_unlock(&dl_mutex);
2060 return res;
2061 }
2062
2063 void* os::get_default_process_handle() {
2064 return (void*)::dlopen(NULL, RTLD_LAZY);
2065 }
2066
2067 static bool _print_ascii_file(const char* filename, outputStream* st) {
2068 int fd = ::open(filename, O_RDONLY);
2069 if (fd == -1) {
2070 return false;
2071 }
2072
2073 char buf[32];
2074 int bytes;
2075 while ((bytes = ::read(fd, buf, sizeof(buf))) > 0) {
2076 st->print_raw(buf, bytes);
2077 }
2078
2079 ::close(fd);
2080
2081 return true;
2082 }
2083
2084 void os::print_dll_info(outputStream *st) {
2085 st->print_cr("Dynamic libraries:");
2086
2087 char fname[32];
2088 pid_t pid = os::Linux::gettid();
2089
2090 jio_snprintf(fname, sizeof(fname), "/proc/%d/maps", pid);
2091
2092 if (!_print_ascii_file(fname, st)) {
2093 st->print("Can not get library information for pid = %d\n", pid);
2094 }
2095 }
2096
2097 int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *param) {
2098 FILE *procmapsFile = NULL;
2099
2100 // Open the procfs maps file for the current process
2101 if ((procmapsFile = fopen("/proc/self/maps", "r")) != NULL) {
2102 // Allocate PATH_MAX for file name plus a reasonable size for other fields.
2103 char line[PATH_MAX + 100];
2104
2105 // Read line by line from 'file'
2106 while (fgets(line, sizeof(line), procmapsFile) != NULL) {
2107 u8 base, top, offset, inode;
2108 char permissions[5];
2109 char device[6];
2110 char name[PATH_MAX + 1];
2111
2112 // Parse fields from line
2113 sscanf(line, "%lx-%lx %4s %lx %5s %ld %s", &base, &top, permissions, &offset, device, &inode, name);
2114
2115 // Filter by device id '00:00' so that we only get file system mapped files.
2116 if (strcmp(device, "00:00") != 0) {
2117
2118 // Call callback with the fields of interest
2119 if(callback(name, (address)base, (address)top, param)) {
2120 // Oops abort, callback aborted
2121 fclose(procmapsFile);
2122 return 1;
2123 }
2124 }
2125 }
2126 fclose(procmapsFile);
2127 }
2128 return 0;
2129 }
2130
2131 void os::print_os_info_brief(outputStream* st) {
2132 os::Linux::print_distro_info(st);
2133
2134 os::Posix::print_uname_info(st);
2135
2136 os::Linux::print_libversion_info(st);
2137
2138 }
2139
2140 void os::print_os_info(outputStream* st) {
2141 st->print("OS:");
2142
2143 os::Linux::print_distro_info(st);
2144
2145 os::Posix::print_uname_info(st);
2146
2147 // Print warning if unsafe chroot environment detected
2148 if (unsafe_chroot_detected) {
2149 st->print("WARNING!! ");
2150 st->print_cr("%s", unstable_chroot_error);
2151 }
2152
2153 os::Linux::print_libversion_info(st);
2154
2155 os::Posix::print_rlimit_info(st);
2156
2157 os::Posix::print_load_average(st);
2158
2159 os::Linux::print_full_memory_info(st);
2160 }
2161
2162 // Try to identify popular distros.
2163 // Most Linux distributions have a /etc/XXX-release file, which contains
2164 // the OS version string. Newer Linux distributions have a /etc/lsb-release
2165 // file that also contains the OS version string. Some have more than one
2166 // /etc/XXX-release file (e.g. Mandrake has both /etc/mandrake-release and
2167 // /etc/redhat-release.), so the order is important.
2168 // Any Linux that is based on Redhat (i.e. Oracle, Mandrake, Sun JDS...) have
2169 // their own specific XXX-release file as well as a redhat-release file.
2170 // Because of this the XXX-release file needs to be searched for before the
2171 // redhat-release file.
2172 // Since Red Hat has a lsb-release file that is not very descriptive the
2173 // search for redhat-release needs to be before lsb-release.
2174 // Since the lsb-release file is the new standard it needs to be searched
2175 // before the older style release files.
2176 // Searching system-release (Red Hat) and os-release (other Linuxes) are a
2177 // next to last resort. The os-release file is a new standard that contains
2178 // distribution information and the system-release file seems to be an old
2179 // standard that has been replaced by the lsb-release and os-release files.
2180 // Searching for the debian_version file is the last resort. It contains
2181 // an informative string like "6.0.6" or "wheezy/sid". Because of this
2182 // "Debian " is printed before the contents of the debian_version file.
2183 void os::Linux::print_distro_info(outputStream* st) {
2184 if (!_print_ascii_file("/etc/oracle-release", st) &&
2185 !_print_ascii_file("/etc/mandriva-release", st) &&
2186 !_print_ascii_file("/etc/mandrake-release", st) &&
2187 !_print_ascii_file("/etc/sun-release", st) &&
2188 !_print_ascii_file("/etc/redhat-release", st) &&
2189 !_print_ascii_file("/etc/lsb-release", st) &&
2190 !_print_ascii_file("/etc/SuSE-release", st) &&
2191 !_print_ascii_file("/etc/turbolinux-release", st) &&
2192 !_print_ascii_file("/etc/gentoo-release", st) &&
2193 !_print_ascii_file("/etc/ltib-release", st) &&
2194 !_print_ascii_file("/etc/angstrom-version", st) &&
2195 !_print_ascii_file("/etc/system-release", st) &&
2196 !_print_ascii_file("/etc/os-release", st)) {
2197
2198 if (file_exists("/etc/debian_version")) {
2199 st->print("Debian ");
2200 _print_ascii_file("/etc/debian_version", st);
2201 } else {
2202 st->print("Linux");
2203 }
2204 }
2205 st->cr();
2206 }
2207
2208 void os::Linux::print_libversion_info(outputStream* st) {
2209 // libc, pthread
2210 st->print("libc:");
2211 st->print("%s ", os::Linux::glibc_version());
2212 st->print("%s ", os::Linux::libpthread_version());
2213 if (os::Linux::is_LinuxThreads()) {
2214 st->print("(%s stack)", os::Linux::is_floating_stack() ? "floating" : "fixed");
2215 }
2216 st->cr();
2217 }
2218
2219 void os::Linux::print_full_memory_info(outputStream* st) {
2220 st->print("\n/proc/meminfo:\n");
2221 _print_ascii_file("/proc/meminfo", st);
2222 st->cr();
2223 }
2224
2225 void os::print_memory_info(outputStream* st) {
2226
2227 st->print("Memory:");
2228 st->print(" %dk page", os::vm_page_size()>>10);
2229
2230 // values in struct sysinfo are "unsigned long"
2231 struct sysinfo si;
2232 sysinfo(&si);
2233
2234 st->print(", physical " UINT64_FORMAT "k",
2235 os::physical_memory() >> 10);
2236 st->print("(" UINT64_FORMAT "k free)",
2237 os::available_memory() >> 10);
2238 st->print(", swap " UINT64_FORMAT "k",
2239 ((jlong)si.totalswap * si.mem_unit) >> 10);
2240 st->print("(" UINT64_FORMAT "k free)",
2241 ((jlong)si.freeswap * si.mem_unit) >> 10);
2242 st->cr();
2243 }
2244
2245 void os::pd_print_cpu_info(outputStream* st) {
2246 st->print("\n/proc/cpuinfo:\n");
2247 if (!_print_ascii_file("/proc/cpuinfo", st)) {
2248 st->print(" <Not Available>");
2249 }
2250 st->cr();
2251 }
2252
2253 void os::print_siginfo(outputStream* st, void* siginfo) {
2254 const siginfo_t* si = (const siginfo_t*)siginfo;
2255
2256 os::Posix::print_siginfo_brief(st, si);
2257 #if INCLUDE_CDS
2258 if (si && (si->si_signo == SIGBUS || si->si_signo == SIGSEGV) &&
2259 UseSharedSpaces) {
2260 FileMapInfo* mapinfo = FileMapInfo::current_info();
2261 if (mapinfo->is_in_shared_space(si->si_addr)) {
2262 st->print("\n\nError accessing class data sharing archive." \
2263 " Mapped file inaccessible during execution, " \
2264 " possible disk/network problem.");
2265 }
2266 }
2267 #endif
2268 st->cr();
2269 }
2270
2271
2272 static void print_signal_handler(outputStream* st, int sig,
2273 char* buf, size_t buflen);
2274
2275 void os::print_signal_handlers(outputStream* st, char* buf, size_t buflen) {
2276 st->print_cr("Signal Handlers:");
2277 print_signal_handler(st, SIGSEGV, buf, buflen);
2278 print_signal_handler(st, SIGBUS , buf, buflen);
2279 print_signal_handler(st, SIGFPE , buf, buflen);
2280 print_signal_handler(st, SIGPIPE, buf, buflen);
2281 print_signal_handler(st, SIGXFSZ, buf, buflen);
2282 print_signal_handler(st, SIGILL , buf, buflen);
2283 print_signal_handler(st, INTERRUPT_SIGNAL, buf, buflen);
2284 print_signal_handler(st, SR_signum, buf, buflen);
2285 print_signal_handler(st, SHUTDOWN1_SIGNAL, buf, buflen);
2286 print_signal_handler(st, SHUTDOWN2_SIGNAL , buf, buflen);
2287 print_signal_handler(st, SHUTDOWN3_SIGNAL , buf, buflen);
2288 print_signal_handler(st, BREAK_SIGNAL, buf, buflen);
2289 #if defined(PPC64)
2290 print_signal_handler(st, SIGTRAP, buf, buflen);
2291 #endif
2292 }
2293
2294 static char saved_jvm_path[MAXPATHLEN] = {0};
2295
2296 // Find the full path to the current module, libjvm.so
2297 void os::jvm_path(char *buf, jint buflen) {
2298 // Error checking.
2299 if (buflen < MAXPATHLEN) {
2300 assert(false, "must use a large-enough buffer");
2301 buf[0] = '\0';
2302 return;
2303 }
2304 // Lazy resolve the path to current module.
2305 if (saved_jvm_path[0] != 0) {
2306 strcpy(buf, saved_jvm_path);
2307 return;
2308 }
2309
2310 char dli_fname[MAXPATHLEN];
2311 bool ret = dll_address_to_library_name(
2312 CAST_FROM_FN_PTR(address, os::jvm_path),
2313 dli_fname, sizeof(dli_fname), NULL);
2314 assert(ret, "cannot locate libjvm");
2315 char *rp = NULL;
2316 if (ret && dli_fname[0] != '\0') {
2317 rp = realpath(dli_fname, buf);
2318 }
2319 if (rp == NULL) {
2320 return;
2321 }
2322
2323 if (Arguments::sun_java_launcher_is_altjvm()) {
2324 // Support for the java launcher's '-XXaltjvm=<path>' option. Typical
2325 // value for buf is "<JAVA_HOME>/jre/lib/<arch>/<vmtype>/libjvm.so".
2326 // If "/jre/lib/" appears at the right place in the string, then
2327 // assume we are installed in a JDK and we're done. Otherwise, check
2328 // for a JAVA_HOME environment variable and fix up the path so it
2329 // looks like libjvm.so is installed there (append a fake suffix
2330 // hotspot/libjvm.so).
2331 const char *p = buf + strlen(buf) - 1;
2332 for (int count = 0; p > buf && count < 5; ++count) {
2333 for (--p; p > buf && *p != '/'; --p)
2334 /* empty */ ;
2335 }
2336
2337 if (strncmp(p, "/jre/lib/", 9) != 0) {
2338 // Look for JAVA_HOME in the environment.
2339 char* java_home_var = ::getenv("JAVA_HOME");
2340 if (java_home_var != NULL && java_home_var[0] != 0) {
2341 char* jrelib_p;
2342 int len;
2343
2344 // Check the current module name "libjvm.so".
2345 p = strrchr(buf, '/');
2346 if (p == NULL) {
2347 return;
2348 }
2349 assert(strstr(p, "/libjvm") == p, "invalid library name");
2350
2351 rp = realpath(java_home_var, buf);
2352 if (rp == NULL) {
2353 return;
2354 }
2355
2356 // determine if this is a legacy image or modules image
2357 // modules image doesn't have "jre" subdirectory
2358 len = strlen(buf);
2359 assert(len < buflen, "Ran out of buffer room");
2360 jrelib_p = buf + len;
2361 snprintf(jrelib_p, buflen-len, "/jre/lib/%s", cpu_arch);
2362 if (0 != access(buf, F_OK)) {
2363 snprintf(jrelib_p, buflen-len, "/lib/%s", cpu_arch);
2364 }
2365
2366 if (0 == access(buf, F_OK)) {
2367 // Use current module name "libjvm.so"
2368 len = strlen(buf);
2369 snprintf(buf + len, buflen-len, "/hotspot/libjvm.so");
2370 } else {
2371 // Go back to path of .so
2372 rp = realpath(dli_fname, buf);
2373 if (rp == NULL) {
2374 return;
2375 }
2376 }
2377 }
2378 }
2379 }
2380
2381 strncpy(saved_jvm_path, buf, MAXPATHLEN);
2382 saved_jvm_path[MAXPATHLEN - 1] = '\0';
2383 }
2384
2385 void os::print_jni_name_prefix_on(outputStream* st, int args_size) {
2386 // no prefix required, not even "_"
2387 }
2388
2389 void os::print_jni_name_suffix_on(outputStream* st, int args_size) {
2390 // no suffix required
2391 }
2392
2393 ////////////////////////////////////////////////////////////////////////////////
2394 // sun.misc.Signal support
2395
2396 static volatile jint sigint_count = 0;
2397
2398 static void UserHandler(int sig, void *siginfo, void *context) {
2399 // 4511530 - sem_post is serialized and handled by the manager thread. When
2400 // the program is interrupted by Ctrl-C, SIGINT is sent to every thread. We
2401 // don't want to flood the manager thread with sem_post requests.
2402 if (sig == SIGINT && Atomic::add(1, &sigint_count) > 1) {
2403 return;
2404 }
2405
2406 // Ctrl-C is pressed during error reporting, likely because the error
2407 // handler fails to abort. Let VM die immediately.
2408 if (sig == SIGINT && is_error_reported()) {
2409 os::die();
2410 }
2411
2412 os::signal_notify(sig);
2413 }
2414
2415 void* os::user_handler() {
2416 return CAST_FROM_FN_PTR(void*, UserHandler);
2417 }
2418
2419 class Semaphore : public StackObj {
2420 public:
2421 Semaphore();
2422 ~Semaphore();
2423 void signal();
2424 void wait();
2425 bool trywait();
2426 bool timedwait(unsigned int sec, int nsec);
2427 private:
2428 sem_t _semaphore;
2429 };
2430
2431 Semaphore::Semaphore() {
2432 sem_init(&_semaphore, 0, 0);
2433 }
2434
2435 Semaphore::~Semaphore() {
2436 sem_destroy(&_semaphore);
2437 }
2438
2439 void Semaphore::signal() {
2440 sem_post(&_semaphore);
2441 }
2442
2443 void Semaphore::wait() {
2444 sem_wait(&_semaphore);
2445 }
2446
2447 bool Semaphore::trywait() {
2448 return sem_trywait(&_semaphore) == 0;
2449 }
2450
2451 bool Semaphore::timedwait(unsigned int sec, int nsec) {
2452
2453 struct timespec ts;
2454 // Semaphore's are always associated with CLOCK_REALTIME
2455 os::Linux::clock_gettime(CLOCK_REALTIME, &ts);
2456 // see unpackTime for discussion on overflow checking
2457 if (sec >= MAX_SECS) {
2458 ts.tv_sec += MAX_SECS;
2459 ts.tv_nsec = 0;
2460 } else {
2461 ts.tv_sec += sec;
2462 ts.tv_nsec += nsec;
2463 if (ts.tv_nsec >= NANOSECS_PER_SEC) {
2464 ts.tv_nsec -= NANOSECS_PER_SEC;
2465 ++ts.tv_sec; // note: this must be <= max_secs
2466 }
2467 }
2468
2469 while (1) {
2470 int result = sem_timedwait(&_semaphore, &ts);
2471 if (result == 0) {
2472 return true;
2473 } else if (errno == EINTR) {
2474 continue;
2475 } else if (errno == ETIMEDOUT) {
2476 return false;
2477 } else {
2478 return false;
2479 }
2480 }
2481 }
2482
2483 extern "C" {
2484 typedef void (*sa_handler_t)(int);
2485 typedef void (*sa_sigaction_t)(int, siginfo_t *, void *);
2486 }
2487
2488 void* os::signal(int signal_number, void* handler) {
2489 struct sigaction sigAct, oldSigAct;
2490
2491 sigfillset(&(sigAct.sa_mask));
2492 sigAct.sa_flags = SA_RESTART|SA_SIGINFO;
2493 sigAct.sa_handler = CAST_TO_FN_PTR(sa_handler_t, handler);
2494
2495 if (sigaction(signal_number, &sigAct, &oldSigAct)) {
2496 // -1 means registration failed
2497 return (void *)-1;
2498 }
2499
2500 return CAST_FROM_FN_PTR(void*, oldSigAct.sa_handler);
2501 }
2502
2503 void os::signal_raise(int signal_number) {
2504 ::raise(signal_number);
2505 }
2506
2507 // The following code is moved from os.cpp for making this
2508 // code platform specific, which it is by its very nature.
2509
2510 // Will be modified when max signal is changed to be dynamic
2511 int os::sigexitnum_pd() {
2512 return NSIG;
2513 }
2514
2515 // a counter for each possible signal value
2516 static volatile jint pending_signals[NSIG+1] = { 0 };
2517
2518 // Linux(POSIX) specific hand shaking semaphore.
2519 static sem_t sig_sem;
2520 static Semaphore sr_semaphore;
2521
2522 void os::signal_init_pd() {
2523 // Initialize signal structures
2524 ::memset((void*)pending_signals, 0, sizeof(pending_signals));
2525
2526 // Initialize signal semaphore
2527 ::sem_init(&sig_sem, 0, 0);
2528 }
2529
2530 void os::signal_notify(int sig) {
2531 Atomic::inc(&pending_signals[sig]);
2532 ::sem_post(&sig_sem);
2533 }
2534
2535 static int check_pending_signals(bool wait) {
2536 Atomic::store(0, &sigint_count);
2537 for (;;) {
2538 for (int i = 0; i < NSIG + 1; i++) {
2539 jint n = pending_signals[i];
2540 if (n > 0 && n == Atomic::cmpxchg(n - 1, &pending_signals[i], n)) {
2541 return i;
2542 }
2543 }
2544 if (!wait) {
2545 return -1;
2546 }
2547 JavaThread *thread = JavaThread::current();
2548 ThreadBlockInVM tbivm(thread);
2549
2550 bool threadIsSuspended;
2551 do {
2552 thread->set_suspend_equivalent();
2553 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
2554 ::sem_wait(&sig_sem);
2555
2556 // were we externally suspended while we were waiting?
2557 threadIsSuspended = thread->handle_special_suspend_equivalent_condition();
2558 if (threadIsSuspended) {
2559 // The semaphore has been incremented, but while we were waiting
2560 // another thread suspended us. We don't want to continue running
2561 // while suspended because that would surprise the thread that
2562 // suspended us.
2563 ::sem_post(&sig_sem);
2564
2565 thread->java_suspend_self();
2566 }
2567 } while (threadIsSuspended);
2568 }
2569 }
2570
2571 int os::signal_lookup() {
2572 return check_pending_signals(false);
2573 }
2574
2575 int os::signal_wait() {
2576 return check_pending_signals(true);
2577 }
2578
2579 ////////////////////////////////////////////////////////////////////////////////
2580 // Virtual Memory
2581
2582 int os::vm_page_size() {
2583 // Seems redundant as all get out
2584 assert(os::Linux::page_size() != -1, "must call os::init");
2585 return os::Linux::page_size();
2586 }
2587
2588 // Solaris allocates memory by pages.
2589 int os::vm_allocation_granularity() {
2590 assert(os::Linux::page_size() != -1, "must call os::init");
2591 return os::Linux::page_size();
2592 }
2593
2594 // Rationale behind this function:
2595 // current (Mon Apr 25 20:12:18 MSD 2005) oprofile drops samples without executable
2596 // mapping for address (see lookup_dcookie() in the kernel module), thus we cannot get
2597 // samples for JITted code. Here we create private executable mapping over the code cache
2598 // and then we can use standard (well, almost, as mapping can change) way to provide
2599 // info for the reporting script by storing timestamp and location of symbol
2600 void linux_wrap_code(char* base, size_t size) {
2601 static volatile jint cnt = 0;
2602
2603 if (!UseOprofile) {
2604 return;
2605 }
2606
2607 char buf[PATH_MAX+1];
2608 int num = Atomic::add(1, &cnt);
2609
2610 snprintf(buf, sizeof(buf), "%s/hs-vm-%d-%d",
2611 os::get_temp_directory(), os::current_process_id(), num);
2612 unlink(buf);
2613
2614 int fd = ::open(buf, O_CREAT | O_RDWR, S_IRWXU);
2615
2616 if (fd != -1) {
2617 off_t rv = ::lseek(fd, size-2, SEEK_SET);
2618 if (rv != (off_t)-1) {
2619 if (::write(fd, "", 1) == 1) {
2620 mmap(base, size,
2621 PROT_READ|PROT_WRITE|PROT_EXEC,
2622 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE, fd, 0);
2623 }
2624 }
2625 ::close(fd);
2626 unlink(buf);
2627 }
2628 }
2629
2630 static bool recoverable_mmap_error(int err) {
2631 // See if the error is one we can let the caller handle. This
2632 // list of errno values comes from JBS-6843484. I can't find a
2633 // Linux man page that documents this specific set of errno
2634 // values so while this list currently matches Solaris, it may
2635 // change as we gain experience with this failure mode.
2636 switch (err) {
2637 case EBADF:
2638 case EINVAL:
2639 case ENOTSUP:
2640 // let the caller deal with these errors
2641 return true;
2642
2643 default:
2644 // Any remaining errors on this OS can cause our reserved mapping
2645 // to be lost. That can cause confusion where different data
2646 // structures think they have the same memory mapped. The worst
2647 // scenario is if both the VM and a library think they have the
2648 // same memory mapped.
2649 return false;
2650 }
2651 }
2652
2653 static void warn_fail_commit_memory(char* addr, size_t size, bool exec,
2654 int err) {
2655 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2656 ", %d) failed; error='%s' (errno=%d)", addr, size, exec,
2657 strerror(err), err);
2658 }
2659
2660 static void warn_fail_commit_memory(char* addr, size_t size,
2661 size_t alignment_hint, bool exec,
2662 int err) {
2663 warning("INFO: os::commit_memory(" PTR_FORMAT ", " SIZE_FORMAT
2664 ", " SIZE_FORMAT ", %d) failed; error='%s' (errno=%d)", addr, size,
2665 alignment_hint, exec, strerror(err), err);
2666 }
2667
2668 // NOTE: Linux kernel does not really reserve the pages for us.
2669 // All it does is to check if there are enough free pages
2670 // left at the time of mmap(). This could be a potential
2671 // problem.
2672 int os::Linux::commit_memory_impl(char* addr, size_t size, bool exec) {
2673 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
2674 uintptr_t res = (uintptr_t) ::mmap(addr, size, prot,
2675 MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS, -1, 0);
2676 if (res != (uintptr_t) MAP_FAILED) {
2677 if (UseNUMAInterleaving) {
2678 numa_make_global(addr, size);
2679 }
2680 return 0;
2681 }
2682
2683 int err = errno; // save errno from mmap() call above
2684
2685 if (!recoverable_mmap_error(err)) {
2686 warn_fail_commit_memory(addr, size, exec, err);
2687 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, "committing reserved memory.");
2688 }
2689
2690 return err;
2691 }
2692
2693 bool os::pd_commit_memory(char* addr, size_t size, bool exec) {
2694 return os::Linux::commit_memory_impl(addr, size, exec) == 0;
2695 }
2696
2697 void os::pd_commit_memory_or_exit(char* addr, size_t size, bool exec,
2698 const char* mesg) {
2699 assert(mesg != NULL, "mesg must be specified");
2700 int err = os::Linux::commit_memory_impl(addr, size, exec);
2701 if (err != 0) {
2702 // the caller wants all commit errors to exit with the specified mesg:
2703 warn_fail_commit_memory(addr, size, exec, err);
2704 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2705 }
2706 }
2707
2708 // Define MAP_HUGETLB here so we can build HotSpot on old systems.
2709 #ifndef MAP_HUGETLB
2710 #define MAP_HUGETLB 0x40000
2711 #endif
2712
2713 // Define MADV_HUGEPAGE here so we can build HotSpot on old systems.
2714 #ifndef MADV_HUGEPAGE
2715 #define MADV_HUGEPAGE 14
2716 #endif
2717
2718 int os::Linux::commit_memory_impl(char* addr, size_t size,
2719 size_t alignment_hint, bool exec) {
2720 int err = os::Linux::commit_memory_impl(addr, size, exec);
2721 if (err == 0) {
2722 realign_memory(addr, size, alignment_hint);
2723 }
2724 return err;
2725 }
2726
2727 bool os::pd_commit_memory(char* addr, size_t size, size_t alignment_hint,
2728 bool exec) {
2729 return os::Linux::commit_memory_impl(addr, size, alignment_hint, exec) == 0;
2730 }
2731
2732 void os::pd_commit_memory_or_exit(char* addr, size_t size,
2733 size_t alignment_hint, bool exec,
2734 const char* mesg) {
2735 assert(mesg != NULL, "mesg must be specified");
2736 int err = os::Linux::commit_memory_impl(addr, size, alignment_hint, exec);
2737 if (err != 0) {
2738 // the caller wants all commit errors to exit with the specified mesg:
2739 warn_fail_commit_memory(addr, size, alignment_hint, exec, err);
2740 vm_exit_out_of_memory(size, OOM_MMAP_ERROR, mesg);
2741 }
2742 }
2743
2744 void os::pd_realign_memory(char *addr, size_t bytes, size_t alignment_hint) {
2745 if (UseTransparentHugePages && alignment_hint > (size_t)vm_page_size()) {
2746 // We don't check the return value: madvise(MADV_HUGEPAGE) may not
2747 // be supported or the memory may already be backed by huge pages.
2748 ::madvise(addr, bytes, MADV_HUGEPAGE);
2749 }
2750 }
2751
2752 void os::pd_free_memory(char *addr, size_t bytes, size_t alignment_hint) {
2753 // This method works by doing an mmap over an existing mmaping and effectively discarding
2754 // the existing pages. However it won't work for SHM-based large pages that cannot be
2755 // uncommitted at all. We don't do anything in this case to avoid creating a segment with
2756 // small pages on top of the SHM segment. This method always works for small pages, so we
2757 // allow that in any case.
2758 if (alignment_hint <= (size_t)os::vm_page_size() || can_commit_large_page_memory()) {
2759 commit_memory(addr, bytes, alignment_hint, !ExecMem);
2760 }
2761 }
2762
2763 void os::numa_make_global(char *addr, size_t bytes) {
2764 Linux::numa_interleave_memory(addr, bytes);
2765 }
2766
2767 // Define for numa_set_bind_policy(int). Setting the argument to 0 will set the
2768 // bind policy to MPOL_PREFERRED for the current thread.
2769 #define USE_MPOL_PREFERRED 0
2770
2771 void os::numa_make_local(char *addr, size_t bytes, int lgrp_hint) {
2772 // To make NUMA and large pages more robust when both enabled, we need to ease
2773 // the requirements on where the memory should be allocated. MPOL_BIND is the
2774 // default policy and it will force memory to be allocated on the specified
2775 // node. Changing this to MPOL_PREFERRED will prefer to allocate the memory on
2776 // the specified node, but will not force it. Using this policy will prevent
2777 // getting SIGBUS when trying to allocate large pages on NUMA nodes with no
2778 // free large pages.
2779 Linux::numa_set_bind_policy(USE_MPOL_PREFERRED);
2780 Linux::numa_tonode_memory(addr, bytes, lgrp_hint);
2781 }
2782
2783 bool os::numa_topology_changed() { return false; }
2784
2785 size_t os::numa_get_groups_num() {
2786 int max_node = Linux::numa_max_node();
2787 return max_node > 0 ? max_node + 1 : 1;
2788 }
2789
2790 int os::numa_get_group_id() {
2791 int cpu_id = Linux::sched_getcpu();
2792 if (cpu_id != -1) {
2793 int lgrp_id = Linux::get_node_by_cpu(cpu_id);
2794 if (lgrp_id != -1) {
2795 return lgrp_id;
2796 }
2797 }
2798 return 0;
2799 }
2800
2801 size_t os::numa_get_leaf_groups(int *ids, size_t size) {
2802 for (size_t i = 0; i < size; i++) {
2803 ids[i] = i;
2804 }
2805 return size;
2806 }
2807
2808 bool os::get_page_info(char *start, page_info* info) {
2809 return false;
2810 }
2811
2812 char *os::scan_pages(char *start, char* end, page_info* page_expected,
2813 page_info* page_found) {
2814 return end;
2815 }
2816
2817
2818 int os::Linux::sched_getcpu_syscall(void) {
2819 unsigned int cpu;
2820 int retval = -1;
2821
2822 #if defined(IA32)
2823 #ifndef SYS_getcpu
2824 #define SYS_getcpu 318
2825 #endif
2826 retval = syscall(SYS_getcpu, &cpu, NULL, NULL);
2827 #elif defined(AMD64)
2828 // Unfortunately we have to bring all these macros here from vsyscall.h
2829 // to be able to compile on old linuxes.
2830 #define __NR_vgetcpu 2
2831 #define VSYSCALL_START (-10UL << 20)
2832 #define VSYSCALL_SIZE 1024
2833 #define VSYSCALL_ADDR(vsyscall_nr) (VSYSCALL_START+VSYSCALL_SIZE*(vsyscall_nr))
2834 typedef long (*vgetcpu_t)(unsigned int *cpu, unsigned int *node, unsigned long *tcache);
2835 vgetcpu_t vgetcpu = (vgetcpu_t)VSYSCALL_ADDR(__NR_vgetcpu);
2836 retval = vgetcpu(&cpu, NULL, NULL);
2837 #endif
2838
2839 return (retval == -1) ? retval : cpu;
2840 }
2841
2842 // Something to do with the numa-aware allocator needs these symbols
2843 extern "C" JNIEXPORT void numa_warn(int number, char *where, ...) { }
2844 extern "C" JNIEXPORT void numa_error(char *where) { }
2845 extern "C" JNIEXPORT int fork1() { return fork(); }
2846
2847
2848 // If we are running with libnuma version > 2, then we should
2849 // be trying to use symbols with versions 1.1
2850 // If we are running with earlier version, which did not have symbol versions,
2851 // we should use the base version.
2852 void* os::Linux::libnuma_dlsym(void* handle, const char *name) {
2853 void *f = dlvsym(handle, name, "libnuma_1.1");
2854 if (f == NULL) {
2855 f = dlsym(handle, name);
2856 }
2857 return f;
2858 }
2859
2860 bool os::Linux::libnuma_init() {
2861 // sched_getcpu() should be in libc.
2862 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2863 dlsym(RTLD_DEFAULT, "sched_getcpu")));
2864
2865 // If it's not, try a direct syscall.
2866 if (sched_getcpu() == -1) {
2867 set_sched_getcpu(CAST_TO_FN_PTR(sched_getcpu_func_t,
2868 (void*)&sched_getcpu_syscall));
2869 }
2870
2871 if (sched_getcpu() != -1) { // Does it work?
2872 void *handle = dlopen("libnuma.so.1", RTLD_LAZY);
2873 if (handle != NULL) {
2874 set_numa_node_to_cpus(CAST_TO_FN_PTR(numa_node_to_cpus_func_t,
2875 libnuma_dlsym(handle, "numa_node_to_cpus")));
2876 set_numa_max_node(CAST_TO_FN_PTR(numa_max_node_func_t,
2877 libnuma_dlsym(handle, "numa_max_node")));
2878 set_numa_available(CAST_TO_FN_PTR(numa_available_func_t,
2879 libnuma_dlsym(handle, "numa_available")));
2880 set_numa_tonode_memory(CAST_TO_FN_PTR(numa_tonode_memory_func_t,
2881 libnuma_dlsym(handle, "numa_tonode_memory")));
2882 set_numa_interleave_memory(CAST_TO_FN_PTR(numa_interleave_memory_func_t,
2883 libnuma_dlsym(handle, "numa_interleave_memory")));
2884 set_numa_set_bind_policy(CAST_TO_FN_PTR(numa_set_bind_policy_func_t,
2885 libnuma_dlsym(handle, "numa_set_bind_policy")));
2886
2887
2888 if (numa_available() != -1) {
2889 set_numa_all_nodes((unsigned long*)libnuma_dlsym(handle, "numa_all_nodes"));
2890 // Create a cpu -> node mapping
2891 _cpu_to_node = new (ResourceObj::C_HEAP, mtInternal) GrowableArray<int>(0, true);
2892 rebuild_cpu_to_node_map();
2893 return true;
2894 }
2895 }
2896 }
2897 return false;
2898 }
2899
2900 // rebuild_cpu_to_node_map() constructs a table mapping cpud id to node id.
2901 // The table is later used in get_node_by_cpu().
2902 void os::Linux::rebuild_cpu_to_node_map() {
2903 const size_t NCPUS = 32768; // Since the buffer size computation is very obscure
2904 // in libnuma (possible values are starting from 16,
2905 // and continuing up with every other power of 2, but less
2906 // than the maximum number of CPUs supported by kernel), and
2907 // is a subject to change (in libnuma version 2 the requirements
2908 // are more reasonable) we'll just hardcode the number they use
2909 // in the library.
2910 const size_t BitsPerCLong = sizeof(long) * CHAR_BIT;
2911
2912 size_t cpu_num = os::active_processor_count();
2913 size_t cpu_map_size = NCPUS / BitsPerCLong;
2914 size_t cpu_map_valid_size =
2915 MIN2((cpu_num + BitsPerCLong - 1) / BitsPerCLong, cpu_map_size);
2916
2917 cpu_to_node()->clear();
2918 cpu_to_node()->at_grow(cpu_num - 1);
2919 size_t node_num = numa_get_groups_num();
2920
2921 unsigned long *cpu_map = NEW_C_HEAP_ARRAY(unsigned long, cpu_map_size, mtInternal);
2922 for (size_t i = 0; i < node_num; i++) {
2923 if (numa_node_to_cpus(i, cpu_map, cpu_map_size * sizeof(unsigned long)) != -1) {
2924 for (size_t j = 0; j < cpu_map_valid_size; j++) {
2925 if (cpu_map[j] != 0) {
2926 for (size_t k = 0; k < BitsPerCLong; k++) {
2927 if (cpu_map[j] & (1UL << k)) {
2928 cpu_to_node()->at_put(j * BitsPerCLong + k, i);
2929 }
2930 }
2931 }
2932 }
2933 }
2934 }
2935 FREE_C_HEAP_ARRAY(unsigned long, cpu_map);
2936 }
2937
2938 int os::Linux::get_node_by_cpu(int cpu_id) {
2939 if (cpu_to_node() != NULL && cpu_id >= 0 && cpu_id < cpu_to_node()->length()) {
2940 return cpu_to_node()->at(cpu_id);
2941 }
2942 return -1;
2943 }
2944
2945 GrowableArray<int>* os::Linux::_cpu_to_node;
2946 os::Linux::sched_getcpu_func_t os::Linux::_sched_getcpu;
2947 os::Linux::numa_node_to_cpus_func_t os::Linux::_numa_node_to_cpus;
2948 os::Linux::numa_max_node_func_t os::Linux::_numa_max_node;
2949 os::Linux::numa_available_func_t os::Linux::_numa_available;
2950 os::Linux::numa_tonode_memory_func_t os::Linux::_numa_tonode_memory;
2951 os::Linux::numa_interleave_memory_func_t os::Linux::_numa_interleave_memory;
2952 os::Linux::numa_set_bind_policy_func_t os::Linux::_numa_set_bind_policy;
2953 unsigned long* os::Linux::_numa_all_nodes;
2954
2955 bool os::pd_uncommit_memory(char* addr, size_t size) {
2956 uintptr_t res = (uintptr_t) ::mmap(addr, size, PROT_NONE,
2957 MAP_PRIVATE|MAP_FIXED|MAP_NORESERVE|MAP_ANONYMOUS, -1, 0);
2958 return res != (uintptr_t) MAP_FAILED;
2959 }
2960
2961 static address get_stack_commited_bottom(address bottom, size_t size) {
2962 address nbot = bottom;
2963 address ntop = bottom + size;
2964
2965 size_t page_sz = os::vm_page_size();
2966 unsigned pages = size / page_sz;
2967
2968 unsigned char vec[1];
2969 unsigned imin = 1, imax = pages + 1, imid;
2970 int mincore_return_value = 0;
2971
2972 assert(imin <= imax, "Unexpected page size");
2973
2974 while (imin < imax) {
2975 imid = (imax + imin) / 2;
2976 nbot = ntop - (imid * page_sz);
2977
2978 // Use a trick with mincore to check whether the page is mapped or not.
2979 // mincore sets vec to 1 if page resides in memory and to 0 if page
2980 // is swapped output but if page we are asking for is unmapped
2981 // it returns -1,ENOMEM
2982 mincore_return_value = mincore(nbot, page_sz, vec);
2983
2984 if (mincore_return_value == -1) {
2985 // Page is not mapped go up
2986 // to find first mapped page
2987 if (errno != EAGAIN) {
2988 assert(errno == ENOMEM, "Unexpected mincore errno");
2989 imax = imid;
2990 }
2991 } else {
2992 // Page is mapped go down
2993 // to find first not mapped page
2994 imin = imid + 1;
2995 }
2996 }
2997
2998 nbot = nbot + page_sz;
2999
3000 // Adjust stack bottom one page up if last checked page is not mapped
3001 if (mincore_return_value == -1) {
3002 nbot = nbot + page_sz;
3003 }
3004
3005 return nbot;
3006 }
3007
3008
3009 // Linux uses a growable mapping for the stack, and if the mapping for
3010 // the stack guard pages is not removed when we detach a thread the
3011 // stack cannot grow beyond the pages where the stack guard was
3012 // mapped. If at some point later in the process the stack expands to
3013 // that point, the Linux kernel cannot expand the stack any further
3014 // because the guard pages are in the way, and a segfault occurs.
3015 //
3016 // However, it's essential not to split the stack region by unmapping
3017 // a region (leaving a hole) that's already part of the stack mapping,
3018 // so if the stack mapping has already grown beyond the guard pages at
3019 // the time we create them, we have to truncate the stack mapping.
3020 // So, we need to know the extent of the stack mapping when
3021 // create_stack_guard_pages() is called.
3022
3023 // We only need this for stacks that are growable: at the time of
3024 // writing thread stacks don't use growable mappings (i.e. those
3025 // creeated with MAP_GROWSDOWN), and aren't marked "[stack]", so this
3026 // only applies to the main thread.
3027
3028 // If the (growable) stack mapping already extends beyond the point
3029 // where we're going to put our guard pages, truncate the mapping at
3030 // that point by munmap()ping it. This ensures that when we later
3031 // munmap() the guard pages we don't leave a hole in the stack
3032 // mapping. This only affects the main/initial thread
3033
3034 bool os::pd_create_stack_guard_pages(char* addr, size_t size) {
3035 if (os::Linux::is_initial_thread()) {
3036 // As we manually grow stack up to bottom inside create_attached_thread(),
3037 // it's likely that os::Linux::initial_thread_stack_bottom is mapped and
3038 // we don't need to do anything special.
3039 // Check it first, before calling heavy function.
3040 uintptr_t stack_extent = (uintptr_t) os::Linux::initial_thread_stack_bottom();
3041 unsigned char vec[1];
3042
3043 if (mincore((address)stack_extent, os::vm_page_size(), vec) == -1) {
3044 // Fallback to slow path on all errors, including EAGAIN
3045 stack_extent = (uintptr_t) get_stack_commited_bottom(
3046 os::Linux::initial_thread_stack_bottom(),
3047 (size_t)addr - stack_extent);
3048 }
3049
3050 if (stack_extent < (uintptr_t)addr) {
3051 ::munmap((void*)stack_extent, (uintptr_t)(addr - stack_extent));
3052 }
3053 }
3054
3055 return os::commit_memory(addr, size, !ExecMem);
3056 }
3057
3058 // If this is a growable mapping, remove the guard pages entirely by
3059 // munmap()ping them. If not, just call uncommit_memory(). This only
3060 // affects the main/initial thread, but guard against future OS changes
3061 // It's safe to always unmap guard pages for initial thread because we
3062 // always place it right after end of the mapped region
3063
3064 bool os::remove_stack_guard_pages(char* addr, size_t size) {
3065 uintptr_t stack_extent, stack_base;
3066
3067 if (os::Linux::is_initial_thread()) {
3068 return ::munmap(addr, size) == 0;
3069 }
3070
3071 return os::uncommit_memory(addr, size);
3072 }
3073
3074 static address _highest_vm_reserved_address = NULL;
3075
3076 // If 'fixed' is true, anon_mmap() will attempt to reserve anonymous memory
3077 // at 'requested_addr'. If there are existing memory mappings at the same
3078 // location, however, they will be overwritten. If 'fixed' is false,
3079 // 'requested_addr' is only treated as a hint, the return value may or
3080 // may not start from the requested address. Unlike Linux mmap(), this
3081 // function returns NULL to indicate failure.
3082 static char* anon_mmap(char* requested_addr, size_t bytes, bool fixed) {
3083 char * addr;
3084 int flags;
3085
3086 flags = MAP_PRIVATE | MAP_NORESERVE | MAP_ANONYMOUS;
3087 if (fixed) {
3088 assert((uintptr_t)requested_addr % os::Linux::page_size() == 0, "unaligned address");
3089 flags |= MAP_FIXED;
3090 }
3091
3092 // Map reserved/uncommitted pages PROT_NONE so we fail early if we
3093 // touch an uncommitted page. Otherwise, the read/write might
3094 // succeed if we have enough swap space to back the physical page.
3095 addr = (char*)::mmap(requested_addr, bytes, PROT_NONE,
3096 flags, -1, 0);
3097
3098 if (addr != MAP_FAILED) {
3099 // anon_mmap() should only get called during VM initialization,
3100 // don't need lock (actually we can skip locking even it can be called
3101 // from multiple threads, because _highest_vm_reserved_address is just a
3102 // hint about the upper limit of non-stack memory regions.)
3103 if ((address)addr + bytes > _highest_vm_reserved_address) {
3104 _highest_vm_reserved_address = (address)addr + bytes;
3105 }
3106 }
3107
3108 return addr == MAP_FAILED ? NULL : addr;
3109 }
3110
3111 // Don't update _highest_vm_reserved_address, because there might be memory
3112 // regions above addr + size. If so, releasing a memory region only creates
3113 // a hole in the address space, it doesn't help prevent heap-stack collision.
3114 //
3115 static int anon_munmap(char * addr, size_t size) {
3116 return ::munmap(addr, size) == 0;
3117 }
3118
3119 char* os::pd_reserve_memory(size_t bytes, char* requested_addr,
3120 size_t alignment_hint) {
3121 return anon_mmap(requested_addr, bytes, (requested_addr != NULL));
3122 }
3123
3124 bool os::pd_release_memory(char* addr, size_t size) {
3125 return anon_munmap(addr, size);
3126 }
3127
3128 static address highest_vm_reserved_address() {
3129 return _highest_vm_reserved_address;
3130 }
3131
3132 static bool linux_mprotect(char* addr, size_t size, int prot) {
3133 // Linux wants the mprotect address argument to be page aligned.
3134 char* bottom = (char*)align_size_down((intptr_t)addr, os::Linux::page_size());
3135
3136 // According to SUSv3, mprotect() should only be used with mappings
3137 // established by mmap(), and mmap() always maps whole pages. Unaligned
3138 // 'addr' likely indicates problem in the VM (e.g. trying to change
3139 // protection of malloc'ed or statically allocated memory). Check the
3140 // caller if you hit this assert.
3141 assert(addr == bottom, "sanity check");
3142
3143 size = align_size_up(pointer_delta(addr, bottom, 1) + size, os::Linux::page_size());
3144 return ::mprotect(bottom, size, prot) == 0;
3145 }
3146
3147 // Set protections specified
3148 bool os::protect_memory(char* addr, size_t bytes, ProtType prot,
3149 bool is_committed) {
3150 unsigned int p = 0;
3151 switch (prot) {
3152 case MEM_PROT_NONE: p = PROT_NONE; break;
3153 case MEM_PROT_READ: p = PROT_READ; break;
3154 case MEM_PROT_RW: p = PROT_READ|PROT_WRITE; break;
3155 case MEM_PROT_RWX: p = PROT_READ|PROT_WRITE|PROT_EXEC; break;
3156 default:
3157 ShouldNotReachHere();
3158 }
3159 // is_committed is unused.
3160 return linux_mprotect(addr, bytes, p);
3161 }
3162
3163 bool os::guard_memory(char* addr, size_t size) {
3164 return linux_mprotect(addr, size, PROT_NONE);
3165 }
3166
3167 bool os::unguard_memory(char* addr, size_t size) {
3168 return linux_mprotect(addr, size, PROT_READ|PROT_WRITE);
3169 }
3170
3171 bool os::Linux::transparent_huge_pages_sanity_check(bool warn,
3172 size_t page_size) {
3173 bool result = false;
3174 void *p = mmap(NULL, page_size * 2, PROT_READ|PROT_WRITE,
3175 MAP_ANONYMOUS|MAP_PRIVATE,
3176 -1, 0);
3177 if (p != MAP_FAILED) {
3178 void *aligned_p = align_ptr_up(p, page_size);
3179
3180 result = madvise(aligned_p, page_size, MADV_HUGEPAGE) == 0;
3181
3182 munmap(p, page_size * 2);
3183 }
3184
3185 if (warn && !result) {
3186 warning("TransparentHugePages is not supported by the operating system.");
3187 }
3188
3189 return result;
3190 }
3191
3192 bool os::Linux::hugetlbfs_sanity_check(bool warn, size_t page_size) {
3193 bool result = false;
3194 void *p = mmap(NULL, page_size, PROT_READ|PROT_WRITE,
3195 MAP_ANONYMOUS|MAP_PRIVATE|MAP_HUGETLB,
3196 -1, 0);
3197
3198 if (p != MAP_FAILED) {
3199 // We don't know if this really is a huge page or not.
3200 FILE *fp = fopen("/proc/self/maps", "r");
3201 if (fp) {
3202 while (!feof(fp)) {
3203 char chars[257];
3204 long x = 0;
3205 if (fgets(chars, sizeof(chars), fp)) {
3206 if (sscanf(chars, "%lx-%*x", &x) == 1
3207 && x == (long)p) {
3208 if (strstr (chars, "hugepage")) {
3209 result = true;
3210 break;
3211 }
3212 }
3213 }
3214 }
3215 fclose(fp);
3216 }
3217 munmap(p, page_size);
3218 }
3219
3220 if (warn && !result) {
3221 warning("HugeTLBFS is not supported by the operating system.");
3222 }
3223
3224 return result;
3225 }
3226
3227 // Set the coredump_filter bits to include largepages in core dump (bit 6)
3228 //
3229 // From the coredump_filter documentation:
3230 //
3231 // - (bit 0) anonymous private memory
3232 // - (bit 1) anonymous shared memory
3233 // - (bit 2) file-backed private memory
3234 // - (bit 3) file-backed shared memory
3235 // - (bit 4) ELF header pages in file-backed private memory areas (it is
3236 // effective only if the bit 2 is cleared)
3237 // - (bit 5) hugetlb private memory
3238 // - (bit 6) hugetlb shared memory
3239 //
3240 static void set_coredump_filter(void) {
3241 FILE *f;
3242 long cdm;
3243
3244 if ((f = fopen("/proc/self/coredump_filter", "r+")) == NULL) {
3245 return;
3246 }
3247
3248 if (fscanf(f, "%lx", &cdm) != 1) {
3249 fclose(f);
3250 return;
3251 }
3252
3253 rewind(f);
3254
3255 if ((cdm & LARGEPAGES_BIT) == 0) {
3256 cdm |= LARGEPAGES_BIT;
3257 fprintf(f, "%#lx", cdm);
3258 }
3259
3260 fclose(f);
3261 }
3262
3263 // Large page support
3264
3265 static size_t _large_page_size = 0;
3266
3267 size_t os::Linux::find_large_page_size() {
3268 size_t large_page_size = 0;
3269
3270 // large_page_size on Linux is used to round up heap size. x86 uses either
3271 // 2M or 4M page, depending on whether PAE (Physical Address Extensions)
3272 // mode is enabled. AMD64/EM64T uses 2M page in 64bit mode. IA64 can use
3273 // page as large as 256M.
3274 //
3275 // Here we try to figure out page size by parsing /proc/meminfo and looking
3276 // for a line with the following format:
3277 // Hugepagesize: 2048 kB
3278 //
3279 // If we can't determine the value (e.g. /proc is not mounted, or the text
3280 // format has been changed), we'll use the largest page size supported by
3281 // the processor.
3282
3283 #ifndef ZERO
3284 large_page_size = IA32_ONLY(4 * M) AMD64_ONLY(2 * M) IA64_ONLY(256 * M) SPARC_ONLY(4 * M)
3285 ARM_ONLY(2 * M) PPC_ONLY(4 * M) AARCH64_ONLY(2 * M);
3286 #endif // ZERO
3287
3288 FILE *fp = fopen("/proc/meminfo", "r");
3289 if (fp) {
3290 while (!feof(fp)) {
3291 int x = 0;
3292 char buf[16];
3293 if (fscanf(fp, "Hugepagesize: %d", &x) == 1) {
3294 if (x && fgets(buf, sizeof(buf), fp) && strcmp(buf, " kB\n") == 0) {
3295 large_page_size = x * K;
3296 break;
3297 }
3298 } else {
3299 // skip to next line
3300 for (;;) {
3301 int ch = fgetc(fp);
3302 if (ch == EOF || ch == (int)'\n') break;
3303 }
3304 }
3305 }
3306 fclose(fp);
3307 }
3308
3309 if (!FLAG_IS_DEFAULT(LargePageSizeInBytes) && LargePageSizeInBytes != large_page_size) {
3310 warning("Setting LargePageSizeInBytes has no effect on this OS. Large page size is "
3311 SIZE_FORMAT "%s.", byte_size_in_proper_unit(large_page_size),
3312 proper_unit_for_byte_size(large_page_size));
3313 }
3314
3315 return large_page_size;
3316 }
3317
3318 size_t os::Linux::setup_large_page_size() {
3319 _large_page_size = Linux::find_large_page_size();
3320 const size_t default_page_size = (size_t)Linux::page_size();
3321 if (_large_page_size > default_page_size) {
3322 _page_sizes[0] = _large_page_size;
3323 _page_sizes[1] = default_page_size;
3324 _page_sizes[2] = 0;
3325 }
3326
3327 return _large_page_size;
3328 }
3329
3330 bool os::Linux::setup_large_page_type(size_t page_size) {
3331 if (FLAG_IS_DEFAULT(UseHugeTLBFS) &&
3332 FLAG_IS_DEFAULT(UseSHM) &&
3333 FLAG_IS_DEFAULT(UseTransparentHugePages)) {
3334
3335 // The type of large pages has not been specified by the user.
3336
3337 // Try UseHugeTLBFS and then UseSHM.
3338 UseHugeTLBFS = UseSHM = true;
3339
3340 // Don't try UseTransparentHugePages since there are known
3341 // performance issues with it turned on. This might change in the future.
3342 UseTransparentHugePages = false;
3343 }
3344
3345 if (UseTransparentHugePages) {
3346 bool warn_on_failure = !FLAG_IS_DEFAULT(UseTransparentHugePages);
3347 if (transparent_huge_pages_sanity_check(warn_on_failure, page_size)) {
3348 UseHugeTLBFS = false;
3349 UseSHM = false;
3350 return true;
3351 }
3352 UseTransparentHugePages = false;
3353 }
3354
3355 if (UseHugeTLBFS) {
3356 bool warn_on_failure = !FLAG_IS_DEFAULT(UseHugeTLBFS);
3357 if (hugetlbfs_sanity_check(warn_on_failure, page_size)) {
3358 UseSHM = false;
3359 return true;
3360 }
3361 UseHugeTLBFS = false;
3362 }
3363
3364 return UseSHM;
3365 }
3366
3367 void os::large_page_init() {
3368 if (!UseLargePages &&
3369 !UseTransparentHugePages &&
3370 !UseHugeTLBFS &&
3371 !UseSHM) {
3372 // Not using large pages.
3373 return;
3374 }
3375
3376 if (!FLAG_IS_DEFAULT(UseLargePages) && !UseLargePages) {
3377 // The user explicitly turned off large pages.
3378 // Ignore the rest of the large pages flags.
3379 UseTransparentHugePages = false;
3380 UseHugeTLBFS = false;
3381 UseSHM = false;
3382 return;
3383 }
3384
3385 size_t large_page_size = Linux::setup_large_page_size();
3386 UseLargePages = Linux::setup_large_page_type(large_page_size);
3387
3388 set_coredump_filter();
3389 }
3390
3391 #ifndef SHM_HUGETLB
3392 #define SHM_HUGETLB 04000
3393 #endif
3394
3395 char* os::Linux::reserve_memory_special_shm(size_t bytes, size_t alignment,
3396 char* req_addr, bool exec) {
3397 // "exec" is passed in but not used. Creating the shared image for
3398 // the code cache doesn't have an SHM_X executable permission to check.
3399 assert(UseLargePages && UseSHM, "only for SHM large pages");
3400 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3401
3402 if (!is_size_aligned(bytes, os::large_page_size()) || alignment > os::large_page_size()) {
3403 return NULL; // Fallback to small pages.
3404 }
3405
3406 key_t key = IPC_PRIVATE;
3407 char *addr;
3408
3409 bool warn_on_failure = UseLargePages &&
3410 (!FLAG_IS_DEFAULT(UseLargePages) ||
3411 !FLAG_IS_DEFAULT(UseSHM) ||
3412 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3413 char msg[128];
3414
3415 // Create a large shared memory region to attach to based on size.
3416 // Currently, size is the total size of the heap
3417 int shmid = shmget(key, bytes, SHM_HUGETLB|IPC_CREAT|SHM_R|SHM_W);
3418 if (shmid == -1) {
3419 // Possible reasons for shmget failure:
3420 // 1. shmmax is too small for Java heap.
3421 // > check shmmax value: cat /proc/sys/kernel/shmmax
3422 // > increase shmmax value: echo "0xffffffff" > /proc/sys/kernel/shmmax
3423 // 2. not enough large page memory.
3424 // > check available large pages: cat /proc/meminfo
3425 // > increase amount of large pages:
3426 // echo new_value > /proc/sys/vm/nr_hugepages
3427 // Note 1: different Linux may use different name for this property,
3428 // e.g. on Redhat AS-3 it is "hugetlb_pool".
3429 // Note 2: it's possible there's enough physical memory available but
3430 // they are so fragmented after a long run that they can't
3431 // coalesce into large pages. Try to reserve large pages when
3432 // the system is still "fresh".
3433 if (warn_on_failure) {
3434 jio_snprintf(msg, sizeof(msg), "Failed to reserve shared memory (errno = %d).", errno);
3435 warning("%s", msg);
3436 }
3437 return NULL;
3438 }
3439
3440 // attach to the region
3441 addr = (char*)shmat(shmid, req_addr, 0);
3442 int err = errno;
3443
3444 // Remove shmid. If shmat() is successful, the actual shared memory segment
3445 // will be deleted when it's detached by shmdt() or when the process
3446 // terminates. If shmat() is not successful this will remove the shared
3447 // segment immediately.
3448 shmctl(shmid, IPC_RMID, NULL);
3449
3450 if ((intptr_t)addr == -1) {
3451 if (warn_on_failure) {
3452 jio_snprintf(msg, sizeof(msg), "Failed to attach shared memory (errno = %d).", err);
3453 warning("%s", msg);
3454 }
3455 return NULL;
3456 }
3457
3458 return addr;
3459 }
3460
3461 static void warn_on_large_pages_failure(char* req_addr, size_t bytes,
3462 int error) {
3463 assert(error == ENOMEM, "Only expect to fail if no memory is available");
3464
3465 bool warn_on_failure = UseLargePages &&
3466 (!FLAG_IS_DEFAULT(UseLargePages) ||
3467 !FLAG_IS_DEFAULT(UseHugeTLBFS) ||
3468 !FLAG_IS_DEFAULT(LargePageSizeInBytes));
3469
3470 if (warn_on_failure) {
3471 char msg[128];
3472 jio_snprintf(msg, sizeof(msg), "Failed to reserve large pages memory req_addr: "
3473 PTR_FORMAT " bytes: " SIZE_FORMAT " (errno = %d).", req_addr, bytes, error);
3474 warning("%s", msg);
3475 }
3476 }
3477
3478 char* os::Linux::reserve_memory_special_huge_tlbfs_only(size_t bytes,
3479 char* req_addr,
3480 bool exec) {
3481 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3482 assert(is_size_aligned(bytes, os::large_page_size()), "Unaligned size");
3483 assert(is_ptr_aligned(req_addr, os::large_page_size()), "Unaligned address");
3484
3485 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3486 char* addr = (char*)::mmap(req_addr, bytes, prot,
3487 MAP_PRIVATE|MAP_ANONYMOUS|MAP_HUGETLB,
3488 -1, 0);
3489
3490 if (addr == MAP_FAILED) {
3491 warn_on_large_pages_failure(req_addr, bytes, errno);
3492 return NULL;
3493 }
3494
3495 assert(is_ptr_aligned(addr, os::large_page_size()), "Must be");
3496
3497 return addr;
3498 }
3499
3500 char* os::Linux::reserve_memory_special_huge_tlbfs_mixed(size_t bytes,
3501 size_t alignment,
3502 char* req_addr,
3503 bool exec) {
3504 size_t large_page_size = os::large_page_size();
3505
3506 assert(bytes >= large_page_size, "Shouldn't allocate large pages for small sizes");
3507
3508 // Allocate small pages.
3509
3510 char* start;
3511 if (req_addr != NULL) {
3512 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3513 assert(is_size_aligned(bytes, alignment), "Must be");
3514 start = os::reserve_memory(bytes, req_addr);
3515 assert(start == NULL || start == req_addr, "Must be");
3516 } else {
3517 start = os::reserve_memory_aligned(bytes, alignment);
3518 }
3519
3520 if (start == NULL) {
3521 return NULL;
3522 }
3523
3524 assert(is_ptr_aligned(start, alignment), "Must be");
3525
3526 if (MemTracker::tracking_level() > NMT_minimal) {
3527 // os::reserve_memory_special will record this memory area.
3528 // Need to release it here to prevent overlapping reservations.
3529 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3530 tkr.record((address)start, bytes);
3531 }
3532
3533 char* end = start + bytes;
3534
3535 // Find the regions of the allocated chunk that can be promoted to large pages.
3536 char* lp_start = (char*)align_ptr_up(start, large_page_size);
3537 char* lp_end = (char*)align_ptr_down(end, large_page_size);
3538
3539 size_t lp_bytes = lp_end - lp_start;
3540
3541 assert(is_size_aligned(lp_bytes, large_page_size), "Must be");
3542
3543 if (lp_bytes == 0) {
3544 // The mapped region doesn't even span the start and the end of a large page.
3545 // Fall back to allocate a non-special area.
3546 ::munmap(start, end - start);
3547 return NULL;
3548 }
3549
3550 int prot = exec ? PROT_READ|PROT_WRITE|PROT_EXEC : PROT_READ|PROT_WRITE;
3551
3552
3553 void* result;
3554
3555 if (start != lp_start) {
3556 result = ::mmap(start, lp_start - start, prot,
3557 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3558 -1, 0);
3559 if (result == MAP_FAILED) {
3560 ::munmap(lp_start, end - lp_start);
3561 return NULL;
3562 }
3563 }
3564
3565 result = ::mmap(lp_start, lp_bytes, prot,
3566 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED|MAP_HUGETLB,
3567 -1, 0);
3568 if (result == MAP_FAILED) {
3569 warn_on_large_pages_failure(req_addr, bytes, errno);
3570 // If the mmap above fails, the large pages region will be unmapped and we
3571 // have regions before and after with small pages. Release these regions.
3572 //
3573 // | mapped | unmapped | mapped |
3574 // ^ ^ ^ ^
3575 // start lp_start lp_end end
3576 //
3577 ::munmap(start, lp_start - start);
3578 ::munmap(lp_end, end - lp_end);
3579 return NULL;
3580 }
3581
3582 if (lp_end != end) {
3583 result = ::mmap(lp_end, end - lp_end, prot,
3584 MAP_PRIVATE|MAP_ANONYMOUS|MAP_FIXED,
3585 -1, 0);
3586 if (result == MAP_FAILED) {
3587 ::munmap(start, lp_end - start);
3588 return NULL;
3589 }
3590 }
3591
3592 return start;
3593 }
3594
3595 char* os::Linux::reserve_memory_special_huge_tlbfs(size_t bytes,
3596 size_t alignment,
3597 char* req_addr,
3598 bool exec) {
3599 assert(UseLargePages && UseHugeTLBFS, "only for Huge TLBFS large pages");
3600 assert(is_ptr_aligned(req_addr, alignment), "Must be");
3601 assert(is_power_of_2(alignment), "Must be");
3602 assert(is_power_of_2(os::large_page_size()), "Must be");
3603 assert(bytes >= os::large_page_size(), "Shouldn't allocate large pages for small sizes");
3604
3605 if (is_size_aligned(bytes, os::large_page_size()) && alignment <= os::large_page_size()) {
3606 return reserve_memory_special_huge_tlbfs_only(bytes, req_addr, exec);
3607 } else {
3608 return reserve_memory_special_huge_tlbfs_mixed(bytes, alignment, req_addr, exec);
3609 }
3610 }
3611
3612 char* os::reserve_memory_special(size_t bytes, size_t alignment,
3613 char* req_addr, bool exec) {
3614 assert(UseLargePages, "only for large pages");
3615
3616 char* addr;
3617 if (UseSHM) {
3618 addr = os::Linux::reserve_memory_special_shm(bytes, alignment, req_addr, exec);
3619 } else {
3620 assert(UseHugeTLBFS, "must be");
3621 addr = os::Linux::reserve_memory_special_huge_tlbfs(bytes, alignment, req_addr, exec);
3622 }
3623
3624 if (addr != NULL) {
3625 if (UseNUMAInterleaving) {
3626 numa_make_global(addr, bytes);
3627 }
3628
3629 // The memory is committed
3630 MemTracker::record_virtual_memory_reserve_and_commit((address)addr, bytes, CALLER_PC);
3631 }
3632
3633 return addr;
3634 }
3635
3636 bool os::Linux::release_memory_special_shm(char* base, size_t bytes) {
3637 // detaching the SHM segment will also delete it, see reserve_memory_special_shm()
3638 return shmdt(base) == 0;
3639 }
3640
3641 bool os::Linux::release_memory_special_huge_tlbfs(char* base, size_t bytes) {
3642 return pd_release_memory(base, bytes);
3643 }
3644
3645 bool os::release_memory_special(char* base, size_t bytes) {
3646 bool res;
3647 if (MemTracker::tracking_level() > NMT_minimal) {
3648 Tracker tkr = MemTracker::get_virtual_memory_release_tracker();
3649 res = os::Linux::release_memory_special_impl(base, bytes);
3650 if (res) {
3651 tkr.record((address)base, bytes);
3652 }
3653
3654 } else {
3655 res = os::Linux::release_memory_special_impl(base, bytes);
3656 }
3657 return res;
3658 }
3659
3660 bool os::Linux::release_memory_special_impl(char* base, size_t bytes) {
3661 assert(UseLargePages, "only for large pages");
3662 bool res;
3663
3664 if (UseSHM) {
3665 res = os::Linux::release_memory_special_shm(base, bytes);
3666 } else {
3667 assert(UseHugeTLBFS, "must be");
3668 res = os::Linux::release_memory_special_huge_tlbfs(base, bytes);
3669 }
3670 return res;
3671 }
3672
3673 size_t os::large_page_size() {
3674 return _large_page_size;
3675 }
3676
3677 // With SysV SHM the entire memory region must be allocated as shared
3678 // memory.
3679 // HugeTLBFS allows application to commit large page memory on demand.
3680 // However, when committing memory with HugeTLBFS fails, the region
3681 // that was supposed to be committed will lose the old reservation
3682 // and allow other threads to steal that memory region. Because of this
3683 // behavior we can't commit HugeTLBFS memory.
3684 bool os::can_commit_large_page_memory() {
3685 return UseTransparentHugePages;
3686 }
3687
3688 bool os::can_execute_large_page_memory() {
3689 return UseTransparentHugePages || UseHugeTLBFS;
3690 }
3691
3692 // Reserve memory at an arbitrary address, only if that area is
3693 // available (and not reserved for something else).
3694
3695 char* os::pd_attempt_reserve_memory_at(size_t bytes, char* requested_addr) {
3696 const int max_tries = 10;
3697 char* base[max_tries];
3698 size_t size[max_tries];
3699 const size_t gap = 0x000000;
3700
3701 // Assert only that the size is a multiple of the page size, since
3702 // that's all that mmap requires, and since that's all we really know
3703 // about at this low abstraction level. If we need higher alignment,
3704 // we can either pass an alignment to this method or verify alignment
3705 // in one of the methods further up the call chain. See bug 5044738.
3706 assert(bytes % os::vm_page_size() == 0, "reserving unexpected size block");
3707
3708 // Repeatedly allocate blocks until the block is allocated at the
3709 // right spot. Give up after max_tries. Note that reserve_memory() will
3710 // automatically update _highest_vm_reserved_address if the call is
3711 // successful. The variable tracks the highest memory address every reserved
3712 // by JVM. It is used to detect heap-stack collision if running with
3713 // fixed-stack LinuxThreads. Because here we may attempt to reserve more
3714 // space than needed, it could confuse the collision detecting code. To
3715 // solve the problem, save current _highest_vm_reserved_address and
3716 // calculate the correct value before return.
3717 address old_highest = _highest_vm_reserved_address;
3718
3719 // Linux mmap allows caller to pass an address as hint; give it a try first,
3720 // if kernel honors the hint then we can return immediately.
3721 char * addr = anon_mmap(requested_addr, bytes, false);
3722 if (addr == requested_addr) {
3723 return requested_addr;
3724 }
3725
3726 if (addr != NULL) {
3727 // mmap() is successful but it fails to reserve at the requested address
3728 anon_munmap(addr, bytes);
3729 }
3730
3731 int i;
3732 for (i = 0; i < max_tries; ++i) {
3733 base[i] = reserve_memory(bytes);
3734
3735 if (base[i] != NULL) {
3736 // Is this the block we wanted?
3737 if (base[i] == requested_addr) {
3738 size[i] = bytes;
3739 break;
3740 }
3741
3742 // Does this overlap the block we wanted? Give back the overlapped
3743 // parts and try again.
3744
3745 size_t top_overlap = requested_addr + (bytes + gap) - base[i];
3746 if (top_overlap >= 0 && top_overlap < bytes) {
3747 unmap_memory(base[i], top_overlap);
3748 base[i] += top_overlap;
3749 size[i] = bytes - top_overlap;
3750 } else {
3751 size_t bottom_overlap = base[i] + bytes - requested_addr;
3752 if (bottom_overlap >= 0 && bottom_overlap < bytes) {
3753 unmap_memory(requested_addr, bottom_overlap);
3754 size[i] = bytes - bottom_overlap;
3755 } else {
3756 size[i] = bytes;
3757 }
3758 }
3759 }
3760 }
3761
3762 // Give back the unused reserved pieces.
3763
3764 for (int j = 0; j < i; ++j) {
3765 if (base[j] != NULL) {
3766 unmap_memory(base[j], size[j]);
3767 }
3768 }
3769
3770 if (i < max_tries) {
3771 _highest_vm_reserved_address = MAX2(old_highest, (address)requested_addr + bytes);
3772 return requested_addr;
3773 } else {
3774 _highest_vm_reserved_address = old_highest;
3775 return NULL;
3776 }
3777 }
3778
3779 size_t os::read(int fd, void *buf, unsigned int nBytes) {
3780 return ::read(fd, buf, nBytes);
3781 }
3782
3783 size_t os::read_at(int fd, void *buf, unsigned int nBytes, jlong offset) {
3784 return ::pread(fd, buf, nBytes, offset);
3785 }
3786
3787 // Short sleep, direct OS call.
3788 //
3789 // Note: certain versions of Linux CFS scheduler (since 2.6.23) do not guarantee
3790 // sched_yield(2) will actually give up the CPU:
3791 //
3792 // * Alone on this pariticular CPU, keeps running.
3793 // * Before the introduction of "skip_buddy" with "compat_yield" disabled
3794 // (pre 2.6.39).
3795 //
3796 // So calling this with 0 is an alternative.
3797 //
3798 void os::naked_short_sleep(jlong ms) {
3799 struct timespec req;
3800
3801 assert(ms < 1000, "Un-interruptable sleep, short time use only");
3802 req.tv_sec = 0;
3803 if (ms > 0) {
3804 req.tv_nsec = (ms % 1000) * 1000000;
3805 } else {
3806 req.tv_nsec = 1;
3807 }
3808
3809 nanosleep(&req, NULL);
3810
3811 return;
3812 }
3813
3814 // Sleep forever; naked call to OS-specific sleep; use with CAUTION
3815 void os::infinite_sleep() {
3816 while (true) { // sleep forever ...
3817 ::sleep(100); // ... 100 seconds at a time
3818 }
3819 }
3820
3821 // Used to convert frequent JVM_Yield() to nops
3822 bool os::dont_yield() {
3823 return DontYieldALot;
3824 }
3825
3826 void os::naked_yield() {
3827 sched_yield();
3828 }
3829
3830 ////////////////////////////////////////////////////////////////////////////////
3831 // thread priority support
3832
3833 // Note: Normal Linux applications are run with SCHED_OTHER policy. SCHED_OTHER
3834 // only supports dynamic priority, static priority must be zero. For real-time
3835 // applications, Linux supports SCHED_RR which allows static priority (1-99).
3836 // However, for large multi-threaded applications, SCHED_RR is not only slower
3837 // than SCHED_OTHER, but also very unstable (my volano tests hang hard 4 out
3838 // of 5 runs - Sep 2005).
3839 //
3840 // The following code actually changes the niceness of kernel-thread/LWP. It
3841 // has an assumption that setpriority() only modifies one kernel-thread/LWP,
3842 // not the entire user process, and user level threads are 1:1 mapped to kernel
3843 // threads. It has always been the case, but could change in the future. For
3844 // this reason, the code should not be used as default (ThreadPriorityPolicy=0).
3845 // It is only used when ThreadPriorityPolicy=1 and requires root privilege.
3846
3847 int os::java_to_os_priority[CriticalPriority + 1] = {
3848 19, // 0 Entry should never be used
3849
3850 4, // 1 MinPriority
3851 3, // 2
3852 2, // 3
3853
3854 1, // 4
3855 0, // 5 NormPriority
3856 -1, // 6
3857
3858 -2, // 7
3859 -3, // 8
3860 -4, // 9 NearMaxPriority
3861
3862 -5, // 10 MaxPriority
3863
3864 -5 // 11 CriticalPriority
3865 };
3866
3867 static int prio_init() {
3868 if (ThreadPriorityPolicy == 1) {
3869 // Only root can raise thread priority. Don't allow ThreadPriorityPolicy=1
3870 // if effective uid is not root. Perhaps, a more elegant way of doing
3871 // this is to test CAP_SYS_NICE capability, but that will require libcap.so
3872 if (geteuid() != 0) {
3873 if (!FLAG_IS_DEFAULT(ThreadPriorityPolicy)) {
3874 warning("-XX:ThreadPriorityPolicy requires root privilege on Linux");
3875 }
3876 ThreadPriorityPolicy = 0;
3877 }
3878 }
3879 if (UseCriticalJavaThreadPriority) {
3880 os::java_to_os_priority[MaxPriority] = os::java_to_os_priority[CriticalPriority];
3881 }
3882 return 0;
3883 }
3884
3885 OSReturn os::set_native_priority(Thread* thread, int newpri) {
3886 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) return OS_OK;
3887
3888 int ret = setpriority(PRIO_PROCESS, thread->osthread()->thread_id(), newpri);
3889 return (ret == 0) ? OS_OK : OS_ERR;
3890 }
3891
3892 OSReturn os::get_native_priority(const Thread* const thread,
3893 int *priority_ptr) {
3894 if (!UseThreadPriorities || ThreadPriorityPolicy == 0) {
3895 *priority_ptr = java_to_os_priority[NormPriority];
3896 return OS_OK;
3897 }
3898
3899 errno = 0;
3900 *priority_ptr = getpriority(PRIO_PROCESS, thread->osthread()->thread_id());
3901 return (*priority_ptr != -1 || errno == 0 ? OS_OK : OS_ERR);
3902 }
3903
3904 // Hint to the underlying OS that a task switch would not be good.
3905 // Void return because it's a hint and can fail.
3906 void os::hint_no_preempt() {}
3907
3908 ////////////////////////////////////////////////////////////////////////////////
3909 // suspend/resume support
3910
3911 // the low-level signal-based suspend/resume support is a remnant from the
3912 // old VM-suspension that used to be for java-suspension, safepoints etc,
3913 // within hotspot. Now there is a single use-case for this:
3914 // - calling get_thread_pc() on the VMThread by the flat-profiler task
3915 // that runs in the watcher thread.
3916 // The remaining code is greatly simplified from the more general suspension
3917 // code that used to be used.
3918 //
3919 // The protocol is quite simple:
3920 // - suspend:
3921 // - sends a signal to the target thread
3922 // - polls the suspend state of the osthread using a yield loop
3923 // - target thread signal handler (SR_handler) sets suspend state
3924 // and blocks in sigsuspend until continued
3925 // - resume:
3926 // - sets target osthread state to continue
3927 // - sends signal to end the sigsuspend loop in the SR_handler
3928 //
3929 // Note that the SR_lock plays no role in this suspend/resume protocol.
3930
3931 static void resume_clear_context(OSThread *osthread) {
3932 osthread->set_ucontext(NULL);
3933 osthread->set_siginfo(NULL);
3934 }
3935
3936 static void suspend_save_context(OSThread *osthread, siginfo_t* siginfo,
3937 ucontext_t* context) {
3938 osthread->set_ucontext(context);
3939 osthread->set_siginfo(siginfo);
3940 }
3941
3942 // Handler function invoked when a thread's execution is suspended or
3943 // resumed. We have to be careful that only async-safe functions are
3944 // called here (Note: most pthread functions are not async safe and
3945 // should be avoided.)
3946 //
3947 // Note: sigwait() is a more natural fit than sigsuspend() from an
3948 // interface point of view, but sigwait() prevents the signal hander
3949 // from being run. libpthread would get very confused by not having
3950 // its signal handlers run and prevents sigwait()'s use with the
3951 // mutex granting granting signal.
3952 //
3953 // Currently only ever called on the VMThread and JavaThreads (PC sampling)
3954 //
3955 static void SR_handler(int sig, siginfo_t* siginfo, ucontext_t* context) {
3956 // Save and restore errno to avoid confusing native code with EINTR
3957 // after sigsuspend.
3958 int old_errno = errno;
3959
3960 Thread* thread = Thread::current();
3961 OSThread* osthread = thread->osthread();
3962 assert(thread->is_VM_thread() || thread->is_Java_thread(), "Must be VMThread or JavaThread");
3963
3964 os::SuspendResume::State current = osthread->sr.state();
3965 if (current == os::SuspendResume::SR_SUSPEND_REQUEST) {
3966 suspend_save_context(osthread, siginfo, context);
3967
3968 // attempt to switch the state, we assume we had a SUSPEND_REQUEST
3969 os::SuspendResume::State state = osthread->sr.suspended();
3970 if (state == os::SuspendResume::SR_SUSPENDED) {
3971 sigset_t suspend_set; // signals for sigsuspend()
3972
3973 // get current set of blocked signals and unblock resume signal
3974 pthread_sigmask(SIG_BLOCK, NULL, &suspend_set);
3975 sigdelset(&suspend_set, SR_signum);
3976
3977 sr_semaphore.signal();
3978 // wait here until we are resumed
3979 while (1) {
3980 sigsuspend(&suspend_set);
3981
3982 os::SuspendResume::State result = osthread->sr.running();
3983 if (result == os::SuspendResume::SR_RUNNING) {
3984 sr_semaphore.signal();
3985 break;
3986 }
3987 }
3988
3989 } else if (state == os::SuspendResume::SR_RUNNING) {
3990 // request was cancelled, continue
3991 } else {
3992 ShouldNotReachHere();
3993 }
3994
3995 resume_clear_context(osthread);
3996 } else if (current == os::SuspendResume::SR_RUNNING) {
3997 // request was cancelled, continue
3998 } else if (current == os::SuspendResume::SR_WAKEUP_REQUEST) {
3999 // ignore
4000 } else {
4001 // ignore
4002 }
4003
4004 errno = old_errno;
4005 }
4006
4007
4008 static int SR_initialize() {
4009 struct sigaction act;
4010 char *s;
4011 // Get signal number to use for suspend/resume
4012 if ((s = ::getenv("_JAVA_SR_SIGNUM")) != 0) {
4013 int sig = ::strtol(s, 0, 10);
4014 if (sig > 0 || sig < _NSIG) {
4015 SR_signum = sig;
4016 }
4017 }
4018
4019 assert(SR_signum > SIGSEGV && SR_signum > SIGBUS,
4020 "SR_signum must be greater than max(SIGSEGV, SIGBUS), see 4355769");
4021
4022 sigemptyset(&SR_sigset);
4023 sigaddset(&SR_sigset, SR_signum);
4024
4025 // Set up signal handler for suspend/resume
4026 act.sa_flags = SA_RESTART|SA_SIGINFO;
4027 act.sa_handler = (void (*)(int)) SR_handler;
4028
4029 // SR_signum is blocked by default.
4030 // 4528190 - We also need to block pthread restart signal (32 on all
4031 // supported Linux platforms). Note that LinuxThreads need to block
4032 // this signal for all threads to work properly. So we don't have
4033 // to use hard-coded signal number when setting up the mask.
4034 pthread_sigmask(SIG_BLOCK, NULL, &act.sa_mask);
4035
4036 if (sigaction(SR_signum, &act, 0) == -1) {
4037 return -1;
4038 }
4039
4040 // Save signal flag
4041 os::Linux::set_our_sigflags(SR_signum, act.sa_flags);
4042 return 0;
4043 }
4044
4045 static int sr_notify(OSThread* osthread) {
4046 int status = pthread_kill(osthread->pthread_id(), SR_signum);
4047 assert_status(status == 0, status, "pthread_kill");
4048 return status;
4049 }
4050
4051 // "Randomly" selected value for how long we want to spin
4052 // before bailing out on suspending a thread, also how often
4053 // we send a signal to a thread we want to resume
4054 static const int RANDOMLY_LARGE_INTEGER = 1000000;
4055 static const int RANDOMLY_LARGE_INTEGER2 = 100;
4056
4057 // returns true on success and false on error - really an error is fatal
4058 // but this seems the normal response to library errors
4059 static bool do_suspend(OSThread* osthread) {
4060 assert(osthread->sr.is_running(), "thread should be running");
4061 assert(!sr_semaphore.trywait(), "semaphore has invalid state");
4062
4063 // mark as suspended and send signal
4064 if (osthread->sr.request_suspend() != os::SuspendResume::SR_SUSPEND_REQUEST) {
4065 // failed to switch, state wasn't running?
4066 ShouldNotReachHere();
4067 return false;
4068 }
4069
4070 if (sr_notify(osthread) != 0) {
4071 ShouldNotReachHere();
4072 }
4073
4074 // managed to send the signal and switch to SUSPEND_REQUEST, now wait for SUSPENDED
4075 while (true) {
4076 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4077 break;
4078 } else {
4079 // timeout
4080 os::SuspendResume::State cancelled = osthread->sr.cancel_suspend();
4081 if (cancelled == os::SuspendResume::SR_RUNNING) {
4082 return false;
4083 } else if (cancelled == os::SuspendResume::SR_SUSPENDED) {
4084 // make sure that we consume the signal on the semaphore as well
4085 sr_semaphore.wait();
4086 break;
4087 } else {
4088 ShouldNotReachHere();
4089 return false;
4090 }
4091 }
4092 }
4093
4094 guarantee(osthread->sr.is_suspended(), "Must be suspended");
4095 return true;
4096 }
4097
4098 static void do_resume(OSThread* osthread) {
4099 assert(osthread->sr.is_suspended(), "thread should be suspended");
4100 assert(!sr_semaphore.trywait(), "invalid semaphore state");
4101
4102 if (osthread->sr.request_wakeup() != os::SuspendResume::SR_WAKEUP_REQUEST) {
4103 // failed to switch to WAKEUP_REQUEST
4104 ShouldNotReachHere();
4105 return;
4106 }
4107
4108 while (true) {
4109 if (sr_notify(osthread) == 0) {
4110 if (sr_semaphore.timedwait(0, 2 * NANOSECS_PER_MILLISEC)) {
4111 if (osthread->sr.is_running()) {
4112 return;
4113 }
4114 }
4115 } else {
4116 ShouldNotReachHere();
4117 }
4118 }
4119
4120 guarantee(osthread->sr.is_running(), "Must be running!");
4121 }
4122
4123 ///////////////////////////////////////////////////////////////////////////////////
4124 // signal handling (except suspend/resume)
4125
4126 // This routine may be used by user applications as a "hook" to catch signals.
4127 // The user-defined signal handler must pass unrecognized signals to this
4128 // routine, and if it returns true (non-zero), then the signal handler must
4129 // return immediately. If the flag "abort_if_unrecognized" is true, then this
4130 // routine will never retun false (zero), but instead will execute a VM panic
4131 // routine kill the process.
4132 //
4133 // If this routine returns false, it is OK to call it again. This allows
4134 // the user-defined signal handler to perform checks either before or after
4135 // the VM performs its own checks. Naturally, the user code would be making
4136 // a serious error if it tried to handle an exception (such as a null check
4137 // or breakpoint) that the VM was generating for its own correct operation.
4138 //
4139 // This routine may recognize any of the following kinds of signals:
4140 // SIGBUS, SIGSEGV, SIGILL, SIGFPE, SIGQUIT, SIGPIPE, SIGXFSZ, SIGUSR1.
4141 // It should be consulted by handlers for any of those signals.
4142 //
4143 // The caller of this routine must pass in the three arguments supplied
4144 // to the function referred to in the "sa_sigaction" (not the "sa_handler")
4145 // field of the structure passed to sigaction(). This routine assumes that
4146 // the sa_flags field passed to sigaction() includes SA_SIGINFO and SA_RESTART.
4147 //
4148 // Note that the VM will print warnings if it detects conflicting signal
4149 // handlers, unless invoked with the option "-XX:+AllowUserSignalHandlers".
4150 //
4151 extern "C" JNIEXPORT int JVM_handle_linux_signal(int signo,
4152 siginfo_t* siginfo,
4153 void* ucontext,
4154 int abort_if_unrecognized);
4155
4156 void signalHandler(int sig, siginfo_t* info, void* uc) {
4157 assert(info != NULL && uc != NULL, "it must be old kernel");
4158 int orig_errno = errno; // Preserve errno value over signal handler.
4159 JVM_handle_linux_signal(sig, info, uc, true);
4160 errno = orig_errno;
4161 }
4162
4163
4164 // This boolean allows users to forward their own non-matching signals
4165 // to JVM_handle_linux_signal, harmlessly.
4166 bool os::Linux::signal_handlers_are_installed = false;
4167
4168 // For signal-chaining
4169 struct sigaction os::Linux::sigact[MAXSIGNUM];
4170 unsigned int os::Linux::sigs = 0;
4171 bool os::Linux::libjsig_is_loaded = false;
4172 typedef struct sigaction *(*get_signal_t)(int);
4173 get_signal_t os::Linux::get_signal_action = NULL;
4174
4175 struct sigaction* os::Linux::get_chained_signal_action(int sig) {
4176 struct sigaction *actp = NULL;
4177
4178 if (libjsig_is_loaded) {
4179 // Retrieve the old signal handler from libjsig
4180 actp = (*get_signal_action)(sig);
4181 }
4182 if (actp == NULL) {
4183 // Retrieve the preinstalled signal handler from jvm
4184 actp = get_preinstalled_handler(sig);
4185 }
4186
4187 return actp;
4188 }
4189
4190 static bool call_chained_handler(struct sigaction *actp, int sig,
4191 siginfo_t *siginfo, void *context) {
4192 // Call the old signal handler
4193 if (actp->sa_handler == SIG_DFL) {
4194 // It's more reasonable to let jvm treat it as an unexpected exception
4195 // instead of taking the default action.
4196 return false;
4197 } else if (actp->sa_handler != SIG_IGN) {
4198 if ((actp->sa_flags & SA_NODEFER) == 0) {
4199 // automaticlly block the signal
4200 sigaddset(&(actp->sa_mask), sig);
4201 }
4202
4203 sa_handler_t hand;
4204 sa_sigaction_t sa;
4205 bool siginfo_flag_set = (actp->sa_flags & SA_SIGINFO) != 0;
4206 // retrieve the chained handler
4207 if (siginfo_flag_set) {
4208 sa = actp->sa_sigaction;
4209 } else {
4210 hand = actp->sa_handler;
4211 }
4212
4213 if ((actp->sa_flags & SA_RESETHAND) != 0) {
4214 actp->sa_handler = SIG_DFL;
4215 }
4216
4217 // try to honor the signal mask
4218 sigset_t oset;
4219 pthread_sigmask(SIG_SETMASK, &(actp->sa_mask), &oset);
4220
4221 // call into the chained handler
4222 if (siginfo_flag_set) {
4223 (*sa)(sig, siginfo, context);
4224 } else {
4225 (*hand)(sig);
4226 }
4227
4228 // restore the signal mask
4229 pthread_sigmask(SIG_SETMASK, &oset, 0);
4230 }
4231 // Tell jvm's signal handler the signal is taken care of.
4232 return true;
4233 }
4234
4235 bool os::Linux::chained_handler(int sig, siginfo_t* siginfo, void* context) {
4236 bool chained = false;
4237 // signal-chaining
4238 if (UseSignalChaining) {
4239 struct sigaction *actp = get_chained_signal_action(sig);
4240 if (actp != NULL) {
4241 chained = call_chained_handler(actp, sig, siginfo, context);
4242 }
4243 }
4244 return chained;
4245 }
4246
4247 struct sigaction* os::Linux::get_preinstalled_handler(int sig) {
4248 if ((((unsigned int)1 << sig) & sigs) != 0) {
4249 return &sigact[sig];
4250 }
4251 return NULL;
4252 }
4253
4254 void os::Linux::save_preinstalled_handler(int sig, struct sigaction& oldAct) {
4255 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4256 sigact[sig] = oldAct;
4257 sigs |= (unsigned int)1 << sig;
4258 }
4259
4260 // for diagnostic
4261 int os::Linux::sigflags[MAXSIGNUM];
4262
4263 int os::Linux::get_our_sigflags(int sig) {
4264 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4265 return sigflags[sig];
4266 }
4267
4268 void os::Linux::set_our_sigflags(int sig, int flags) {
4269 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4270 sigflags[sig] = flags;
4271 }
4272
4273 void os::Linux::set_signal_handler(int sig, bool set_installed) {
4274 // Check for overwrite.
4275 struct sigaction oldAct;
4276 sigaction(sig, (struct sigaction*)NULL, &oldAct);
4277
4278 void* oldhand = oldAct.sa_sigaction
4279 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4280 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4281 if (oldhand != CAST_FROM_FN_PTR(void*, SIG_DFL) &&
4282 oldhand != CAST_FROM_FN_PTR(void*, SIG_IGN) &&
4283 oldhand != CAST_FROM_FN_PTR(void*, (sa_sigaction_t)signalHandler)) {
4284 if (AllowUserSignalHandlers || !set_installed) {
4285 // Do not overwrite; user takes responsibility to forward to us.
4286 return;
4287 } else if (UseSignalChaining) {
4288 // save the old handler in jvm
4289 save_preinstalled_handler(sig, oldAct);
4290 // libjsig also interposes the sigaction() call below and saves the
4291 // old sigaction on it own.
4292 } else {
4293 fatal(err_msg("Encountered unexpected pre-existing sigaction handler "
4294 "%#lx for signal %d.", (long)oldhand, sig));
4295 }
4296 }
4297
4298 struct sigaction sigAct;
4299 sigfillset(&(sigAct.sa_mask));
4300 sigAct.sa_handler = SIG_DFL;
4301 if (!set_installed) {
4302 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4303 } else {
4304 sigAct.sa_sigaction = signalHandler;
4305 sigAct.sa_flags = SA_SIGINFO|SA_RESTART;
4306 }
4307 // Save flags, which are set by ours
4308 assert(sig > 0 && sig < MAXSIGNUM, "vm signal out of expected range");
4309 sigflags[sig] = sigAct.sa_flags;
4310
4311 int ret = sigaction(sig, &sigAct, &oldAct);
4312 assert(ret == 0, "check");
4313
4314 void* oldhand2 = oldAct.sa_sigaction
4315 ? CAST_FROM_FN_PTR(void*, oldAct.sa_sigaction)
4316 : CAST_FROM_FN_PTR(void*, oldAct.sa_handler);
4317 assert(oldhand2 == oldhand, "no concurrent signal handler installation");
4318 }
4319
4320 // install signal handlers for signals that HotSpot needs to
4321 // handle in order to support Java-level exception handling.
4322
4323 void os::Linux::install_signal_handlers() {
4324 if (!signal_handlers_are_installed) {
4325 signal_handlers_are_installed = true;
4326
4327 // signal-chaining
4328 typedef void (*signal_setting_t)();
4329 signal_setting_t begin_signal_setting = NULL;
4330 signal_setting_t end_signal_setting = NULL;
4331 begin_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4332 dlsym(RTLD_DEFAULT, "JVM_begin_signal_setting"));
4333 if (begin_signal_setting != NULL) {
4334 end_signal_setting = CAST_TO_FN_PTR(signal_setting_t,
4335 dlsym(RTLD_DEFAULT, "JVM_end_signal_setting"));
4336 get_signal_action = CAST_TO_FN_PTR(get_signal_t,
4337 dlsym(RTLD_DEFAULT, "JVM_get_signal_action"));
4338 libjsig_is_loaded = true;
4339 assert(UseSignalChaining, "should enable signal-chaining");
4340 }
4341 if (libjsig_is_loaded) {
4342 // Tell libjsig jvm is setting signal handlers
4343 (*begin_signal_setting)();
4344 }
4345
4346 set_signal_handler(SIGSEGV, true);
4347 set_signal_handler(SIGPIPE, true);
4348 set_signal_handler(SIGBUS, true);
4349 set_signal_handler(SIGILL, true);
4350 set_signal_handler(SIGFPE, true);
4351 #if defined(PPC64)
4352 set_signal_handler(SIGTRAP, true);
4353 #endif
4354 set_signal_handler(SIGXFSZ, true);
4355
4356 if (libjsig_is_loaded) {
4357 // Tell libjsig jvm finishes setting signal handlers
4358 (*end_signal_setting)();
4359 }
4360
4361 // We don't activate signal checker if libjsig is in place, we trust ourselves
4362 // and if UserSignalHandler is installed all bets are off.
4363 // Log that signal checking is off only if -verbose:jni is specified.
4364 if (CheckJNICalls) {
4365 if (libjsig_is_loaded) {
4366 if (PrintJNIResolving) {
4367 tty->print_cr("Info: libjsig is activated, all active signal checking is disabled");
4368 }
4369 check_signals = false;
4370 }
4371 if (AllowUserSignalHandlers) {
4372 if (PrintJNIResolving) {
4373 tty->print_cr("Info: AllowUserSignalHandlers is activated, all active signal checking is disabled");
4374 }
4375 check_signals = false;
4376 }
4377 }
4378 }
4379 }
4380
4381 // This is the fastest way to get thread cpu time on Linux.
4382 // Returns cpu time (user+sys) for any thread, not only for current.
4383 // POSIX compliant clocks are implemented in the kernels 2.6.16+.
4384 // It might work on 2.6.10+ with a special kernel/glibc patch.
4385 // For reference, please, see IEEE Std 1003.1-2004:
4386 // http://www.unix.org/single_unix_specification
4387
4388 jlong os::Linux::fast_thread_cpu_time(clockid_t clockid) {
4389 struct timespec tp;
4390 int rc = os::Linux::clock_gettime(clockid, &tp);
4391 assert(rc == 0, "clock_gettime is expected to return 0 code");
4392
4393 return (tp.tv_sec * NANOSECS_PER_SEC) + tp.tv_nsec;
4394 }
4395
4396 /////
4397 // glibc on Linux platform uses non-documented flag
4398 // to indicate, that some special sort of signal
4399 // trampoline is used.
4400 // We will never set this flag, and we should
4401 // ignore this flag in our diagnostic
4402 #ifdef SIGNIFICANT_SIGNAL_MASK
4403 #undef SIGNIFICANT_SIGNAL_MASK
4404 #endif
4405 #define SIGNIFICANT_SIGNAL_MASK (~0x04000000)
4406
4407 static const char* get_signal_handler_name(address handler,
4408 char* buf, int buflen) {
4409 int offset;
4410 bool found = os::dll_address_to_library_name(handler, buf, buflen, &offset);
4411 if (found) {
4412 // skip directory names
4413 const char *p1, *p2;
4414 p1 = buf;
4415 size_t len = strlen(os::file_separator());
4416 while ((p2 = strstr(p1, os::file_separator())) != NULL) p1 = p2 + len;
4417 jio_snprintf(buf, buflen, "%s+0x%x", p1, offset);
4418 } else {
4419 jio_snprintf(buf, buflen, PTR_FORMAT, handler);
4420 }
4421 return buf;
4422 }
4423
4424 static void print_signal_handler(outputStream* st, int sig,
4425 char* buf, size_t buflen) {
4426 struct sigaction sa;
4427
4428 sigaction(sig, NULL, &sa);
4429
4430 // See comment for SIGNIFICANT_SIGNAL_MASK define
4431 sa.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4432
4433 st->print("%s: ", os::exception_name(sig, buf, buflen));
4434
4435 address handler = (sa.sa_flags & SA_SIGINFO)
4436 ? CAST_FROM_FN_PTR(address, sa.sa_sigaction)
4437 : CAST_FROM_FN_PTR(address, sa.sa_handler);
4438
4439 if (handler == CAST_FROM_FN_PTR(address, SIG_DFL)) {
4440 st->print("SIG_DFL");
4441 } else if (handler == CAST_FROM_FN_PTR(address, SIG_IGN)) {
4442 st->print("SIG_IGN");
4443 } else {
4444 st->print("[%s]", get_signal_handler_name(handler, buf, buflen));
4445 }
4446
4447 st->print(", sa_mask[0]=");
4448 os::Posix::print_signal_set_short(st, &sa.sa_mask);
4449
4450 address rh = VMError::get_resetted_sighandler(sig);
4451 // May be, handler was resetted by VMError?
4452 if (rh != NULL) {
4453 handler = rh;
4454 sa.sa_flags = VMError::get_resetted_sigflags(sig) & SIGNIFICANT_SIGNAL_MASK;
4455 }
4456
4457 st->print(", sa_flags=");
4458 os::Posix::print_sa_flags(st, sa.sa_flags);
4459
4460 // Check: is it our handler?
4461 if (handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler) ||
4462 handler == CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler)) {
4463 // It is our signal handler
4464 // check for flags, reset system-used one!
4465 if ((int)sa.sa_flags != os::Linux::get_our_sigflags(sig)) {
4466 st->print(
4467 ", flags was changed from " PTR32_FORMAT ", consider using jsig library",
4468 os::Linux::get_our_sigflags(sig));
4469 }
4470 }
4471 st->cr();
4472 }
4473
4474
4475 #define DO_SIGNAL_CHECK(sig) \
4476 do { \
4477 if (!sigismember(&check_signal_done, sig)) { \
4478 os::Linux::check_signal_handler(sig); \
4479 } \
4480 } while (0)
4481
4482 // This method is a periodic task to check for misbehaving JNI applications
4483 // under CheckJNI, we can add any periodic checks here
4484
4485 void os::run_periodic_checks() {
4486 if (check_signals == false) return;
4487
4488 // SEGV and BUS if overridden could potentially prevent
4489 // generation of hs*.log in the event of a crash, debugging
4490 // such a case can be very challenging, so we absolutely
4491 // check the following for a good measure:
4492 DO_SIGNAL_CHECK(SIGSEGV);
4493 DO_SIGNAL_CHECK(SIGILL);
4494 DO_SIGNAL_CHECK(SIGFPE);
4495 DO_SIGNAL_CHECK(SIGBUS);
4496 DO_SIGNAL_CHECK(SIGPIPE);
4497 DO_SIGNAL_CHECK(SIGXFSZ);
4498 #if defined(PPC64)
4499 DO_SIGNAL_CHECK(SIGTRAP);
4500 #endif
4501
4502 // ReduceSignalUsage allows the user to override these handlers
4503 // see comments at the very top and jvm_solaris.h
4504 if (!ReduceSignalUsage) {
4505 DO_SIGNAL_CHECK(SHUTDOWN1_SIGNAL);
4506 DO_SIGNAL_CHECK(SHUTDOWN2_SIGNAL);
4507 DO_SIGNAL_CHECK(SHUTDOWN3_SIGNAL);
4508 DO_SIGNAL_CHECK(BREAK_SIGNAL);
4509 }
4510
4511 DO_SIGNAL_CHECK(SR_signum);
4512 DO_SIGNAL_CHECK(INTERRUPT_SIGNAL);
4513 }
4514
4515 typedef int (*os_sigaction_t)(int, const struct sigaction *, struct sigaction *);
4516
4517 static os_sigaction_t os_sigaction = NULL;
4518
4519 void os::Linux::check_signal_handler(int sig) {
4520 char buf[O_BUFLEN];
4521 address jvmHandler = NULL;
4522
4523
4524 struct sigaction act;
4525 if (os_sigaction == NULL) {
4526 // only trust the default sigaction, in case it has been interposed
4527 os_sigaction = (os_sigaction_t)dlsym(RTLD_DEFAULT, "sigaction");
4528 if (os_sigaction == NULL) return;
4529 }
4530
4531 os_sigaction(sig, (struct sigaction*)NULL, &act);
4532
4533
4534 act.sa_flags &= SIGNIFICANT_SIGNAL_MASK;
4535
4536 address thisHandler = (act.sa_flags & SA_SIGINFO)
4537 ? CAST_FROM_FN_PTR(address, act.sa_sigaction)
4538 : CAST_FROM_FN_PTR(address, act.sa_handler);
4539
4540
4541 switch (sig) {
4542 case SIGSEGV:
4543 case SIGBUS:
4544 case SIGFPE:
4545 case SIGPIPE:
4546 case SIGILL:
4547 case SIGXFSZ:
4548 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)signalHandler);
4549 break;
4550
4551 case SHUTDOWN1_SIGNAL:
4552 case SHUTDOWN2_SIGNAL:
4553 case SHUTDOWN3_SIGNAL:
4554 case BREAK_SIGNAL:
4555 jvmHandler = (address)user_handler();
4556 break;
4557
4558 case INTERRUPT_SIGNAL:
4559 jvmHandler = CAST_FROM_FN_PTR(address, SIG_DFL);
4560 break;
4561
4562 default:
4563 if (sig == SR_signum) {
4564 jvmHandler = CAST_FROM_FN_PTR(address, (sa_sigaction_t)SR_handler);
4565 } else {
4566 return;
4567 }
4568 break;
4569 }
4570
4571 if (thisHandler != jvmHandler) {
4572 tty->print("Warning: %s handler ", exception_name(sig, buf, O_BUFLEN));
4573 tty->print("expected:%s", get_signal_handler_name(jvmHandler, buf, O_BUFLEN));
4574 tty->print_cr(" found:%s", get_signal_handler_name(thisHandler, buf, O_BUFLEN));
4575 // No need to check this sig any longer
4576 sigaddset(&check_signal_done, sig);
4577 // Running under non-interactive shell, SHUTDOWN2_SIGNAL will be reassigned SIG_IGN
4578 if (sig == SHUTDOWN2_SIGNAL && !isatty(fileno(stdin))) {
4579 tty->print_cr("Running in non-interactive shell, %s handler is replaced by shell",
4580 exception_name(sig, buf, O_BUFLEN));
4581 }
4582 } else if(os::Linux::get_our_sigflags(sig) != 0 && (int)act.sa_flags != os::Linux::get_our_sigflags(sig)) {
4583 tty->print("Warning: %s handler flags ", exception_name(sig, buf, O_BUFLEN));
4584 tty->print("expected:" PTR32_FORMAT, os::Linux::get_our_sigflags(sig));
4585 tty->print_cr(" found:" PTR32_FORMAT, act.sa_flags);
4586 // No need to check this sig any longer
4587 sigaddset(&check_signal_done, sig);
4588 }
4589
4590 // Dump all the signal
4591 if (sigismember(&check_signal_done, sig)) {
4592 print_signal_handlers(tty, buf, O_BUFLEN);
4593 }
4594 }
4595
4596 extern void report_error(char* file_name, int line_no, char* title,
4597 char* format, ...);
4598
4599 extern bool signal_name(int signo, char* buf, size_t len);
4600
4601 const char* os::exception_name(int exception_code, char* buf, size_t size) {
4602 if (0 < exception_code && exception_code <= SIGRTMAX) {
4603 // signal
4604 if (!signal_name(exception_code, buf, size)) {
4605 jio_snprintf(buf, size, "SIG%d", exception_code);
4606 }
4607 return buf;
4608 } else {
4609 return NULL;
4610 }
4611 }
4612
4613 // this is called _before_ the most of global arguments have been parsed
4614 void os::init(void) {
4615 char dummy; // used to get a guess on initial stack address
4616 // first_hrtime = gethrtime();
4617
4618 // With LinuxThreads the JavaMain thread pid (primordial thread)
4619 // is different than the pid of the java launcher thread.
4620 // So, on Linux, the launcher thread pid is passed to the VM
4621 // via the sun.java.launcher.pid property.
4622 // Use this property instead of getpid() if it was correctly passed.
4623 // See bug 6351349.
4624 pid_t java_launcher_pid = (pid_t) Arguments::sun_java_launcher_pid();
4625
4626 _initial_pid = (java_launcher_pid > 0) ? java_launcher_pid : getpid();
4627
4628 clock_tics_per_sec = sysconf(_SC_CLK_TCK);
4629
4630 init_random(1234567);
4631
4632 ThreadCritical::initialize();
4633
4634 Linux::set_page_size(sysconf(_SC_PAGESIZE));
4635 if (Linux::page_size() == -1) {
4636 fatal(err_msg("os_linux.cpp: os::init: sysconf failed (%s)",
4637 strerror(errno)));
4638 }
4639 init_page_sizes((size_t) Linux::page_size());
4640
4641 Linux::initialize_system_info();
4642
4643 // main_thread points to the aboriginal thread
4644 Linux::_main_thread = pthread_self();
4645
4646 Linux::clock_init();
4647 initial_time_count = javaTimeNanos();
4648
4649 // pthread_condattr initialization for monotonic clock
4650 int status;
4651 pthread_condattr_t* _condattr = os::Linux::condAttr();
4652 if ((status = pthread_condattr_init(_condattr)) != 0) {
4653 fatal(err_msg("pthread_condattr_init: %s", strerror(status)));
4654 }
4655 // Only set the clock if CLOCK_MONOTONIC is available
4656 if (os::supports_monotonic_clock()) {
4657 if ((status = pthread_condattr_setclock(_condattr, CLOCK_MONOTONIC)) != 0) {
4658 if (status == EINVAL) {
4659 warning("Unable to use monotonic clock with relative timed-waits" \
4660 " - changes to the time-of-day clock may have adverse affects");
4661 } else {
4662 fatal(err_msg("pthread_condattr_setclock: %s", strerror(status)));
4663 }
4664 }
4665 }
4666 // else it defaults to CLOCK_REALTIME
4667
4668 pthread_mutex_init(&dl_mutex, NULL);
4669
4670 // If the pagesize of the VM is greater than 8K determine the appropriate
4671 // number of initial guard pages. The user can change this with the
4672 // command line arguments, if needed.
4673 if (vm_page_size() > (int)Linux::vm_default_page_size()) {
4674 StackYellowPages = 1;
4675 StackRedPages = 1;
4676 StackShadowPages = round_to((StackShadowPages*Linux::vm_default_page_size()), vm_page_size()) / vm_page_size();
4677 }
4678
4679 // retrieve entry point for pthread_setname_np
4680 Linux::_pthread_setname_np =
4681 (int(*)(pthread_t, const char*))dlsym(RTLD_DEFAULT, "pthread_setname_np");
4682
4683 }
4684
4685 // To install functions for atexit system call
4686 extern "C" {
4687 static void perfMemory_exit_helper() {
4688 perfMemory_exit();
4689 }
4690 }
4691
4692 // this is called _after_ the global arguments have been parsed
4693 jint os::init_2(void) {
4694 Linux::fast_thread_clock_init();
4695
4696 // Allocate a single page and mark it as readable for safepoint polling
4697 address polling_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4698 guarantee(polling_page != MAP_FAILED, "os::init_2: failed to allocate polling page");
4699
4700 os::set_polling_page(polling_page);
4701
4702 #ifndef PRODUCT
4703 if (Verbose && PrintMiscellaneous) {
4704 tty->print("[SafePoint Polling address: " INTPTR_FORMAT "]\n",
4705 (intptr_t)polling_page);
4706 }
4707 #endif
4708
4709 if (!UseMembar) {
4710 address mem_serialize_page = (address) ::mmap(NULL, Linux::page_size(), PROT_READ | PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
4711 guarantee(mem_serialize_page != MAP_FAILED, "mmap Failed for memory serialize page");
4712 os::set_memory_serialize_page(mem_serialize_page);
4713
4714 #ifndef PRODUCT
4715 if (Verbose && PrintMiscellaneous) {
4716 tty->print("[Memory Serialize Page address: " INTPTR_FORMAT "]\n",
4717 (intptr_t)mem_serialize_page);
4718 }
4719 #endif
4720 }
4721
4722 // initialize suspend/resume support - must do this before signal_sets_init()
4723 if (SR_initialize() != 0) {
4724 perror("SR_initialize failed");
4725 return JNI_ERR;
4726 }
4727
4728 Linux::signal_sets_init();
4729 Linux::install_signal_handlers();
4730
4731 // Check minimum allowable stack size for thread creation and to initialize
4732 // the java system classes, including StackOverflowError - depends on page
4733 // size. Add a page for compiler2 recursion in main thread.
4734 // Add in 2*BytesPerWord times page size to account for VM stack during
4735 // class initialization depending on 32 or 64 bit VM.
4736 os::Linux::min_stack_allowed = MAX2(os::Linux::min_stack_allowed,
4737 (size_t)(StackYellowPages+StackRedPages+StackShadowPages) * Linux::page_size() +
4738 (2*BytesPerWord COMPILER2_PRESENT(+1)) * Linux::vm_default_page_size());
4739
4740 size_t threadStackSizeInBytes = ThreadStackSize * K;
4741 if (threadStackSizeInBytes != 0 &&
4742 threadStackSizeInBytes < os::Linux::min_stack_allowed) {
4743 tty->print_cr("\nThe stack size specified is too small, "
4744 "Specify at least %dk",
4745 os::Linux::min_stack_allowed/ K);
4746 return JNI_ERR;
4747 }
4748
4749 // Make the stack size a multiple of the page size so that
4750 // the yellow/red zones can be guarded.
4751 JavaThread::set_stack_size_at_create(round_to(threadStackSizeInBytes,
4752 vm_page_size()));
4753
4754 Linux::capture_initial_stack(JavaThread::stack_size_at_create());
4755
4756 #if defined(IA32)
4757 workaround_expand_exec_shield_cs_limit();
4758 #endif
4759
4760 Linux::libpthread_init();
4761 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4762 tty->print_cr("[HotSpot is running with %s, %s(%s)]\n",
4763 Linux::glibc_version(), Linux::libpthread_version(),
4764 Linux::is_floating_stack() ? "floating stack" : "fixed stack");
4765 }
4766
4767 if (UseNUMA) {
4768 if (!Linux::libnuma_init()) {
4769 UseNUMA = false;
4770 } else {
4771 if ((Linux::numa_max_node() < 1)) {
4772 // There's only one node(they start from 0), disable NUMA.
4773 UseNUMA = false;
4774 }
4775 }
4776 // With SHM and HugeTLBFS large pages we cannot uncommit a page, so there's no way
4777 // we can make the adaptive lgrp chunk resizing work. If the user specified
4778 // both UseNUMA and UseLargePages (or UseSHM/UseHugeTLBFS) on the command line - warn and
4779 // disable adaptive resizing.
4780 if (UseNUMA && UseLargePages && !can_commit_large_page_memory()) {
4781 if (FLAG_IS_DEFAULT(UseNUMA)) {
4782 UseNUMA = false;
4783 } else {
4784 if (FLAG_IS_DEFAULT(UseLargePages) &&
4785 FLAG_IS_DEFAULT(UseSHM) &&
4786 FLAG_IS_DEFAULT(UseHugeTLBFS)) {
4787 UseLargePages = false;
4788 } else {
4789 warning("UseNUMA is not fully compatible with SHM/HugeTLBFS large pages, disabling adaptive resizing");
4790 UseAdaptiveSizePolicy = false;
4791 UseAdaptiveNUMAChunkSizing = false;
4792 }
4793 }
4794 }
4795 if (!UseNUMA && ForceNUMA) {
4796 UseNUMA = true;
4797 }
4798 }
4799
4800 if (MaxFDLimit) {
4801 // set the number of file descriptors to max. print out error
4802 // if getrlimit/setrlimit fails but continue regardless.
4803 struct rlimit nbr_files;
4804 int status = getrlimit(RLIMIT_NOFILE, &nbr_files);
4805 if (status != 0) {
4806 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4807 perror("os::init_2 getrlimit failed");
4808 }
4809 } else {
4810 nbr_files.rlim_cur = nbr_files.rlim_max;
4811 status = setrlimit(RLIMIT_NOFILE, &nbr_files);
4812 if (status != 0) {
4813 if (PrintMiscellaneous && (Verbose || WizardMode)) {
4814 perror("os::init_2 setrlimit failed");
4815 }
4816 }
4817 }
4818 }
4819
4820 // Initialize lock used to serialize thread creation (see os::create_thread)
4821 Linux::set_createThread_lock(new Mutex(Mutex::leaf, "createThread_lock", false));
4822
4823 // at-exit methods are called in the reverse order of their registration.
4824 // atexit functions are called on return from main or as a result of a
4825 // call to exit(3C). There can be only 32 of these functions registered
4826 // and atexit() does not set errno.
4827
4828 if (PerfAllowAtExitRegistration) {
4829 // only register atexit functions if PerfAllowAtExitRegistration is set.
4830 // atexit functions can be delayed until process exit time, which
4831 // can be problematic for embedded VM situations. Embedded VMs should
4832 // call DestroyJavaVM() to assure that VM resources are released.
4833
4834 // note: perfMemory_exit_helper atexit function may be removed in
4835 // the future if the appropriate cleanup code can be added to the
4836 // VM_Exit VMOperation's doit method.
4837 if (atexit(perfMemory_exit_helper) != 0) {
4838 warning("os::init_2 atexit(perfMemory_exit_helper) failed");
4839 }
4840 }
4841
4842 // initialize thread priority policy
4843 prio_init();
4844
4845 return JNI_OK;
4846 }
4847
4848 // Mark the polling page as unreadable
4849 void os::make_polling_page_unreadable(void) {
4850 if (!guard_memory((char*)_polling_page, Linux::page_size())) {
4851 fatal("Could not disable polling page");
4852 }
4853 }
4854
4855 // Mark the polling page as readable
4856 void os::make_polling_page_readable(void) {
4857 if (!linux_mprotect((char *)_polling_page, Linux::page_size(), PROT_READ)) {
4858 fatal("Could not enable polling page");
4859 }
4860 }
4861
4862 int os::active_processor_count() {
4863 // Linux doesn't yet have a (official) notion of processor sets,
4864 // so just return the number of online processors.
4865 int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
4866 assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
4867 return online_cpus;
4868 }
4869
4870 void os::set_native_thread_name(const char *name) {
4871 if (Linux::_pthread_setname_np) {
4872 char buf [16]; // according to glibc manpage, 16 chars incl. '/0'
4873 snprintf(buf, sizeof(buf), "%s", name);
4874 buf[sizeof(buf) - 1] = '\0';
4875 const int rc = Linux::_pthread_setname_np(pthread_self(), buf);
4876 // ERANGE should not happen; all other errors should just be ignored.
4877 assert(rc != ERANGE, "pthread_setname_np failed");
4878 }
4879 }
4880
4881 bool os::distribute_processes(uint length, uint* distribution) {
4882 // Not yet implemented.
4883 return false;
4884 }
4885
4886 bool os::bind_to_processor(uint processor_id) {
4887 // Not yet implemented.
4888 return false;
4889 }
4890
4891 ///
4892
4893 void os::SuspendedThreadTask::internal_do_task() {
4894 if (do_suspend(_thread->osthread())) {
4895 SuspendedThreadTaskContext context(_thread, _thread->osthread()->ucontext());
4896 do_task(context);
4897 do_resume(_thread->osthread());
4898 }
4899 }
4900
4901 class PcFetcher : public os::SuspendedThreadTask {
4902 public:
4903 PcFetcher(Thread* thread) : os::SuspendedThreadTask(thread) {}
4904 ExtendedPC result();
4905 protected:
4906 void do_task(const os::SuspendedThreadTaskContext& context);
4907 private:
4908 ExtendedPC _epc;
4909 };
4910
4911 ExtendedPC PcFetcher::result() {
4912 guarantee(is_done(), "task is not done yet.");
4913 return _epc;
4914 }
4915
4916 void PcFetcher::do_task(const os::SuspendedThreadTaskContext& context) {
4917 Thread* thread = context.thread();
4918 OSThread* osthread = thread->osthread();
4919 if (osthread->ucontext() != NULL) {
4920 _epc = os::Linux::ucontext_get_pc((ucontext_t *) context.ucontext());
4921 } else {
4922 // NULL context is unexpected, double-check this is the VMThread
4923 guarantee(thread->is_VM_thread(), "can only be called for VMThread");
4924 }
4925 }
4926
4927 // Suspends the target using the signal mechanism and then grabs the PC before
4928 // resuming the target. Used by the flat-profiler only
4929 ExtendedPC os::get_thread_pc(Thread* thread) {
4930 // Make sure that it is called by the watcher for the VMThread
4931 assert(Thread::current()->is_Watcher_thread(), "Must be watcher");
4932 assert(thread->is_VM_thread(), "Can only be called for VMThread");
4933
4934 PcFetcher fetcher(thread);
4935 fetcher.run();
4936 return fetcher.result();
4937 }
4938
4939 int os::Linux::safe_cond_timedwait(pthread_cond_t *_cond,
4940 pthread_mutex_t *_mutex,
4941 const struct timespec *_abstime) {
4942 if (is_NPTL()) {
4943 return pthread_cond_timedwait(_cond, _mutex, _abstime);
4944 } else {
4945 // 6292965: LinuxThreads pthread_cond_timedwait() resets FPU control
4946 // word back to default 64bit precision if condvar is signaled. Java
4947 // wants 53bit precision. Save and restore current value.
4948 int fpu = get_fpu_control_word();
4949 int status = pthread_cond_timedwait(_cond, _mutex, _abstime);
4950 set_fpu_control_word(fpu);
4951 return status;
4952 }
4953 }
4954
4955 ////////////////////////////////////////////////////////////////////////////////
4956 // debug support
4957
4958 bool os::find(address addr, outputStream* st) {
4959 Dl_info dlinfo;
4960 memset(&dlinfo, 0, sizeof(dlinfo));
4961 if (dladdr(addr, &dlinfo) != 0) {
4962 st->print(PTR_FORMAT ": ", addr);
4963 if (dlinfo.dli_sname != NULL && dlinfo.dli_saddr != NULL) {
4964 st->print("%s+%#x", dlinfo.dli_sname,
4965 addr - (intptr_t)dlinfo.dli_saddr);
4966 } else if (dlinfo.dli_fbase != NULL) {
4967 st->print("<offset %#x>", addr - (intptr_t)dlinfo.dli_fbase);
4968 } else {
4969 st->print("<absolute address>");
4970 }
4971 if (dlinfo.dli_fname != NULL) {
4972 st->print(" in %s", dlinfo.dli_fname);
4973 }
4974 if (dlinfo.dli_fbase != NULL) {
4975 st->print(" at " PTR_FORMAT, dlinfo.dli_fbase);
4976 }
4977 st->cr();
4978
4979 if (Verbose) {
4980 // decode some bytes around the PC
4981 address begin = clamp_address_in_page(addr-40, addr, os::vm_page_size());
4982 address end = clamp_address_in_page(addr+40, addr, os::vm_page_size());
4983 address lowest = (address) dlinfo.dli_sname;
4984 if (!lowest) lowest = (address) dlinfo.dli_fbase;
4985 if (begin < lowest) begin = lowest;
4986 Dl_info dlinfo2;
4987 if (dladdr(end, &dlinfo2) != 0 && dlinfo2.dli_saddr != dlinfo.dli_saddr
4988 && end > dlinfo2.dli_saddr && dlinfo2.dli_saddr > begin) {
4989 end = (address) dlinfo2.dli_saddr;
4990 }
4991 Disassembler::decode(begin, end, st);
4992 }
4993 return true;
4994 }
4995 return false;
4996 }
4997
4998 ////////////////////////////////////////////////////////////////////////////////
4999 // misc
5000
5001 // This does not do anything on Linux. This is basically a hook for being
5002 // able to use structured exception handling (thread-local exception filters)
5003 // on, e.g., Win32.
5004 void
5005 os::os_exception_wrapper(java_call_t f, JavaValue* value, methodHandle* method,
5006 JavaCallArguments* args, Thread* thread) {
5007 f(value, method, args, thread);
5008 }
5009
5010 void os::print_statistics() {
5011 }
5012
5013 int os::message_box(const char* title, const char* message) {
5014 int i;
5015 fdStream err(defaultStream::error_fd());
5016 for (i = 0; i < 78; i++) err.print_raw("=");
5017 err.cr();
5018 err.print_raw_cr(title);
5019 for (i = 0; i < 78; i++) err.print_raw("-");
5020 err.cr();
5021 err.print_raw_cr(message);
5022 for (i = 0; i < 78; i++) err.print_raw("=");
5023 err.cr();
5024
5025 char buf[16];
5026 // Prevent process from exiting upon "read error" without consuming all CPU
5027 while (::read(0, buf, sizeof(buf)) <= 0) { ::sleep(100); }
5028
5029 return buf[0] == 'y' || buf[0] == 'Y';
5030 }
5031
5032 int os::stat(const char *path, struct stat *sbuf) {
5033 char pathbuf[MAX_PATH];
5034 if (strlen(path) > MAX_PATH - 1) {
5035 errno = ENAMETOOLONG;
5036 return -1;
5037 }
5038 os::native_path(strcpy(pathbuf, path));
5039 return ::stat(pathbuf, sbuf);
5040 }
5041
5042 bool os::check_heap(bool force) {
5043 return true;
5044 }
5045
5046 // Is a (classpath) directory empty?
5047 bool os::dir_is_empty(const char* path) {
5048 DIR *dir = NULL;
5049 struct dirent *ptr;
5050
5051 dir = opendir(path);
5052 if (dir == NULL) return true;
5053
5054 // Scan the directory
5055 bool result = true;
5056 char buf[sizeof(struct dirent) + MAX_PATH];
5057 while (result && (ptr = ::readdir(dir)) != NULL) {
5058 if (strcmp(ptr->d_name, ".") != 0 && strcmp(ptr->d_name, "..") != 0) {
5059 result = false;
5060 }
5061 }
5062 closedir(dir);
5063 return result;
5064 }
5065
5066 // This code originates from JDK's sysOpen and open64_w
5067 // from src/solaris/hpi/src/system_md.c
5068
5069 int os::open(const char *path, int oflag, int mode) {
5070 if (strlen(path) > MAX_PATH - 1) {
5071 errno = ENAMETOOLONG;
5072 return -1;
5073 }
5074
5075 // All file descriptors that are opened in the Java process and not
5076 // specifically destined for a subprocess should have the close-on-exec
5077 // flag set. If we don't set it, then careless 3rd party native code
5078 // might fork and exec without closing all appropriate file descriptors
5079 // (e.g. as we do in closeDescriptors in UNIXProcess.c), and this in
5080 // turn might:
5081 //
5082 // - cause end-of-file to fail to be detected on some file
5083 // descriptors, resulting in mysterious hangs, or
5084 //
5085 // - might cause an fopen in the subprocess to fail on a system
5086 // suffering from bug 1085341.
5087 //
5088 // (Yes, the default setting of the close-on-exec flag is a Unix
5089 // design flaw)
5090 //
5091 // See:
5092 // 1085341: 32-bit stdio routines should support file descriptors >255
5093 // 4843136: (process) pipe file descriptor from Runtime.exec not being closed
5094 // 6339493: (process) Runtime.exec does not close all file descriptors on Solaris 9
5095 //
5096 // Modern Linux kernels (after 2.6.23 2007) support O_CLOEXEC with open().
5097 // O_CLOEXEC is preferable to using FD_CLOEXEC on an open file descriptor
5098 // because it saves a system call and removes a small window where the flag
5099 // is unset. On ancient Linux kernels the O_CLOEXEC flag will be ignored
5100 // and we fall back to using FD_CLOEXEC (see below).
5101 #ifdef O_CLOEXEC
5102 oflag |= O_CLOEXEC;
5103 #endif
5104
5105 int fd = ::open64(path, oflag, mode);
5106 if (fd == -1) return -1;
5107
5108 //If the open succeeded, the file might still be a directory
5109 {
5110 struct stat64 buf64;
5111 int ret = ::fstat64(fd, &buf64);
5112 int st_mode = buf64.st_mode;
5113
5114 if (ret != -1) {
5115 if ((st_mode & S_IFMT) == S_IFDIR) {
5116 errno = EISDIR;
5117 ::close(fd);
5118 return -1;
5119 }
5120 } else {
5121 ::close(fd);
5122 return -1;
5123 }
5124 }
5125
5126 #ifdef FD_CLOEXEC
5127 // Validate that the use of the O_CLOEXEC flag on open above worked.
5128 // With recent kernels, we will perform this check exactly once.
5129 static sig_atomic_t O_CLOEXEC_is_known_to_work = 0;
5130 if (!O_CLOEXEC_is_known_to_work) {
5131 int flags = ::fcntl(fd, F_GETFD);
5132 if (flags != -1) {
5133 if ((flags & FD_CLOEXEC) != 0)
5134 O_CLOEXEC_is_known_to_work = 1;
5135 else
5136 ::fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
5137 }
5138 }
5139 #endif
5140
5141 return fd;
5142 }
5143
5144
5145 // create binary file, rewriting existing file if required
5146 int os::create_binary_file(const char* path, bool rewrite_existing) {
5147 int oflags = O_WRONLY | O_CREAT;
5148 if (!rewrite_existing) {
5149 oflags |= O_EXCL;
5150 }
5151 return ::open64(path, oflags, S_IREAD | S_IWRITE);
5152 }
5153
5154 // return current position of file pointer
5155 jlong os::current_file_offset(int fd) {
5156 return (jlong)::lseek64(fd, (off64_t)0, SEEK_CUR);
5157 }
5158
5159 // move file pointer to the specified offset
5160 jlong os::seek_to_file_offset(int fd, jlong offset) {
5161 return (jlong)::lseek64(fd, (off64_t)offset, SEEK_SET);
5162 }
5163
5164 // This code originates from JDK's sysAvailable
5165 // from src/solaris/hpi/src/native_threads/src/sys_api_td.c
5166
5167 int os::available(int fd, jlong *bytes) {
5168 jlong cur, end;
5169 int mode;
5170 struct stat64 buf64;
5171
5172 if (::fstat64(fd, &buf64) >= 0) {
5173 mode = buf64.st_mode;
5174 if (S_ISCHR(mode) || S_ISFIFO(mode) || S_ISSOCK(mode)) {
5175 // XXX: is the following call interruptible? If so, this might
5176 // need to go through the INTERRUPT_IO() wrapper as for other
5177 // blocking, interruptible calls in this file.
5178 int n;
5179 if (::ioctl(fd, FIONREAD, &n) >= 0) {
5180 *bytes = n;
5181 return 1;
5182 }
5183 }
5184 }
5185 if ((cur = ::lseek64(fd, 0L, SEEK_CUR)) == -1) {
5186 return 0;
5187 } else if ((end = ::lseek64(fd, 0L, SEEK_END)) == -1) {
5188 return 0;
5189 } else if (::lseek64(fd, cur, SEEK_SET) == -1) {
5190 return 0;
5191 }
5192 *bytes = end - cur;
5193 return 1;
5194 }
5195
5196 // Map a block of memory.
5197 char* os::pd_map_memory(int fd, const char* file_name, size_t file_offset,
5198 char *addr, size_t bytes, bool read_only,
5199 bool allow_exec) {
5200 int prot;
5201 int flags = MAP_PRIVATE;
5202
5203 if (read_only) {
5204 prot = PROT_READ;
5205 } else {
5206 prot = PROT_READ | PROT_WRITE;
5207 }
5208
5209 if (allow_exec) {
5210 prot |= PROT_EXEC;
5211 }
5212
5213 if (addr != NULL) {
5214 flags |= MAP_FIXED;
5215 }
5216
5217 char* mapped_address = (char*)mmap(addr, (size_t)bytes, prot, flags,
5218 fd, file_offset);
5219 if (mapped_address == MAP_FAILED) {
5220 return NULL;
5221 }
5222 return mapped_address;
5223 }
5224
5225
5226 // Remap a block of memory.
5227 char* os::pd_remap_memory(int fd, const char* file_name, size_t file_offset,
5228 char *addr, size_t bytes, bool read_only,
5229 bool allow_exec) {
5230 // same as map_memory() on this OS
5231 return os::map_memory(fd, file_name, file_offset, addr, bytes, read_only,
5232 allow_exec);
5233 }
5234
5235
5236 // Unmap a block of memory.
5237 bool os::pd_unmap_memory(char* addr, size_t bytes) {
5238 return munmap(addr, bytes) == 0;
5239 }
5240
5241 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time);
5242
5243 static clockid_t thread_cpu_clockid(Thread* thread) {
5244 pthread_t tid = thread->osthread()->pthread_id();
5245 clockid_t clockid;
5246
5247 // Get thread clockid
5248 int rc = os::Linux::pthread_getcpuclockid(tid, &clockid);
5249 assert(rc == 0, "pthread_getcpuclockid is expected to return 0 code");
5250 return clockid;
5251 }
5252
5253 // current_thread_cpu_time(bool) and thread_cpu_time(Thread*, bool)
5254 // are used by JVM M&M and JVMTI to get user+sys or user CPU time
5255 // of a thread.
5256 //
5257 // current_thread_cpu_time() and thread_cpu_time(Thread*) returns
5258 // the fast estimate available on the platform.
5259
5260 jlong os::current_thread_cpu_time() {
5261 if (os::Linux::supports_fast_thread_cpu_time()) {
5262 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5263 } else {
5264 // return user + sys since the cost is the same
5265 return slow_thread_cpu_time(Thread::current(), true /* user + sys */);
5266 }
5267 }
5268
5269 jlong os::thread_cpu_time(Thread* thread) {
5270 // consistent with what current_thread_cpu_time() returns
5271 if (os::Linux::supports_fast_thread_cpu_time()) {
5272 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5273 } else {
5274 return slow_thread_cpu_time(thread, true /* user + sys */);
5275 }
5276 }
5277
5278 jlong os::current_thread_cpu_time(bool user_sys_cpu_time) {
5279 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5280 return os::Linux::fast_thread_cpu_time(CLOCK_THREAD_CPUTIME_ID);
5281 } else {
5282 return slow_thread_cpu_time(Thread::current(), user_sys_cpu_time);
5283 }
5284 }
5285
5286 jlong os::thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5287 if (user_sys_cpu_time && os::Linux::supports_fast_thread_cpu_time()) {
5288 return os::Linux::fast_thread_cpu_time(thread_cpu_clockid(thread));
5289 } else {
5290 return slow_thread_cpu_time(thread, user_sys_cpu_time);
5291 }
5292 }
5293
5294 // -1 on error.
5295 static jlong slow_thread_cpu_time(Thread *thread, bool user_sys_cpu_time) {
5296 pid_t tid = thread->osthread()->thread_id();
5297 char *s;
5298 char stat[2048];
5299 int statlen;
5300 char proc_name[64];
5301 int count;
5302 long sys_time, user_time;
5303 char cdummy;
5304 int idummy;
5305 long ldummy;
5306 FILE *fp;
5307
5308 snprintf(proc_name, 64, "/proc/self/task/%d/stat", tid);
5309 fp = fopen(proc_name, "r");
5310 if (fp == NULL) return -1;
5311 statlen = fread(stat, 1, 2047, fp);
5312 stat[statlen] = '\0';
5313 fclose(fp);
5314
5315 // Skip pid and the command string. Note that we could be dealing with
5316 // weird command names, e.g. user could decide to rename java launcher
5317 // to "java 1.4.2 :)", then the stat file would look like
5318 // 1234 (java 1.4.2 :)) R ... ...
5319 // We don't really need to know the command string, just find the last
5320 // occurrence of ")" and then start parsing from there. See bug 4726580.
5321 s = strrchr(stat, ')');
5322 if (s == NULL) return -1;
5323
5324 // Skip blank chars
5325 do { s++; } while (s && isspace(*s));
5326
5327 count = sscanf(s,"%c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu",
5328 &cdummy, &idummy, &idummy, &idummy, &idummy, &idummy,
5329 &ldummy, &ldummy, &ldummy, &ldummy, &ldummy,
5330 &user_time, &sys_time);
5331 if (count != 13) return -1;
5332 if (user_sys_cpu_time) {
5333 return ((jlong)sys_time + (jlong)user_time) * (1000000000 / clock_tics_per_sec);
5334 } else {
5335 return (jlong)user_time * (1000000000 / clock_tics_per_sec);
5336 }
5337 }
5338
5339 void os::current_thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5340 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5341 info_ptr->may_skip_backward = false; // elapsed time not wall time
5342 info_ptr->may_skip_forward = false; // elapsed time not wall time
5343 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5344 }
5345
5346 void os::thread_cpu_time_info(jvmtiTimerInfo *info_ptr) {
5347 info_ptr->max_value = ALL_64_BITS; // will not wrap in less than 64 bits
5348 info_ptr->may_skip_backward = false; // elapsed time not wall time
5349 info_ptr->may_skip_forward = false; // elapsed time not wall time
5350 info_ptr->kind = JVMTI_TIMER_TOTAL_CPU; // user+system time is returned
5351 }
5352
5353 bool os::is_thread_cpu_time_supported() {
5354 return true;
5355 }
5356
5357 // System loadavg support. Returns -1 if load average cannot be obtained.
5358 // Linux doesn't yet have a (official) notion of processor sets,
5359 // so just return the system wide load average.
5360 int os::loadavg(double loadavg[], int nelem) {
5361 return ::getloadavg(loadavg, nelem);
5362 }
5363
5364 void os::pause() {
5365 char filename[MAX_PATH];
5366 if (PauseAtStartupFile && PauseAtStartupFile[0]) {
5367 jio_snprintf(filename, MAX_PATH, PauseAtStartupFile);
5368 } else {
5369 jio_snprintf(filename, MAX_PATH, "./vm.paused.%d", current_process_id());
5370 }
5371
5372 int fd = ::open(filename, O_WRONLY | O_CREAT | O_TRUNC, 0666);
5373 if (fd != -1) {
5374 struct stat buf;
5375 ::close(fd);
5376 while (::stat(filename, &buf) == 0) {
5377 (void)::poll(NULL, 0, 100);
5378 }
5379 } else {
5380 jio_fprintf(stderr,
5381 "Could not open pause file '%s', continuing immediately.\n", filename);
5382 }
5383 }
5384
5385
5386 // Refer to the comments in os_solaris.cpp park-unpark. The next two
5387 // comment paragraphs are worth repeating here:
5388 //
5389 // Assumption:
5390 // Only one parker can exist on an event, which is why we allocate
5391 // them per-thread. Multiple unparkers can coexist.
5392 //
5393 // _Event serves as a restricted-range semaphore.
5394 // -1 : thread is blocked, i.e. there is a waiter
5395 // 0 : neutral: thread is running or ready,
5396 // could have been signaled after a wait started
5397 // 1 : signaled - thread is running or ready
5398 //
5399 // Beware -- Some versions of NPTL embody a flaw where pthread_cond_timedwait() can
5400 // hang indefinitely. For instance NPTL 0.60 on 2.4.21-4ELsmp is vulnerable.
5401 // For specifics regarding the bug see GLIBC BUGID 261237 :
5402 // http://www.mail-archive.com/debian-glibc@lists.debian.org/msg10837.html.
5403 // Briefly, pthread_cond_timedwait() calls with an expiry time that's not in the future
5404 // will either hang or corrupt the condvar, resulting in subsequent hangs if the condvar
5405 // is used. (The simple C test-case provided in the GLIBC bug report manifests the
5406 // hang). The JVM is vulernable via sleep(), Object.wait(timo), LockSupport.parkNanos()
5407 // and monitorenter when we're using 1-0 locking. All those operations may result in
5408 // calls to pthread_cond_timedwait(). Using LD_ASSUME_KERNEL to use an older version
5409 // of libpthread avoids the problem, but isn't practical.
5410 //
5411 // Possible remedies:
5412 //
5413 // 1. Establish a minimum relative wait time. 50 to 100 msecs seems to work.
5414 // This is palliative and probabilistic, however. If the thread is preempted
5415 // between the call to compute_abstime() and pthread_cond_timedwait(), more
5416 // than the minimum period may have passed, and the abstime may be stale (in the
5417 // past) resultin in a hang. Using this technique reduces the odds of a hang
5418 // but the JVM is still vulnerable, particularly on heavily loaded systems.
5419 //
5420 // 2. Modify park-unpark to use per-thread (per ParkEvent) pipe-pairs instead
5421 // of the usual flag-condvar-mutex idiom. The write side of the pipe is set
5422 // NDELAY. unpark() reduces to write(), park() reduces to read() and park(timo)
5423 // reduces to poll()+read(). This works well, but consumes 2 FDs per extant
5424 // thread.
5425 //
5426 // 3. Embargo pthread_cond_timedwait() and implement a native "chron" thread
5427 // that manages timeouts. We'd emulate pthread_cond_timedwait() by enqueuing
5428 // a timeout request to the chron thread and then blocking via pthread_cond_wait().
5429 // This also works well. In fact it avoids kernel-level scalability impediments
5430 // on certain platforms that don't handle lots of active pthread_cond_timedwait()
5431 // timers in a graceful fashion.
5432 //
5433 // 4. When the abstime value is in the past it appears that control returns
5434 // correctly from pthread_cond_timedwait(), but the condvar is left corrupt.
5435 // Subsequent timedwait/wait calls may hang indefinitely. Given that, we
5436 // can avoid the problem by reinitializing the condvar -- by cond_destroy()
5437 // followed by cond_init() -- after all calls to pthread_cond_timedwait().
5438 // It may be possible to avoid reinitialization by checking the return
5439 // value from pthread_cond_timedwait(). In addition to reinitializing the
5440 // condvar we must establish the invariant that cond_signal() is only called
5441 // within critical sections protected by the adjunct mutex. This prevents
5442 // cond_signal() from "seeing" a condvar that's in the midst of being
5443 // reinitialized or that is corrupt. Sadly, this invariant obviates the
5444 // desirable signal-after-unlock optimization that avoids futile context switching.
5445 //
5446 // I'm also concerned that some versions of NTPL might allocate an auxilliary
5447 // structure when a condvar is used or initialized. cond_destroy() would
5448 // release the helper structure. Our reinitialize-after-timedwait fix
5449 // put excessive stress on malloc/free and locks protecting the c-heap.
5450 //
5451 // We currently use (4). See the WorkAroundNTPLTimedWaitHang flag.
5452 // It may be possible to refine (4) by checking the kernel and NTPL verisons
5453 // and only enabling the work-around for vulnerable environments.
5454
5455 // utility to compute the abstime argument to timedwait:
5456 // millis is the relative timeout time
5457 // abstime will be the absolute timeout time
5458 // TODO: replace compute_abstime() with unpackTime()
5459
5460 static struct timespec* compute_abstime(timespec* abstime, jlong millis) {
5461 if (millis < 0) millis = 0;
5462
5463 jlong seconds = millis / 1000;
5464 millis %= 1000;
5465 if (seconds > 50000000) { // see man cond_timedwait(3T)
5466 seconds = 50000000;
5467 }
5468
5469 if (os::supports_monotonic_clock()) {
5470 struct timespec now;
5471 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5472 assert_status(status == 0, status, "clock_gettime");
5473 abstime->tv_sec = now.tv_sec + seconds;
5474 long nanos = now.tv_nsec + millis * NANOSECS_PER_MILLISEC;
5475 if (nanos >= NANOSECS_PER_SEC) {
5476 abstime->tv_sec += 1;
5477 nanos -= NANOSECS_PER_SEC;
5478 }
5479 abstime->tv_nsec = nanos;
5480 } else {
5481 struct timeval now;
5482 int status = gettimeofday(&now, NULL);
5483 assert(status == 0, "gettimeofday");
5484 abstime->tv_sec = now.tv_sec + seconds;
5485 long usec = now.tv_usec + millis * 1000;
5486 if (usec >= 1000000) {
5487 abstime->tv_sec += 1;
5488 usec -= 1000000;
5489 }
5490 abstime->tv_nsec = usec * 1000;
5491 }
5492 return abstime;
5493 }
5494
5495 void os::PlatformEvent::park() { // AKA "down()"
5496 // Transitions for _Event:
5497 // -1 => -1 : illegal
5498 // 1 => 0 : pass - return immediately
5499 // 0 => -1 : block; then set _Event to 0 before returning
5500
5501 // Invariant: Only the thread associated with the Event/PlatformEvent
5502 // may call park().
5503 // TODO: assert that _Assoc != NULL or _Assoc == Self
5504 assert(_nParked == 0, "invariant");
5505
5506 int v;
5507 for (;;) {
5508 v = _Event;
5509 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5510 }
5511 guarantee(v >= 0, "invariant");
5512 if (v == 0) {
5513 // Do this the hard way by blocking ...
5514 int status = pthread_mutex_lock(_mutex);
5515 assert_status(status == 0, status, "mutex_lock");
5516 guarantee(_nParked == 0, "invariant");
5517 ++_nParked;
5518 while (_Event < 0) {
5519 status = pthread_cond_wait(_cond, _mutex);
5520 // for some reason, under 2.7 lwp_cond_wait() may return ETIME ...
5521 // Treat this the same as if the wait was interrupted
5522 if (status == ETIME) { status = EINTR; }
5523 assert_status(status == 0 || status == EINTR, status, "cond_wait");
5524 }
5525 --_nParked;
5526
5527 _Event = 0;
5528 status = pthread_mutex_unlock(_mutex);
5529 assert_status(status == 0, status, "mutex_unlock");
5530 // Paranoia to ensure our locked and lock-free paths interact
5531 // correctly with each other.
5532 OrderAccess::fence();
5533 }
5534 guarantee(_Event >= 0, "invariant");
5535 }
5536
5537 int os::PlatformEvent::park(jlong millis) {
5538 // Transitions for _Event:
5539 // -1 => -1 : illegal
5540 // 1 => 0 : pass - return immediately
5541 // 0 => -1 : block; then set _Event to 0 before returning
5542
5543 guarantee(_nParked == 0, "invariant");
5544
5545 int v;
5546 for (;;) {
5547 v = _Event;
5548 if (Atomic::cmpxchg(v-1, &_Event, v) == v) break;
5549 }
5550 guarantee(v >= 0, "invariant");
5551 if (v != 0) return OS_OK;
5552
5553 // We do this the hard way, by blocking the thread.
5554 // Consider enforcing a minimum timeout value.
5555 struct timespec abst;
5556 compute_abstime(&abst, millis);
5557
5558 int ret = OS_TIMEOUT;
5559 int status = pthread_mutex_lock(_mutex);
5560 assert_status(status == 0, status, "mutex_lock");
5561 guarantee(_nParked == 0, "invariant");
5562 ++_nParked;
5563
5564 // Object.wait(timo) will return because of
5565 // (a) notification
5566 // (b) timeout
5567 // (c) thread.interrupt
5568 //
5569 // Thread.interrupt and object.notify{All} both call Event::set.
5570 // That is, we treat thread.interrupt as a special case of notification.
5571 // We ignore spurious OS wakeups unless FilterSpuriousWakeups is false.
5572 // We assume all ETIME returns are valid.
5573 //
5574 // TODO: properly differentiate simultaneous notify+interrupt.
5575 // In that case, we should propagate the notify to another waiter.
5576
5577 while (_Event < 0) {
5578 status = os::Linux::safe_cond_timedwait(_cond, _mutex, &abst);
5579 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5580 pthread_cond_destroy(_cond);
5581 pthread_cond_init(_cond, os::Linux::condAttr());
5582 }
5583 assert_status(status == 0 || status == EINTR ||
5584 status == ETIME || status == ETIMEDOUT,
5585 status, "cond_timedwait");
5586 if (!FilterSpuriousWakeups) break; // previous semantics
5587 if (status == ETIME || status == ETIMEDOUT) break;
5588 // We consume and ignore EINTR and spurious wakeups.
5589 }
5590 --_nParked;
5591 if (_Event >= 0) {
5592 ret = OS_OK;
5593 }
5594 _Event = 0;
5595 status = pthread_mutex_unlock(_mutex);
5596 assert_status(status == 0, status, "mutex_unlock");
5597 assert(_nParked == 0, "invariant");
5598 // Paranoia to ensure our locked and lock-free paths interact
5599 // correctly with each other.
5600 OrderAccess::fence();
5601 return ret;
5602 }
5603
5604 void os::PlatformEvent::unpark() {
5605 // Transitions for _Event:
5606 // 0 => 1 : just return
5607 // 1 => 1 : just return
5608 // -1 => either 0 or 1; must signal target thread
5609 // That is, we can safely transition _Event from -1 to either
5610 // 0 or 1.
5611 // See also: "Semaphores in Plan 9" by Mullender & Cox
5612 //
5613 // Note: Forcing a transition from "-1" to "1" on an unpark() means
5614 // that it will take two back-to-back park() calls for the owning
5615 // thread to block. This has the benefit of forcing a spurious return
5616 // from the first park() call after an unpark() call which will help
5617 // shake out uses of park() and unpark() without condition variables.
5618
5619 if (Atomic::xchg(1, &_Event) >= 0) return;
5620
5621 // Wait for the thread associated with the event to vacate
5622 int status = pthread_mutex_lock(_mutex);
5623 assert_status(status == 0, status, "mutex_lock");
5624 int AnyWaiters = _nParked;
5625 assert(AnyWaiters == 0 || AnyWaiters == 1, "invariant");
5626 if (AnyWaiters != 0 && WorkAroundNPTLTimedWaitHang) {
5627 AnyWaiters = 0;
5628 pthread_cond_signal(_cond);
5629 }
5630 status = pthread_mutex_unlock(_mutex);
5631 assert_status(status == 0, status, "mutex_unlock");
5632 if (AnyWaiters != 0) {
5633 // Note that we signal() *after* dropping the lock for "immortal" Events.
5634 // This is safe and avoids a common class of futile wakeups. In rare
5635 // circumstances this can cause a thread to return prematurely from
5636 // cond_{timed}wait() but the spurious wakeup is benign and the victim
5637 // will simply re-test the condition and re-park itself.
5638 // This provides particular benefit if the underlying platform does not
5639 // provide wait morphing.
5640 status = pthread_cond_signal(_cond);
5641 assert_status(status == 0, status, "cond_signal");
5642 }
5643 }
5644
5645
5646 // JSR166
5647 // -------------------------------------------------------
5648
5649 // The solaris and linux implementations of park/unpark are fairly
5650 // conservative for now, but can be improved. They currently use a
5651 // mutex/condvar pair, plus a a count.
5652 // Park decrements count if > 0, else does a condvar wait. Unpark
5653 // sets count to 1 and signals condvar. Only one thread ever waits
5654 // on the condvar. Contention seen when trying to park implies that someone
5655 // is unparking you, so don't wait. And spurious returns are fine, so there
5656 // is no need to track notifications.
5657
5658 // This code is common to linux and solaris and will be moved to a
5659 // common place in dolphin.
5660 //
5661 // The passed in time value is either a relative time in nanoseconds
5662 // or an absolute time in milliseconds. Either way it has to be unpacked
5663 // into suitable seconds and nanoseconds components and stored in the
5664 // given timespec structure.
5665 // Given time is a 64-bit value and the time_t used in the timespec is only
5666 // a signed-32-bit value (except on 64-bit Linux) we have to watch for
5667 // overflow if times way in the future are given. Further on Solaris versions
5668 // prior to 10 there is a restriction (see cond_timedwait) that the specified
5669 // number of seconds, in abstime, is less than current_time + 100,000,000.
5670 // As it will be 28 years before "now + 100000000" will overflow we can
5671 // ignore overflow and just impose a hard-limit on seconds using the value
5672 // of "now + 100,000,000". This places a limit on the timeout of about 3.17
5673 // years from "now".
5674
5675 static void unpackTime(timespec* absTime, bool isAbsolute, jlong time) {
5676 assert(time > 0, "convertTime");
5677 time_t max_secs = 0;
5678
5679 if (!os::supports_monotonic_clock() || isAbsolute) {
5680 struct timeval now;
5681 int status = gettimeofday(&now, NULL);
5682 assert(status == 0, "gettimeofday");
5683
5684 max_secs = now.tv_sec + MAX_SECS;
5685
5686 if (isAbsolute) {
5687 jlong secs = time / 1000;
5688 if (secs > max_secs) {
5689 absTime->tv_sec = max_secs;
5690 } else {
5691 absTime->tv_sec = secs;
5692 }
5693 absTime->tv_nsec = (time % 1000) * NANOSECS_PER_MILLISEC;
5694 } else {
5695 jlong secs = time / NANOSECS_PER_SEC;
5696 if (secs >= MAX_SECS) {
5697 absTime->tv_sec = max_secs;
5698 absTime->tv_nsec = 0;
5699 } else {
5700 absTime->tv_sec = now.tv_sec + secs;
5701 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_usec*1000;
5702 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5703 absTime->tv_nsec -= NANOSECS_PER_SEC;
5704 ++absTime->tv_sec; // note: this must be <= max_secs
5705 }
5706 }
5707 }
5708 } else {
5709 // must be relative using monotonic clock
5710 struct timespec now;
5711 int status = os::Linux::clock_gettime(CLOCK_MONOTONIC, &now);
5712 assert_status(status == 0, status, "clock_gettime");
5713 max_secs = now.tv_sec + MAX_SECS;
5714 jlong secs = time / NANOSECS_PER_SEC;
5715 if (secs >= MAX_SECS) {
5716 absTime->tv_sec = max_secs;
5717 absTime->tv_nsec = 0;
5718 } else {
5719 absTime->tv_sec = now.tv_sec + secs;
5720 absTime->tv_nsec = (time % NANOSECS_PER_SEC) + now.tv_nsec;
5721 if (absTime->tv_nsec >= NANOSECS_PER_SEC) {
5722 absTime->tv_nsec -= NANOSECS_PER_SEC;
5723 ++absTime->tv_sec; // note: this must be <= max_secs
5724 }
5725 }
5726 }
5727 assert(absTime->tv_sec >= 0, "tv_sec < 0");
5728 assert(absTime->tv_sec <= max_secs, "tv_sec > max_secs");
5729 assert(absTime->tv_nsec >= 0, "tv_nsec < 0");
5730 assert(absTime->tv_nsec < NANOSECS_PER_SEC, "tv_nsec >= nanos_per_sec");
5731 }
5732
5733 void Parker::park(bool isAbsolute, jlong time) {
5734 // Ideally we'd do something useful while spinning, such
5735 // as calling unpackTime().
5736
5737 // Optional fast-path check:
5738 // Return immediately if a permit is available.
5739 // We depend on Atomic::xchg() having full barrier semantics
5740 // since we are doing a lock-free update to _counter.
5741 if (Atomic::xchg(0, &_counter) > 0) return;
5742
5743 Thread* thread = Thread::current();
5744 assert(thread->is_Java_thread(), "Must be JavaThread");
5745 JavaThread *jt = (JavaThread *)thread;
5746
5747 // Optional optimization -- avoid state transitions if there's an interrupt pending.
5748 // Check interrupt before trying to wait
5749 if (Thread::is_interrupted(thread, false)) {
5750 return;
5751 }
5752
5753 // Next, demultiplex/decode time arguments
5754 timespec absTime;
5755 if (time < 0 || (isAbsolute && time == 0)) { // don't wait at all
5756 return;
5757 }
5758 if (time > 0) {
5759 unpackTime(&absTime, isAbsolute, time);
5760 }
5761
5762
5763 // Enter safepoint region
5764 // Beware of deadlocks such as 6317397.
5765 // The per-thread Parker:: mutex is a classic leaf-lock.
5766 // In particular a thread must never block on the Threads_lock while
5767 // holding the Parker:: mutex. If safepoints are pending both the
5768 // the ThreadBlockInVM() CTOR and DTOR may grab Threads_lock.
5769 ThreadBlockInVM tbivm(jt);
5770
5771 // Don't wait if cannot get lock since interference arises from
5772 // unblocking. Also. check interrupt before trying wait
5773 if (Thread::is_interrupted(thread, false) || pthread_mutex_trylock(_mutex) != 0) {
5774 return;
5775 }
5776
5777 int status;
5778 if (_counter > 0) { // no wait needed
5779 _counter = 0;
5780 status = pthread_mutex_unlock(_mutex);
5781 assert(status == 0, "invariant");
5782 // Paranoia to ensure our locked and lock-free paths interact
5783 // correctly with each other and Java-level accesses.
5784 OrderAccess::fence();
5785 return;
5786 }
5787
5788 #ifdef ASSERT
5789 // Don't catch signals while blocked; let the running threads have the signals.
5790 // (This allows a debugger to break into the running thread.)
5791 sigset_t oldsigs;
5792 sigset_t* allowdebug_blocked = os::Linux::allowdebug_blocked_signals();
5793 pthread_sigmask(SIG_BLOCK, allowdebug_blocked, &oldsigs);
5794 #endif
5795
5796 OSThreadWaitState osts(thread->osthread(), false /* not Object.wait() */);
5797 jt->set_suspend_equivalent();
5798 // cleared by handle_special_suspend_equivalent_condition() or java_suspend_self()
5799
5800 assert(_cur_index == -1, "invariant");
5801 if (time == 0) {
5802 _cur_index = REL_INDEX; // arbitrary choice when not timed
5803 status = pthread_cond_wait(&_cond[_cur_index], _mutex);
5804 } else {
5805 _cur_index = isAbsolute ? ABS_INDEX : REL_INDEX;
5806 status = os::Linux::safe_cond_timedwait(&_cond[_cur_index], _mutex, &absTime);
5807 if (status != 0 && WorkAroundNPTLTimedWaitHang) {
5808 pthread_cond_destroy(&_cond[_cur_index]);
5809 pthread_cond_init(&_cond[_cur_index], isAbsolute ? NULL : os::Linux::condAttr());
5810 }
5811 }
5812 _cur_index = -1;
5813 assert_status(status == 0 || status == EINTR ||
5814 status == ETIME || status == ETIMEDOUT,
5815 status, "cond_timedwait");
5816
5817 #ifdef ASSERT
5818 pthread_sigmask(SIG_SETMASK, &oldsigs, NULL);
5819 #endif
5820
5821 _counter = 0;
5822 status = pthread_mutex_unlock(_mutex);
5823 assert_status(status == 0, status, "invariant");
5824 // Paranoia to ensure our locked and lock-free paths interact
5825 // correctly with each other and Java-level accesses.
5826 OrderAccess::fence();
5827
5828 // If externally suspended while waiting, re-suspend
5829 if (jt->handle_special_suspend_equivalent_condition()) {
5830 jt->java_suspend_self();
5831 }
5832 }
5833
5834 void Parker::unpark() {
5835 int status = pthread_mutex_lock(_mutex);
5836 assert(status == 0, "invariant");
5837 const int s = _counter;
5838 _counter = 1;
5839 if (s < 1) {
5840 // thread might be parked
5841 if (_cur_index != -1) {
5842 // thread is definitely parked
5843 if (WorkAroundNPTLTimedWaitHang) {
5844 status = pthread_cond_signal(&_cond[_cur_index]);
5845 assert(status == 0, "invariant");
5846 status = pthread_mutex_unlock(_mutex);
5847 assert(status == 0, "invariant");
5848 } else {
5849 status = pthread_mutex_unlock(_mutex);
5850 assert(status == 0, "invariant");
5851 status = pthread_cond_signal(&_cond[_cur_index]);
5852 assert(status == 0, "invariant");
5853 }
5854 } else {
5855 pthread_mutex_unlock(_mutex);
5856 assert(status == 0, "invariant");
5857 }
5858 } else {
5859 pthread_mutex_unlock(_mutex);
5860 assert(status == 0, "invariant");
5861 }
5862 }
5863
5864
5865 extern char** environ;
5866
5867 #ifndef __NR_fork
5868 #define __NR_fork IA32_ONLY(2) IA64_ONLY(not defined) AMD64_ONLY(57) AARCH64_ONLY(1079)
5869 #endif
5870
5871 #ifndef __NR_execve
5872 #define __NR_execve IA32_ONLY(11) IA64_ONLY(1033) AMD64_ONLY(59) AARCH64_ONLY(221)
5873 #endif
5874
5875 // Run the specified command in a separate process. Return its exit value,
5876 // or -1 on failure (e.g. can't fork a new process).
5877 // Unlike system(), this function can be called from signal handler. It
5878 // doesn't block SIGINT et al.
5879 int os::fork_and_exec(char* cmd) {
5880 const char * argv[4] = {"sh", "-c", cmd, NULL};
5881
5882 // fork() in LinuxThreads/NPTL is not async-safe. It needs to run
5883 // pthread_atfork handlers and reset pthread library. All we need is a
5884 // separate process to execve. Make a direct syscall to fork process.
5885 // On IA64 there's no fork syscall, we have to use fork() and hope for
5886 // the best...
5887 pid_t pid = NOT_IA64(syscall(__NR_fork);)
5888 IA64_ONLY(fork();)
5889
5890 if (pid < 0) {
5891 // fork failed
5892 return -1;
5893
5894 } else if (pid == 0) {
5895 // child process
5896
5897 // execve() in LinuxThreads will call pthread_kill_other_threads_np()
5898 // first to kill every thread on the thread list. Because this list is
5899 // not reset by fork() (see notes above), execve() will instead kill
5900 // every thread in the parent process. We know this is the only thread
5901 // in the new process, so make a system call directly.
5902 // IA64 should use normal execve() from glibc to match the glibc fork()
5903 // above.
5904 NOT_IA64(syscall(__NR_execve, "/bin/sh", argv, environ);)
5905 IA64_ONLY(execve("/bin/sh", (char* const*)argv, environ);)
5906
5907 // execve failed
5908 _exit(-1);
5909
5910 } else {
5911 // copied from J2SE ..._waitForProcessExit() in UNIXProcess_md.c; we don't
5912 // care about the actual exit code, for now.
5913
5914 int status;
5915
5916 // Wait for the child process to exit. This returns immediately if
5917 // the child has already exited. */
5918 while (waitpid(pid, &status, 0) < 0) {
5919 switch (errno) {
5920 case ECHILD: return 0;
5921 case EINTR: break;
5922 default: return -1;
5923 }
5924 }
5925
5926 if (WIFEXITED(status)) {
5927 // The child exited normally; get its exit code.
5928 return WEXITSTATUS(status);
5929 } else if (WIFSIGNALED(status)) {
5930 // The child exited because of a signal
5931 // The best value to return is 0x80 + signal number,
5932 // because that is what all Unix shells do, and because
5933 // it allows callers to distinguish between process exit and
5934 // process death by signal.
5935 return 0x80 + WTERMSIG(status);
5936 } else {
5937 // Unknown exit code; pass it through
5938 return status;
5939 }
5940 }
5941 }
5942
5943 // is_headless_jre()
5944 //
5945 // Test for the existence of xawt/libmawt.so or libawt_xawt.so
5946 // in order to report if we are running in a headless jre
5947 //
5948 // Since JDK8 xawt/libmawt.so was moved into the same directory
5949 // as libawt.so, and renamed libawt_xawt.so
5950 //
5951 bool os::is_headless_jre() {
5952 struct stat statbuf;
5953 char buf[MAXPATHLEN];
5954 char libmawtpath[MAXPATHLEN];
5955 const char *xawtstr = "/xawt/libmawt.so";
5956 const char *new_xawtstr = "/libawt_xawt.so";
5957 char *p;
5958
5959 // Get path to libjvm.so
5960 os::jvm_path(buf, sizeof(buf));
5961
5962 // Get rid of libjvm.so
5963 p = strrchr(buf, '/');
5964 if (p == NULL) {
5965 return false;
5966 } else {
5967 *p = '\0';
5968 }
5969
5970 // Get rid of client or server
5971 p = strrchr(buf, '/');
5972 if (p == NULL) {
5973 return false;
5974 } else {
5975 *p = '\0';
5976 }
5977
5978 // check xawt/libmawt.so
5979 strcpy(libmawtpath, buf);
5980 strcat(libmawtpath, xawtstr);
5981 if (::stat(libmawtpath, &statbuf) == 0) return false;
5982
5983 // check libawt_xawt.so
5984 strcpy(libmawtpath, buf);
5985 strcat(libmawtpath, new_xawtstr);
5986 if (::stat(libmawtpath, &statbuf) == 0) return false;
5987
5988 return true;
5989 }
5990
5991 // Get the default path to the core file
5992 // Returns the length of the string
5993 int os::get_core_path(char* buffer, size_t bufferSize) {
5994 /*
5995 * Max length of /proc/sys/kernel/core_pattern is 128 characters.
5996 * See https://www.kernel.org/doc/Documentation/sysctl/kernel.txt
5997 */
5998 const int core_pattern_len = 129;
5999 char core_pattern[core_pattern_len] = {0};
6000
6001 int core_pattern_file = ::open("/proc/sys/kernel/core_pattern", O_RDONLY);
6002 if (core_pattern_file != -1) {
6003 ssize_t ret = ::read(core_pattern_file, core_pattern, core_pattern_len);
6004 ::close(core_pattern_file);
6005
6006 if (ret > 0) {
6007 char *last_char = core_pattern + strlen(core_pattern) - 1;
6008
6009 if (*last_char == '\n') {
6010 *last_char = '\0';
6011 }
6012 }
6013 }
6014
6015 if (strlen(core_pattern) == 0) {
6016 return 0;
6017 }
6018
6019 char *pid_pos = strstr(core_pattern, "%p");
6020 size_t written;
6021
6022 if (core_pattern[0] == '/') {
6023 written = jio_snprintf(buffer, bufferSize, core_pattern);
6024 } else {
6025 char cwd[PATH_MAX];
6026
6027 const char* p = get_current_directory(cwd, PATH_MAX);
6028 if (p == NULL) {
6029 assert(p != NULL, "failed to get current directory");
6030 return 0;
6031 }
6032
6033 if (core_pattern[0] == '|') {
6034 written = jio_snprintf(buffer, bufferSize,
6035 "\"%s\" (or dumping to %s/core.%d)",
6036 &core_pattern[1], p, current_process_id());
6037 } else {
6038 written = jio_snprintf(buffer, bufferSize, "%s/%s", p, core_pattern);
6039 }
6040 }
6041
6042 if ((written >= 0) && (written < bufferSize)
6043 && (pid_pos == NULL) && (core_pattern[0] != '|')) {
6044 int core_uses_pid_file = ::open("/proc/sys/kernel/core_uses_pid", O_RDONLY);
6045
6046 if (core_uses_pid_file != -1) {
6047 char core_uses_pid = 0;
6048 ssize_t ret = ::read(core_uses_pid_file, &core_uses_pid, 1);
6049 ::close(core_uses_pid_file);
6050
6051 if (core_uses_pid == '1'){
6052 jio_snprintf(buffer + written, bufferSize - written,
6053 ".%d", current_process_id());
6054 }
6055 }
6056 }
6057
6058 return strlen(buffer);
6059 }
6060
6061 /////////////// Unit tests ///////////////
6062
6063 #ifndef PRODUCT
6064
6065 #define test_log(...) \
6066 do { \
6067 if (VerboseInternalVMTests) { \
6068 tty->print_cr(__VA_ARGS__); \
6069 tty->flush(); \
6070 } \
6071 } while (false)
6072
6073 class TestReserveMemorySpecial : AllStatic {
6074 public:
6075 static void small_page_write(void* addr, size_t size) {
6076 size_t page_size = os::vm_page_size();
6077
6078 char* end = (char*)addr + size;
6079 for (char* p = (char*)addr; p < end; p += page_size) {
6080 *p = 1;
6081 }
6082 }
6083
6084 static void test_reserve_memory_special_huge_tlbfs_only(size_t size) {
6085 if (!UseHugeTLBFS) {
6086 return;
6087 }
6088
6089 test_log("test_reserve_memory_special_huge_tlbfs_only(" SIZE_FORMAT ")", size);
6090
6091 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_only(size, NULL, false);
6092
6093 if (addr != NULL) {
6094 small_page_write(addr, size);
6095
6096 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6097 }
6098 }
6099
6100 static void test_reserve_memory_special_huge_tlbfs_only() {
6101 if (!UseHugeTLBFS) {
6102 return;
6103 }
6104
6105 size_t lp = os::large_page_size();
6106
6107 for (size_t size = lp; size <= lp * 10; size += lp) {
6108 test_reserve_memory_special_huge_tlbfs_only(size);
6109 }
6110 }
6111
6112 static void test_reserve_memory_special_huge_tlbfs_mixed(size_t size, size_t alignment) {
6113 if (!UseHugeTLBFS) {
6114 return;
6115 }
6116
6117 test_log("test_reserve_memory_special_huge_tlbfs_mixed(" SIZE_FORMAT ", " SIZE_FORMAT ")",
6118 size, alignment);
6119
6120 assert(size >= os::large_page_size(), "Incorrect input to test");
6121
6122 char* addr = os::Linux::reserve_memory_special_huge_tlbfs_mixed(size, alignment, NULL, false);
6123
6124 if (addr != NULL) {
6125 small_page_write(addr, size);
6126
6127 os::Linux::release_memory_special_huge_tlbfs(addr, size);
6128 }
6129 }
6130
6131 static void test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(size_t size) {
6132 size_t lp = os::large_page_size();
6133 size_t ag = os::vm_allocation_granularity();
6134
6135 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6136 test_reserve_memory_special_huge_tlbfs_mixed(size, alignment);
6137 }
6138 }
6139
6140 static void test_reserve_memory_special_huge_tlbfs_mixed() {
6141 size_t lp = os::large_page_size();
6142 size_t ag = os::vm_allocation_granularity();
6143
6144 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp);
6145 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + ag);
6146 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp + lp / 2);
6147 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2);
6148 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + ag);
6149 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 - ag);
6150 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 2 + lp / 2);
6151 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10);
6152 test_reserve_memory_special_huge_tlbfs_mixed_all_alignments(lp * 10 + lp / 2);
6153 }
6154
6155 static void test_reserve_memory_special_huge_tlbfs() {
6156 if (!UseHugeTLBFS) {
6157 return;
6158 }
6159
6160 test_reserve_memory_special_huge_tlbfs_only();
6161 test_reserve_memory_special_huge_tlbfs_mixed();
6162 }
6163
6164 static void test_reserve_memory_special_shm(size_t size, size_t alignment) {
6165 if (!UseSHM) {
6166 return;
6167 }
6168
6169 test_log("test_reserve_memory_special_shm(" SIZE_FORMAT ", " SIZE_FORMAT ")", size, alignment);
6170
6171 char* addr = os::Linux::reserve_memory_special_shm(size, alignment, NULL, false);
6172
6173 if (addr != NULL) {
6174 assert(is_ptr_aligned(addr, alignment), "Check");
6175 assert(is_ptr_aligned(addr, os::large_page_size()), "Check");
6176
6177 small_page_write(addr, size);
6178
6179 os::Linux::release_memory_special_shm(addr, size);
6180 }
6181 }
6182
6183 static void test_reserve_memory_special_shm() {
6184 size_t lp = os::large_page_size();
6185 size_t ag = os::vm_allocation_granularity();
6186
6187 for (size_t size = ag; size < lp * 3; size += ag) {
6188 for (size_t alignment = ag; is_size_aligned(size, alignment); alignment *= 2) {
6189 test_reserve_memory_special_shm(size, alignment);
6190 }
6191 }
6192 }
6193
6194 static void test() {
6195 test_reserve_memory_special_huge_tlbfs();
6196 test_reserve_memory_special_shm();
6197 }
6198 };
6199
6200 void TestReserveMemorySpecial_test() {
6201 TestReserveMemorySpecial::test();
6202 }
6203
6204 #endif